The Next Generational Shift In Enterprise Infrastructure Has Arrived. If SlideShare is broken, please download report here: https://www.scribd.com/document/352452857/2017-Enterprise-Almanac
On Starlink, presented by Geoff Huston at NZNOG 2024
2017 Enterprise Almanac
1. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC
2017 ENTERPRISE ALMANAC
USHERING IN A NEW ERA OF ENTERPRISE TECH
By Michael Yamnitsky
#2017Almanac
2. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 2
For the past four years at Work-Bench, we’ve been investing in a total reimagining of the enterprise technology stack.
We’re in the midst of a once in a decade tectonic shift of infrastructure that powers the Fortune 1000 and is unlike
anything we’ve seen before. Whereas consumer tech has the Mary Meeker Internet Trends report for an aggregate
view of industry trends, enterprise technology was missing a comprehensive overview of the key trends - so we’re
launching the Enterprise Almanac to share our thinking on the trends reshaping enterprise technology.
Our primary aim is to help founders see the forest from the trees. For Fortune 1000 executives and other players in
the ecosystem, it will help cut through the noise and marketing hype to see what really matters. It’s wishful thinking,
but we also hope new talent gets excited about enterprise after reading this report. By no means will most of the
predictions be correct, but our purpose is to start the discussion by putting this stake in the ground.
Please share any and all feedback via email at michael@work-bench.com or on Twitter at @ItsYamnitsky.
PREAMBLE
INTRODUCING THE WORK-BENCH ENTERPRISE ALMANAC
MICHAEL YAMNITSKY
Venture Partner, Work-Bench
3. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 3
PREAMBLE
ABOUT WORK-BENCH
About Us
Work-Bench is an enterprise technology focused venture fund.
Our Thesis
Customer-centricity. We make it our focus to deeply understand the business and IT needs of the
Fortune 1000 in order to make more informed decisions in our search for the next enterprise giants.
This is highly informed by our backgrounds in corporate IT at leading Wall Street banks and as Industry
Analysts which is unique in the venture business.
Our Model
Our model flows directly from our thesis. We leverage our deep corporate network in New York City and
beyond as a way to identify trends, pick the winners, and secure customers for our portfolio companies.
4. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 4
PREAMBLE
TABLE OF CONTENTS
I. 2017 Macro Perspective
The Next Generational Shift In Enterprise Technology Has Arrived
II. Four Vertical Themes
1. Machine Intelligence
2. Cloud Native
3. Cybersecurity
4. Internet of Things (IoT)
III. Tips for Entrepreneurs
5. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 5
PREAMBLE
SPECIAL THANKS
Special thanks to…
Team Work-Bench
Jonathan Lehr, Jessica Lin, Vipin Chamakkala, Kelley Mak, Mickey Graham, and Dash Adam who added significant
contributions and healthy debate for the content of this presentation.
Friends in the Enterprise Tech ecosystem
Scott Coleman (Ignition Partners), Lenny Pruss (Amplify Partners), Lonne Jaffe (Insight Venture Partners), Frank Gillett
(Forrester Research), Aaref Hilaly (Sequoia), Bradford Cross (Merlon Intelligence), Matt Turck (Firstmark Capital), Drew
Conway (Alluvium), Diego Oppenheimer (Algorithmia), Greg Smithies (Versive), Tim Eades & Keith Stewart (vArmour),
Divya Venkatachari (Cisco), Ed Anuff (Apigee/Google), and Winter Mead (Sapphire Ventures) for contributing their
thoughts.
Work-Bench Founders and CEOs
For keeping me honest and never failing to surprise us with with where technology can take us on this pale blue dot.
6. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 6
PREAMBLE
SMALL DISCLAIMER
Our views are shaped by anecdotal evidence based on our interactions with entrepreneurs, corporate
customers, and big tech leaders. Take that for what it’s worth. We’ve done our best to separate fact from
opinion by highlighting opinionated perspectives in blue.
You’ll notice many qualitative details, but a dearth of data in this report. The trends we discuss are indeed
early – they can’t be rigorously quantified in customer surveys ran by Forrester and Gartner, nor can they be
segmented out of spending figures by IDC. CBInsights and Pitchbook provide valuable fundraising data,
but since history has shown there’s a disequilibrium between market potential and fundraising in early
market crests, we’ve decided to keep funding figures to a minimum. Our intent was to avoid cherry picking
funding data to serve our purpose and make unfair claims of causality.
We have disclosed our investments where appropriate.
7. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 7
2017 Macro Perspective
The Next Generational Shift In Enterprise Technology Has Arrived
8. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 8
MACRO PERSPECTIVE
SPEED, SCALE, CX DEFINES VALUE IN TODAY’S POST-INTERNET ECONOMY
The top 5 most valuable US public companies in 2017
Market leadership requires a combination of customer experience, scale, speed, standards, and insight
9. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 9
MACRO PERSPECTIVE
STARK EVIDENCE OF THIS AS 3 OF THEM TRANSFORM ENTERPRISE TECH
The core tenants of these powerful companies (speed, scale, standards) led them to expose their internal capabilities to global
companies around the world and evolve into “megaclouds” dominating growth in the enterprise IT market
10. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 10
MACRO PERSPECTIVE
MEGACLOUDS ARE FIGHTING TO BE #1 PLUMBING FOR DIGITAL BUSINESS
$17.1 Billion (2017 Revenue Est.)
40% YoY Growth
$6.1 Billion (2017 Revenue Est.)
81% YoY Growth
$950 Million (2017 Revenue Est.)
75% YoY Growth
PRODUCT STRATEGY
The monocloud that’s good enough for most things,
not amazing for anything. Heading down proprietary
path as most services are integrally tied to their public
cloud architecture.
GTM STRATEGY
Aggressive enterprise sales: lock-in, land-and-expand.
BIG EXISTENTIAL QUESTION
Amazon can’t allocate 30 top PhDs to solve a single
problem. Who will hit Amazon in the achilles heel?
PRODUCT STRATEGY
Play to internal strengths: Underserved enterprise
workloads like legacy Microsoft products, platform and
application services for modern enterprise apps.
GTM STRATEGY
Strong enterprise support model.
BIG EXISTENTIAL QUESTION
Will enterprise chops trump Amazon’s scale and scope?
PRODUCT STRATEGY
Google shines strength in machine learning, developer tools,
and container orchestration (Kubernetes).
GTM STRATEGY
Historically Google hasn’t catered to the enterprise with sales
& support. They’re apparently trying to change this though.
BIG EXISTENTIAL QUESTION
Can Diane Green, Sam Ramji, and the first-class GTM team
from Apigee bring Google from enterprise 0 to hero?
PLAYER #1 (CATEGORY LEADER):
MOMENTUM AND BRAND NAME
PLAYER #2 (FOR NOW):
ENTERPRISE HERITAGE
PLAYER #3 (KILLER PRODUCTS):
BUT WHERE’S THE ENTERPRISE LOVE?
Besides a few serious regional players like Alibaba, global enterprises have 3 main
marketplace bazaars to choose from to power their digital transformation
Source: Estimates from Bank of America Merrill Lynch’s “Server & Enterprise Software: Cloud Wars 9: AI : From faster to smarter powered by ABC.” May 8, 2017. Revenue includes PaaS & IaaS.
11. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 11
… The opportunity is massive, so
megaclouds have gotten a little bit
territorial…
Darth vader death grip?
12. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 12
MACRO PERSPECTIVE
TECHNOLOGY GIANTS ARE CRUMBLING AT THE HEELS OF MEGACLOUDS
Software division spin out in 2016 20 straight quarters of YoY revenue decline
13. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 13
MACRO PERSPECTIVE
ENTREPRENEURS IN SILICON VALLEY AREN’T IMMUNE EITHER
Amazon’s ReInvent is a startup bloodbath
Amazon Lightsail = DigitalOcean killer?
AWS X-Ray = testing and debugging startup killer?
Amazon Pinpoint = mobile analytics startup killer?
The list goes on…
Megaclouds put tremendous pressure on startups within the cloud ecosystem —
the Amazon clan will either build it on their own or make it difficult for startups to scale
Amazon mines the startup ecosystem and
replicates their most popular software tools…
…And makes it difficult to create a
tech company around deep IP by
hosting the latest open source software
at a fraction of anyone else’s cost
14. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 14
… But the rules are about to change again
with the resurgence of Artificial Intelligence
15. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 15
MACRO PERSPECTIVE
AI STARTUP FUNDRAISING AT RECORD HIGHS
Source: CBInsights “The 2016 AI Recap: Startups See Record High In Deals And Funding”
16. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 16
MACRO PERSPECTIVE
TECH POWERHOUSES ARE PLAYING DEFENSE WITH ACQUI-HIRES
Source: CBInsights “The Race For AI: Google, Twitter, Intel, Apple In A Rush To Grab Artificial Intelligence Startups”
Race for AI: top acquirers of AI startups
17. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 17
MACRO PERSPECTIVE
STRATEGY IS TO KEEP AI ON A LEASH BY DEMOCRATIZING IT
“One of the most exciting things we all
can do is demystify machine learning
and AI. It’s important for this to be
accessible by all people” Sundar
Pichai, CEO of Google
Democratizing AI retains its disruptive power, allowing megaclouds to maintain
their power grip through sheer scale of delivering
Source: https://www.wired.com/2017/05/sundar-pichai-sees-googles-future-smartest-cloud/
18. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 18
MACRO PERSPECTIVE
JEFF BEZOS ADMITS AI VALUE IS IN REAL WORLD APPLICATION IN OPERATIONS
Source: http://www.businessinsider.com/jeff-bezos-shareholder-letter-on-ai-and-machine-learning-2017-4
“But much of what we do with machine learning
happens beneath the surface. Machine learning
drives our algorithms for demand forecasting,
product search ranking, product and deals
recommendations, merchandising placements,
fraud detection, translations, and much more.
Though less visible, much of the impact of
machine learning will be of this type — quietly
but meaningfully improving core operations.”
19. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 19
MACRO PERSPECTIVE
WE BELIEVE A “MINI” AI CRASH IS IMMINENT
Large tech companies are ending their
talent shopping spree, leaving many AI
startups with inflated valuations and no real
business in the dust. Salesforce is already
showing signs it’s getting over acqui-hires as it
pivots to internally developing Einstein after
gobbling up startups in 2015 and 2016.
Too many competitive Series A & B deals in AI
that have come across our desks at Work-
Bench over the past year have had valuations
in the high double digits, with many even
$100M+. This can’t last long.
20. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 20
MACRO PERSPECTIVE
LIKE THE INTERNET ECONOMY, AI VALUE WILL BE CREATED AFTER THE CRASH
After the crash…
The successful companies will focus on using
AI to enable business process transformation
21. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 21
MACRO PERSPECTIVE
AI COMPANIES WILL CREATE VALUE THROUGH SYSTEMS OF INTELLIGENCE
Systems of Intelligence are highly focused analytical systems intended to solve business challenges and
objectives (i.e. increase revenue and customer experience, improve operations, reduce risk)
Value created by:
• Integrating data from multiple
sources include non-tradition
information rich channels
• Novel new forms of data capture
• Cleverly optimizing the data
preparation and AI training process
Value created by:
• Embedding domain experts into the
debugging and hyper-parameter
tuning process
• Incorporating feedback from human
experts into the system of record (SOR)
Value created by:
• Designing products from data
capabilities up to user experience
and not the other way around
• Software UI as invisible as possible >
fancy GUIs. Name of the game is
making the workflow as seamless as
possible.
Original Framework Source: Jerry Chen’s “The New Moats - Why Systems of Intelligence are the Next Defensible Business Model”
22. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 22
MACRO PERSPECTIVE
EXAMPLES OF SYSTEMS OF INTELLIGENCE
• Manufacturing: predictive maintenance on factory equipment in a
manufacturing facility
• Insurance: Automation of consumer car insurance filing claims
• Pharmaceuticals: Optimization of R&D resource allocation for a
portfolio of drug candidates in clinical trial
•Financial services: Automated customer interactions with retail
banking “chat bots” in natural language
23. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 23
Systems of Intelligence are like Ford Assembly Lines and
Toyota Production Systems – powerful weapons for
competitive process advantage.
24. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 24
MACRO PERSPECTIVE
SYSTEMS OF INTELLIGENCE BARRIER TO ENTRY LIES IN TIGHT INTEGRATION
Domain
expertise
AI
Data
Data-driven
product design
• Cloud moat =
unbundling “capabilities” into
individually deployed “microservices”
for scale advantage
Domain expertise
baked into product /
operational lifecycle
Turning the
intelligence into
action anchors
product design
Processes to
maintain and
enhance data
• Systems of intelligence moat =
bundling “capabilities” into
processes advantage
25. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 25
MACRO PERSPECTIVE
BUILDING A SYSTEM OF INTELLIGENCE REQUIRES GETTING INTO THE DETAILS
Deep domain expertise Rich, highly relevant data sets
26. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 26
…Achilles heel of megaclouds = lack of
focus on the details of real-world
applications…
27. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 27
Next-generation of successful entrepreneurs
will build systems of intelligence
30. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 30
TL;DR
MACHINE INTELLIGENCE PREDICTIONS
1
2
3
Vertical AI continues to be a disruptive force in the enterprise, with ‘niche’ markets presenting massive
opportunities.
There won’t be a Twilio, but there will be a Github of AI.
Technology advances enable AI applications to expand beyond the limitations of large, well-labeled data
sets.
One caveat: complex vertical AI operating models = protracted path to product/market fit.
Invisible apps will ride consumerization wave faster than Slack.
4
5
32. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 32
MACHINE INTELLIGENCE
…BETWEEN TALENT AND OPPORTUNITY…
Most AI talent works here…
… to optimize the output of…
Meanwhile the world changing
opportunities are out here…
33. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 33
MACHINE INTELLIGENCE
…HYPE AND POTENTIAL…
Potential: 30 years ago Michael Porter predicted an IT
automated value chain we’re yet to fully achieve
Hype: “Chatbots” rose and fell in the course of 9 months…
…no one wanted to talk to a real human, why
would they want to talk to a bot?
Source: Michael Porter, “How Information Gives You Competitive Advantage” HBR
34. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 34
MACHINE INTELLIGENCE
… EXPECTATIONS AND REALITY…
What the press thinks of AI entrepreneurs… How entrepreneurs really feel right
now…
36. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 36
MACHINE INTELLIGENCE
THE OPPORTUNITY IS GETTING MORE OBVIOUS…
Fortune 1000 CEOs… AI Entrepreneurs…
37. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 37
MACHINE INTELLIGENCE
REFLECTING ON LESSONS LEARNED FROM THE RECENT PAST…
AI Masquerade Ball
Is it Ava? Or Jake from
State Farm?
Your smartest people + our
smartest people in a room?
14 months, 150 consultant
“ERP” projects
Action Jack’s black box? Something in between?
?
38. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 38
MACHINE INTELLIGENCE
…AND DIGGING INTO THE ARCHIVES TO SEE WHAT WORKED IN THE PAST…
1960s 1970-1980s
An algorithmic dream AI winter
Late 2000s
Internet Big Data AI Masquerade Ball14 months, 150 consultant
“ERP” projects
1990-2000s Mid 2010s 2016
Is it Ava? Or Jake from
State Farm?
1985 2017+
Sybase PowerBuilder AI-powered business
process revolution
Artificial intelligence…
… meets business process re-engineering
39. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 39
MACHINE INTELLIGENCE
… TO WRITE THE FUTURE PLAYBOOK FOR BUSINESS PROCESS AUTOMATION
2 winning personas for AI-powered enterprise software
Invisible apps Vertical AI
Value prop
Buyer persona
Human
Strategy
Data source
AI
HumanAI
Biz app APIs (i.e. Gmail, Salesforce)
ID the killer app, ride on top of
established data set, create a data
label moat to protect against new
entrants
Packaged software to automate
common business processes
Hybrid app/services to automate company-specific processes.
Value prop
Buyer persona
Strategy
Data source Proprietary IT systems
Implement with pilot customer, facilitate niche search
and user exploration in app to train the AI, ID MVP
that can scale with respect to customer
implementation and sell that before expanding scope
Automate a business ‘task’
Eliminate headcount, make those remaining more
efficient
BU leader/CxOEmployee
Operating model
Operating model
invisible apps move faster; vertical AI is more complex to implement but stickier
40. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 40
MACHINE INTELLIGENCE
INVISIBLE APPS HAVE A MORE OBVIOUS TRAJECTORY
Prediction: By 2018 at least one breakthrough invisible app will
grow faster than the early days of Slack or Salesforce
Invisible apps
•Impact: Dominant force disrupting the workforce over the
next five years because of deadly combination of task
automation + wide reach, ease of deployment of
consumerized SaaS
•Key distinction: end-to-end automation of a business task
so the value proposition is cost reduction. Otherwise
merits of AI = more efficient UX and it’s just a productivity
play like any other SaaS app.
•GTM differentiation: Shorter AI training periods leveraging
structure and rich semantics of biz app data. Busy execs,
consultants and sales people will purchase and expense
access to invisible apps in true consumerized fashion.
Examples of invisible apps:
Buyer persona
Employee
* Work-Bench portfolio company
*
41. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 41
MACHINE INTELLIGENCE
VERTICAL AI IS TRICKIER GIVEN THE MORE COMPLEX OPERATING MODEL…
Algorithm
Annotation
Action
Feedbackloop
Invisible apps
Algorithm
Vendor
Annotation
Action
Vertical AI
Customer
Annotation
Feedbackloop
Defensibility driven by the data moat
Deals with sensitive information and drives functionality changes
requiring company-specific process expertise. Note: not all vertical
AI have customer annotation.
Deals with routine discrepancies
“Lock-in” dynamic with integration services and
customer side annotation
Note: there is some “human-in-the-loop” in that the user’s interactivity
with the software drives model refinement, but the onus is not on the
customer to explicitly “train” the AI like in many cases of vertical AI
42. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 42
MACHINE INTELLIGENCE
…AND HARDER TO GET TO DATA SETS…
Invisible apps are intelligence systems riding on
top of SoRs
Vertical AI require customization to get to legacy systems
and unstructured data
Structured
data
Unstructured
data
43. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 43
MACHINE INTELLIGENCE
…SO THEY’RE SUBJECT TO TOUGHER REQT’S FOR PRODUCT/MARKET FIT
Does it automate a task end-to-end with a
high degree of accuracy?
Invisible apps Vertical AI
Is it easy for employees to use?
Does it significantly augment humans?
Is the UX intuitive and enjoyable to use?
Does the AI require the customer’s
intervention? Who? Data scientists?
Business analysts?
This is arguably the trickiest part to
building vertical AI and thus where
startups should differentiate
What level of abstraction from the guts of
the system will the UX need?
44. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 44
MACHINE INTELLIGENCE
VERTICAL OPPS ARE MASSIVE: EX. FINANCIAL SERVICES
There’s a multi-billion dollar differential
in investment bank cost structure, and
compliance is dominating expenditure
post 2008 financial regulation.
Regulation is such a powerful force on Wall
Street that compliance officers seem to be
running the business and driving divisional
efficiency initiatives.
Outcome: increase speed, reduce human overhead
Vertical AI can help firms reduce
compliance headcount by
automating the mind numbingly
repetitive tasks within compliance:
BU leader/CxO
Business process: Compliance workflows (i.e. KYC, AML)
Industry: Financial ServicesBuyer persona
* Work-Bench portfolio company
*
45. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 45
MACHINE INTELLIGENCE
VERTICAL OPPS ARE MASSIVE: EX. PHARMA
Business process: Pre-clinical informatics research (i.e. drug predictions)
Industry: Pharmaceuticals
Massive pre-clinical research cost ($1B+),
vast timeline (3-4 years), high failure rates
(99.5%).
Opportunity to simulate pre-clinical
processes by identifying molecule
combinations most likely to succeed in
clinical trial.
Outcome: increase speed, decrease R&D expenditure
Vertical AI can help firms speed the
development pipeline and focus
clinical trial efforts on drugs with
higher likelihood of success by
automating informatics research.
Service-provider business model
Proprietary drug pipeline
business model
Source: Joseph A. DiMasi, Henry G. Grabowski, Ronald W. Hansen “Innovation In The Pharmaceutical Industry: New Estimates of R&D Costs”; Harvard Business School “The Medicines Case”
BU leader/CxO
Buyer persona
+ some startups innovating here
46. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 46
MACHINE INTELLIGENCE
VERTICAL OPPS ARE MASSIVE FOR A REASON…
Higher margins =
larger margin differential across firms =
wider gap for AI to be used as a
competitive advantage
Source: Factset
47. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 47
New forces extend the possibilities in
enterprise AI
48. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 48
MACHINE INTELLIGENCE
UNSTRUCTURED DATA PREP = MORE USE CASES, FASTER TO MARKET
Automated data prep has historically only worked
for the 10% of structured data…
Now automated data prep possible for the 90%
of unstructured data in large enterprises…
Result: expansion of possible uses cases within
enterprises as data prep is an initial step in any AI
process
70% of time in AI development spent on data prep
* Work-Bench portfolio company
*
49. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 49
MACHINE INTELLIGENCE
INDUSTRIALIZED DATA ANNOTATION = MORE ACCURATE AI
Deep learning common among AI elite; special sauce turning to data annotation in ebb/flow pattern between data and algos
From a single highly tuned data
annotation console …
…to many highly-tuned data annotation consoles…
Modular enough to be outsourced…
2014 2015 2016 2017
Machine
learning
Data
annotation
Industrialization of data
annotation
Deep
learning
Levelof
Yellow brick road towards
autonomous AI
Industrialization of data annotation
50. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 50
MACHINE INTELLIGENCE
DEEP LEARNING SUBSTITUTES = MORE INFERENCE, LESS DATA + EASIER TO USE
Learning from fewer examples Transfer knowledge between tasks
Deep Learning and more exotic forms of AI are great in theory, but difficult to implement in practice due to the intensive parameter tuning and amount
of data required to train an algorithm…
New trend: make AI methods that require less data more accessible by adding representation schemes from “traditional” ML.
Example: Hierarchical temporal memory
Combining exotic bayesian networks with
common decision tree structure of neural nets to
bring deep-learning like algos to natural
language understanding.
More elastic anomaly detection Put language into context
Example: Deep Forest
Ensemble method in the woodworks that limits
the number of hyper-parameters relative to
deep learning and adds common (albeit vast,
pun intended) decision tree structure.
Ease the training curve Lower the level of expertise required
Expand the possibilities:
51. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 51
MACHINE INTELLIGENCE
BAYESIAN LEARNING = PARTICULARLY PROMISING AS COMPLEXITY RISES
As the industry continues to explore more complex
machine learning challenges…
… The need for easier to use substitutes for deep
learning like bayesian learning will rise
Experthumaneffort
Problem difficulty
Deep learning
Bayesian learning
52. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 52
MACHINE INTELLIGENCE
COMPETITIVE FORCES IN AI-POWERED SOFTWARE ARE CHANGING
Data moat? Algorithmic differentiation? Data product differentiation?
Weaker: still a significant barrier, but
it’s faster to develop and thus harder
to sustain a data moat.
Weaker: tough to sustain with open
source, but there is some value in
novel training, profiling, debugging,
and testing processes.
Competitive forces will be in flux as the AI landscape continues to develop at rapid speed.
Here is where things currently stand and directionally where they are going:
Stronger: The key value driver
moving forward is developing
products bottoms up, from data
and analytical capabilities to
features and user experience, and
creating a virtuous loop between
the two.
Direction = whether this factor will be more or less significant 12-24 months from now
Locus of focus shifting from the
quantity you own to the process
you use to sustain these assets*
Example: Merlon intelligence
designs its automated compliance
workflow software to BOTH shorten
insights to action and gather
feedback from users as new data
that feeds into the models.*For more on this topic, see Matt Turck’s “The Power of Data Network Effects”
53. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 53
MACHINE INTELLIGENCE
FORGET AI ELITISM, MAINSTREAM NOW BETTER EQUIPPED/EAGER TO BUILD AI
“Suits and Hoodies” at Goldman Sachs
•Ambitious attitudes: “AI is a competitive differentiator. We
want to own the model, we don’t want Palantir to own it.”
•Smart recruiting tactics: Avoiding talent wars with the web-
scales by sourcing data scientists in India rather than US, and
Masters-levels rather than PhDs.
•…But some skepticism, particularly around deep learning:
“We have lots of existing regression models that are finely
tuned. Deep learning is just going to be incremental and more
expensive right?”
After missing out on the internet and struggling with mobile, Corporate America wants in early on AI
Source: Quotes from interviews with machine learning executives at top-tier Wall Street banks
54. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 54
MACHINE INTELLIGENCE
WITH A PLETHORA OPEN SOURCE IT WILL BE EASY, RIGHT?
Out-of-the-box deep learning
with differentiated AI training
Widely available backend computing libraries… … Intuitive interfaces
Good for recursive neural nets Good for convolutional neural
nets, speedy/flexible but no
support for Keras
Higher-level APIs
55. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 55
MACHINE INTELLIGENCE
WELL THERE’S A STEEP LEARNING CURVE…
Source: S. Zayd Enam, Stanford AI Lab
Root-cause analysis in AI is vastly more complex than regular
software…
Algorithmdesign
Implementation
Two dimensions of
investigation:
• Algorithm design
• Implementation
Four dimensions of investigation:
• Algorithm design
• Implementation
• Choice of model
• Data
It takes more experience to debug AI efficiently than to debug ‘traditional’ software
+ longer time cycle testing the fix
Software development = hours
Machine learning = days
Why?
• Re-training algorithm on dataset is time
consuming, pushing a code change to
production is not.
56. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 56
MACHINE INTELLIGENCE
…AND BRICK WALLS SILOING INTERNAL EFFORTS = LACK OF LEVERAGE
Business Unit #1
•Enterprise data science functions are decentralizing to get more funding/buy-in from across the enterprise.
•Most organizations lack culture of collaborative data exchange, and data governance teams slow projects down.
•Data
•Algos
Common organizational processes mean algos can be used
across business units to solve different use cases
Data governance overlords
Business Unit #2
•Data
•Algos
Business Unit #3
•Data
•Algos
+
Siloed data science initiatives
= Enterprise are not getting
leverage with their data science
efforts
57. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 57
MACHINE INTELLIGENCE
DISPARATE DATA SCIENCE OPERATIONS = NEED FOR SEAMLESS PIPELINE
‘Central intelligence’
• Small group of high paid
technical experts responsible
for initial model development
Product owners
Backend AI trainers
• Less sophisticated/cheaper/
often internationally based data
science support function
responsible for preparing data
and hyper-parameter tuning
models developed by ‘central
intelligence’
• Domain experts responsible for
application of models in BUsDomain expertise
Training data Production-ready models
IT ops
This role is the newest addition to the
enterprise data science function highly
underserved from a SW perspective
• Infrastructure owners responsible
for deploying models in a cost-
efficient manner
How do you manage
resources and costs running
complicated deep learning
models?
How do you keep everyone up to speed
on latest model commits and updates?
App Dev
APIs
58. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 58
MACHINE INTELLIGENCE
ML PIPELINES EVOLVING INTO A PLATFORM TO BUILD & DEPLOY ML
Historical lineage of ML models ready
to be leveraged across the org
No-ops deployment
Modeling tools and platforms
Git push like model employment
with smart orchestration and
serverless computing
Data
Data lake
Data store, version control,
parallel processing
Algos
Tools to build, optimize, language-
convert, containerize, and data connect
ML models. Level of abstraction varies
depending on the role targeted (i.e.
less sophisticated data scientists vs.
hardcore deep learning engineers)
Collaboration
= area for differentiation for vendors
ML platforms help enterprises centralize, reuse, and deploy their models at scale.
Value will be in tight integration of ML workflows spanning the entire pipeline.
* Work-Bench portfolio company
*
*
*
59. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 59
MACHINE INTELLIGENCE
WHAT IT MEANS: FUTURE = GITHUB + HEROKU FOR AI, FUTURE ≠ TWILIO FOR AI
No-ops deployment
Modeling tools and platforms
Data
Data lake
Algos
Collaboration
= area for differentiation
With deflationary pressure from open source, we expect
MLaaS or “Twilio for AI” vendors with differentiated IP and
talented teams will try to pivot towards ‘Github for AI’, but
will most likely get acqui-hired or resort to selling their data
sets to sustain their business.
YES NO
acquired by Cisco
* Work-Bench portfolio company
*
*
*
61. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 61
TL;DR
CLOUD NATIVE PREDICTIONS
1
2
2017 is shaping up to be a pivotal year for Fortune 1000 deployments of cloud native infrastructure.
Container orchestration is the ‘VMware’ anchoring the cloud native ecosystem. Exactly who will play
this critical role will become clearer this year.
3 Cloud native is reshaping databases, middleware, big data, developer tools, and business models.
62. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 62
CLOUD NATIVE
APPLICATION INFRASTRUCTURE TRANSFORMATION = WELL UNDER WAY
Yesterday
Apps
on VMs
Containers
on VMs
Containers
orchestrated on
bare metal
The container disruption = slowly shifting enterprise infrastructure away from virtual machines (VMs)
Today Tomorrow
63. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 63
CLOUD NATIVE
WHY CONTAINERS OVER VMS?
Containers are lighter — 10’s-100’s of MBs vs. multiple GBs, just the right size for component based microservices
Containers are faster — they can be spun up and down in seconds vs. minutes to realize the true agility, resilience,
and portability of cloud computing
Containers are more efficient — you can fit 4-8 times as many app components (or microservices) on a bare metal
container server than you can on a VM because of the way containers share OS resources to free up space
Container are simply a better unit of deployment for the cloud than VMs
64. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 64
CLOUD NATIVE
A NEW CLOUD NATIVE STACK IS BEING DEVELOPED AROUND CONTAINERS
Developer crave for speed and simplicity combined with 4-8X potential server efficiency gains across $726B in global IT
infrastructure spend more than justifies the new economy of container-centric IT infrastructure dubbed “cloud native”
Sources: Forrester Research, Cloud Native Computing Foundation/Redpoint Ventures
65. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 65
CLOUD NATIVE
WHY THIS MATTERS FOR THE FORTUNE 1000
World’s largest custodian of
assets to be largest non web-
scale to go cloud native
•Use case: mission critical
internal workloads and
partner facing developer
platform called “NEXEN”
•Business case: transactional
velocity, cost cutting
•Tool & vendors: Apache
Mesos, Kubernetes,
OpenStack, Docker
Major media company goes “no-ops”
with self-service cloud native PaaS
•Use case: rapid-application
development platform to meet
demands of its deadline-driven
business
•Business case: developer productivity
•Tool & vendors: Kubernetes, Docker
Major education company goes cloud
native to efficiently scale its growing
customer base
•Use case: core digital learning platform
•Business case: rapid scalability
•Tool & vendors: Kubernetes, Docker
66. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 66
CLOUD NATIVE
FORTUNE 1000S WANT DEVELOPMENT AGILITY LIKE THE WEB-SCALES…
Microservices fulfills on the promises of service-oriented architecture by decoupling apps into single-purpose services that
communicate with other microservices via APIs or messages
Sources: PWC “Agile coding in enterprise IT: Code small and local”
Speedier app delivery
Rapid cycle times and easier
updates
Higher code-to-server density ratio
67. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 67
CLOUD NATIVE
…AND TO SCALE AS EFFICIENTLY AS THE WEB-SCALES TOO
Average server utilization by type of environment
10-15%
50-70%
Cloud-native
Virtualized
Sources: Gartner, Codeship
Non-virtualized
5-10%
?
Hallmark vendor
Infrastructure type
Example users
68. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 68
CLOUD NATIVE
DESPITE ORGANIZATIONAL HURDLES ENTERPRISES NEED TO OVERCOME…
“Why the f&ck don’t we use
Kubernetes?”
Tech leadership to ops Ops
“…”
“We’re going to be placing you on a special
projects team to migrate all of our workloads
over to Kubernetes… After that, we’re going
to have to let you go.”
Devs
The cloud native organizational disconnect = ops getting the short end of the stick
69. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 69
CLOUD NATIVE
…2017 IS SHAPING UP TO BE A PIVOTAL YEAR FOR CLOUD NATIVE ADOPTION
• Container adoption is crossing the chasm: 11% of global developers reported using Docker containers for
deployments in late 2016.
• Megaclouds are increasing their integration and support for container orchestration: Amazon natively integrates
with Mesos, Microsoft Azure container services supports Kubernetes, Docker Swarm, and Mesos, Google naturally
integrates with Kubernetes, and even Oracle now supports Kubernetes!
• Developers are embracing new programming models like functional pipelines (i.e. serverless) and the agent
model to ease their migration to microservices.
Sources: Forrester Research, Google Trends
“Serverless” on Google Trends
71. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 71
CLOUD NATIVE
ORCHESTRATION = CENTRAL NERVOUS SYSTEM OF CLOUD NATIVE = $$$
Container orchestration tools are the data center operating systems of the future. They automate container deployments by
spinning up and managing deployment of containers in production applications to fully realize the agility, resiliency, and
portability benefits of containers.
Major OSS container orchestrators
Best bet for greenfield apps
• Largest open source initiative by Google
• Fully featured orchestrator for enterprise apps
• Several commercial vendors in ecosystem as with
Hadoop
• Major Partners: Google, Rackspace, RedHat, Intel,
CoreOS, Oracle
Most mature solution for scale out apps
• More mature project than Kubernetes and Nomad
• Integrates well with existing Hadoop stack
• Not so self-service: bring your own service discovery,
highly skilled operators, and maintenance staff
• Major Partners: Microsoft, HP
À la carte option for running micro
services on existing infrastructure
• Individual open source projects for service scheduling,
discovery, and secrets management that together are
competitive to Kubernetes
• Existing companies with legacy inertia use Nomad for
service discover and secrets management
• Managed by Hashicorp
72. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 72
CLOUD NATIVE
CONTAINER ORCHESTRATION VENDORS ARE THE NEXT BIG IT VENDORS
Commercial software vendors bring container orchestration to life
LEGACY FUTURETRANSITION
VIRTUALIZED SERVERS
EXISTING PAAS PLATFORMS ADDING
CONTAINER SUPPORT
NEW PLATFORMS BUILT GROUND UP
FOR CONTAINERS
* Work-Bench portfolio company
*
73. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 73
Kubernetes is a disruptive pirate ship —
built by Google, with sails set straight for Amazon
74. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 74
CLOUD NATIVE
A FEW REASONS WHY KUBERNETES WILL WIN THE RACE FOR DATA CENTER OS
#1. Maximizes ease of use as the industry’s
most fully-featured orchestrator
#2. Vibrant open source community
#3. Like with Hadoop, lack of vendor
dominance encourages the community to
freely innovate on OSS foundation
#4. Technically sophisticated stack built from
the ground up with the right level of
abstractions for users to build and deploy
applications using containers
Kubernetes-first OSS projects emerging
“Istio currently only supports the Kubernetes platform,
although we plan support for additional platforms such
as Cloud Foundry, and Mesos in the near future.”
76. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 76
CLOUD NATIVE
DATABASES FINALLY CATCHING UP WITH DEMANDS OF CLOUD NATIVE
Google Spanner
Cloud native needs databases that can keep up. Problem = databases are sluggish beasts that never quite benefitted from the
pace of innovation the rest of the industry enjoyed.
Former Google VP of Infrastructure Eric Brewer summarized the engineering challenges of developing database infrastructure
with the CAP Theorem: you can only achieve two of the following guarantees for your database: 1) transactional integrity,
2) availability, 3) and scalability.
Until now. New solutions emerging that dispel CAP Theorem:
Transactional integrity Availability Scalability
Microsoft Cosmos CockroachDB
* Work-Bench portfolio company
*
77. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 77
CLOUD NATIVE
APPLICATION MONITORING CATCHING UP AS WELL
Buoyant
New problems… + new enablers…
Network-riddled app logs
Latency in event time stamping
Distributed tracing techniques and
community support
Machine learning
Powerful stream processing
Interbred app-service dependencies
* Work-Bench portfolio company
*
79. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 79
CLOUD NATIVE
… POINT TO ANOTHER ROUND OF APM CONSOLIDATION?
Will new monitoring entrants evolve standalone or will APM leaders AppDynamics/New Relic lead the charge?
On-premise: fragmented APM Cloud: consolidated APM
Cloud Native: early
monitoring fragmentation
Buoyant
Cloud Native: consolidated
APM?
?
Either way, both categories expand with shift to cloud native
80. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 80
CLOUD NATIVE
WITH CLOUD NATIVE, SOFTWARE EATS MIDDLEWARE
Middleware
• Then: App servers, complex event processing
servers, and ESBs were complex new
technologies in the first generation of cloud
requiring dedicated servers to manage them.
• Now: many of these functions are now
distributed directly across the application
code, thinning the traditional middleware
layer in the stack.
81. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 81
CLOUD NATIVE
NEW PROGRAMMING MODELS ABSTRACT MIDDLEWARE FUNCTIONS INTO CODE
Pivotal Cloud
Foundry Spring
Cloud Services
Microsoft
Azure
Functions
Microsoft Azure
Service Fabric
IBM Bluemix
OpenWhisk
Amazon
AWS
Lambda
Google
Cloud
Functions
Functional pipelines (aka “serverless”)
= no more complex event processor
Ephemeral snippets of code govern app data traffic, rendering the need for
dedicated services for business rules obsolete.
Actor model = no more bloated app servers
Software actors implement concurrency via asynchronous messages and spread proxies across
physical servers to manage consistency. App servers are thus relieved of resiliency duties.
Middleware, once a core layer of the IT stack, is shedding significant weight as middleware
functions now reside in distributed code.
Iron.io Serverless
82. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 82
CLOUD NATIVE
SERVICE MESHES ABSTRACT NETWORK FUNCTIONS VIA LIGHTWEIGHT PROXIES
Service discovery is the new IP
address and DNS
New system of dynamically routing services to manage latency in large scale
distributed systems
gRPC and REST are the new TCP/IP
Service meshes use new protocols developed for communications at the service level rather
than underlying network
Service meshes are lightweight network proxies governing service-to-service communications for tasks such as service
discovery, load balancing, and monitoring in highly complex distributed systems
LinkerD Buoyant Istio
83. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 83
CLOUD NATIVE
MIDDLEWARE MARKET ISN’T DEAD, IT’S JUST EVOLVING
Replacing middleware pipes is a new software-human middleware layer tackling the more specialized functions of
complex modern apps:
Infrastructure
Systems of engagement
Next-generation middleware
ML pipeline
Helping data
scientists add
intelligence and
automation to
software
Ingesting and
interpreting real-time
information from
around the world
Streaming platform
84. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 84
CLOUD NATIVE
STREAMING IS THE NEW COMPLEX EVENT PROCESSING SERVER & ESB
Streaming platform
Data lake/distributed database
Code
Container OS
Container engine
Container orchestration
External integrations
Operations and container image
management
Businesslogicas
distributedcode
“Data stack” Cloud Native stack
ML pipeline
Stream
processing
Messaging
system
Code
Next-gen complex
event processing server
Next-gen ESB
Lightweight network proxies for load balance
and service discovery
85. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 85
CLOUD NATIVE
THE RACE IS ON FOR STREAM PROCESSING
It’s a story of multi-purpose convenience vs. purpose-built performance, with support for
cloud-native schedulers becoming a must
•Already considered
“legacy” in Silicon
Valley with Spark
demonstrating
considerably more
more horsepower
• Doesn’t work out of
the box at scale and
frustrating to set up
and manage
Streaming native systems Multi-purpose batch systems with streaming bolt-on
• New streaming
library on popular
distributed log with
mid-2016 release
• Unproven scalability/
stability, support for
cloud-native
schedulers
• New streaming
library developed at
Twitter with promise
of better scalability
and manageability
than Storm
• Architecture
supports cloud-
native schedulers
and Storm migration
• One stop shop for
batch, streaming,
and ML that plays
well with Hadoop
• “Near” real-time
streaming is good
enough but not great
with respect to scale,
throughput, and
latency
• Streaming and
batch in one
system incurs
latency
• Limited
production use
cases and
unclear
development
path
• Tied closely with
YARN architecture
• Latency issues as
a multi-purpose
system
87. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 87
CLOUD NATIVE
STREAMING INTEGRATIONS = DATA AND APPS WILL LIVE IN ONE STACK
Streaming platform
Data lake/distributed database
Code
Container OS
Container engine
Container orchestration
External integrations
Operations and container image
management
“Data stack” Cloud Native stack
ML pipeline
Stream
processing
Messaging
system
Code
Cloud native schedulers take on
packaging and deployment of big data
workloads
Specialized processing libraries
instead of all-in-one clusters tied
to a specific scheduler
Data and app stacks have been separate until now… Container orchestrators like Kubernetes and
Mesos distribute data workloads better than Hadoop’s Yarn. Spark, Kafka, Herron and other new
school stream processing engines all integrate directly with container orchestrators.
88. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 88
CLOUD NATIVE
MORE BROADLY, HADOOP IS LOSING ITS DOMINANCE
Markets demands moving to machine learning,
where Hadoop has no shine
•Spark > ML for Hadoop, and with cloud-based
object store, you don’t need Hadoop for Spark.
•ML is best run on highly specialized chips
like Google’s TPUs and NVIDIA’s DGX-1 rather
than the commodity hardware Hadoop was
developed for.
Megaclouds ate Hadoop’s lunch
•Megaclouds use the Hadoop distribution in their cloud services,
but by unbundling the underlying file system (HDFS) from the
cluster manager (YARN) and making these components inter-
changible with alternatives, Hadoop is losing it’s position as a
central nervous system for the data stack.
89. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 89
CLOUD NATIVE
SERVERLESS COMPUTING MAKES FINANCIAL SENSE OF MICROSERVICES…
App components
Resource
utilization
“Pay-as-you-go”: theoretical maximum
utilization of infrastructure
Spend per server: “pay-as-you-go” vs. serverless
“Serverless”: actual instance run rate
idle time
run time
Microservices = more shallow utilization across a wider footprint = uneconomical with
sever-based units of measurement in “pay-as-you-go” business model
Serverless lowers operating costs for software vendors.
Still TBD whether vendors decide to pass these savings down to customers.
90. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 90
CLOUD NATIVE
…WILL IT FORTIFY MEGACLOUD LOCK-IN OR DISSOLVE IT?
Amazon sees Lambda as another form of lock-in.
It wouldn’t be trivial for Amazon to change their posture
because architecturally, functions are tied to AWS public
cloud and it would take extensive work with partner
VMware to extend functions into private cloud.
Google wants to make functions more
extensible to promote multi-clouds and
combat Amazon’s lock-in grip. They
hope to commoditize AWS by lowering
switching costs with serverless.
Google Functions
91. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 91
CLOUD NATIVE
ML WILL SOON PLAY AN INTEGRAL ROLE FOR INFRA OPS AND APPDEV
ML-powered cluster management
Improving resource efficiency by
adding a Netflix-like recommendation
system to allocation of app
components across resources
Hyperpilot
ML-powered static code analysis
Analyzing code commits to figure
out how to better allocate developer
resources across the organization
ML-powered IT ops
Bots for new employee on-boarding
tasks
Electric.ai
IT is by nature a data-driven organization, making it the perfect function to infuse with the power of AI
Examples in the market:
* Work-Bench portfolio company
*
93. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 93
TL;DR
SECURITY PREDICTIONS FOR 2017
1
SecDevOps blurs the lines between networking and application security as the race for cloud-native
security products intensifies.
2
3
Beyond the 1%: SOIs as consumable microservices will bring advanced security technology to the 99% of
companies who previously couldn’t afford.
The security ecosystem is re-organizing itself into Systems of Intelligence (SOI). Systems of record (SORs)
must become SOIs or risk being relegated to “plumbing.”
In the sweeping wave of industry consolidation, legacy security companies will buy up security analytics
and Security Operations, Analytics, and Reporting (SOAR) companies in yet another bout to stay relevant.
4
95. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 95
ENDPOINT
CYBERSECURITY
LAST 5 YEARS = EVOLUTION FROM SECURITY PRODUCTS TO SOR PLATFORMS
ANTIVIRUS SIEM FIREWALL DLP IAM MALWARE
It used to be about product oligopolies…
NETWORK // HOST
APPLICATION // CODE
Now the center of attention is around a new breed of monopolistic Systems of Record
platforms assembling themselves around layers of the IT stack…
?
* Work-Bench portfolio company
*
96. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 96
CYBERSECURITY
SECURITY ECOSYSTEM EVOLVING INTO A SYSTEM OF INTELLIGENCE
How do you get the most comprehensive observability of IT systems
in the most seamless fashion?
**Note: each value driver is sized based on its ability to create sustainable competitive advantage.
Original framework source: Jerry Chen (Greylock Partners) “The New Moats”
…and have a little bit of this
SORs are getting here…
How do you make sense of and take action based on the wealth of new
information generated by modern security systems?
With new security tool overload and highly understaffed sec orgs,
how do you make workflows more seamless? How do you foster
collaboration amongst security teams? Can you use automation?
AI
Domain
expertise
Data-driven
product design
Data
Primary value driver**
SORs have a natural first mover advantage to put all the pieces together
* Work-Bench portfolio company
*
97. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 97
CYBERSECURITY
SYSTEMS OF INTELLIGENCE PRINCIPLES BEHIND BREAKTHROUGH SORS
Ease of implementation and management
Scalability in complex IT environments
Short time-to-value
ENDPOINT
NETWORK
APPLICATION
How do you seamlessly instrument into IT systems?
How do you create a System of Record platform?
How do you build great data-driven products with
attributes CISOs really care about?
• Agents on web servers or application run-times
• In-line forward/reverse proxy or agent
• Traffic access point off firewall or web proxy
• App APIs
• Distributed sensors that act like agents but
are decoupled from underlying hardware
• Traditional software agents
• Traditional software agents
Get better data… …to build a better product:
First you land customers with a single product, the
“thin edge of the wedge.” Designing a product starts
with what type of data you can peek into:
Data
Data-driven
product
design
98. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 98
CYBERSECURITY
…AS THE BASIS FOR BEATING LEGACY CO'S AT PRODUCT OLIGOPOLY GAME
You expand beyond the “thin edge of the wedge” by leveraging data/instrumentation advantages to extend product scope
and displace product-centric companies in adjacent categories.
Instrumentation: lightweight endpoint agents collaborating in a distributed system
POLICY MGMT INCIDENCE
RESPONSE
CONFIGURATION
MGMT
OS PATCH
MANAGEMENT
Land
Expand Expand
VULNERABILITY
MGMT
POLICY MGMT DECEPTIONFIREWALLMICRO-
SEGMENTATION
Land
Expand Expand
SECURITY
ANALYTICS
Instrumentation: distributed network sensor system
Distributed security architecture is
a common thread here because it
brings outsized speed & scale to
the process of obtaining data to
build a host of security “products.”
* Work-Bench portfolio company
*
99. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 99
CYBERSECURITY
REASONS WHY A PLATFORM STRATEGY IS SO CRITICAL TO SUCCESS…
• Long term sustainable competitive advantage.
With superior performance characteristics and a
growing treasure trove of data, you can evolve with
the rapidly changing industry pace, much like the
ancient force trumps the latest imperial weapon. It’s
difficult to stay competitive with a single-product
strategy because hackers will eventually figure out
workarounds, rendering your product obsolete.
FireEye’s uncertain future is a case in point.
• Cost-reduction value proposition. Platforms are
stickier than products. The comprehensiveness of
your platform allow customers to rip-and-replace
their old tools in a cost reduction play.
100. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 100
CYBERSECURITY
SOR LANDSCAPE = MORE COMPLICATED AS NETWORK/APP LINES BLUR
NETWORK // HOST
TL;DR: Network and app layer are looking to achieve the same goal of bringing X-Ray vision of apps to
security. Culture + technology factor into this shift. In startup race, network/host layer leaders have first
mover advantage over new entrants to gain X-ray vision up the stack.
APPLICATION // CODE
101. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 101
CYBERSECURITY
DEVSECOPS: THE CULTURAL ASPECT OF NETWORK/APP BLUR
App Dev
VP Infrastructure
• Owns the Systems of Record (including legacy security tools)
• Assembling the Systems of Intelligence to bridge the
divide (creating a pane of glass for security)
CISO
• Making pane of glass more complicated by using new tools with rich
insight into app activity without VP Infrastructure buy-in
App Dev is bringing new infrastructure and tools to the table, so security teams must
keep up with the rich insights into applications these tools generate
102. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 102
CYBERSECURITY
ABSTRACTED INFRA: THE TECHNICAL ASPECT OF NETWORK/APP BLUR
NETWORK
ENDPOINT
Old-world app X-ray vision = server = mostly network data in firewall + some data from endpoint
New-world app X-ray vision =
LAYER 7?
OS HOST?
RUNTIME?
CONTAINER?
Dispersed across more abstracted infrastructure Hidden in data-rich DevOps tools
Container orchestration
CI/CD toolchain
X-ray vision
103. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 103
CYBERSECURITY
BLURRING NETWORK/APP LINES CREATE NEW BATTLE FOR CLOUD NATIVE SEC
NETWORKING
OS HOST
CONTAINER
CODE IMAGE
New “app dev first” entrants = “thin edge of the wedge”
point products with hopes to become new Systems of
Record in cloud native security world.
Host layer Systems of Record = extending product
capabilities to ensure compatibility with containers and
cloud native architecture
Thin edge = vulnerability managementThin edge = WAF bandaid
CONTAINER ORCHESTRATION
With application logic distributed across individual
microservices callable via APIs on the network, east-west
traffic visibility via deep packet inspection is critical
Security tools must limit network activity between containers
running on distributed hosts and observe communication
interdependencies between containers on the same host OS
* Work-Bench portfolio company
*
*
*
*
*
104. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 104
CYBERSECURITY
SOR PLATFORMS WILL HAVE TO EVOLVE INTO SOIS TO REMAIN COMPETITIVE
**SOAR: new term dubbed by Gartner for “Security Operations, Analysis and Reporting” technologies that support workflow management.
Note: these categories are not mutually exclusive in that several Systems of Record vendors have Systems of Intelligence capabilities and vice versa.
Original framework source: Jerry Chen (Greylock Partners) “The New Moats”
Or these guys?
Chicken and egg problem:
how do we partner to get
the SOR data?
Do we acquire one of
these guys?
Security analytics
SOAR**
* Work-Bench portfolio company
*
*
105. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 105
CYBERSECURITY
SOI ENTRANTS = SECURITY ANALYTICS, ON THE HUNT FOR DATA
Security analytics work across Systems of Record (SORs) to make sense of all the data. With SORs
developing security analytics capabilities themselves, they must prove out the value of generating
insights across SORs if they are to endure as independent vendors.
Acquired by Oracle
Acquired by HP
Example: Versive partners across the SORs to build a “deep
and wide” data moat in security analytics
Deeply instrumented
in the data center
Best visibility into mission
critical workloads
Flexibility to pull
data from end user
devices selectively
* Work-Bench portfolio company
*
*
*
106. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 106
CYBERSECURITY
SOE ENTRANTS = SOAR, SECURITY WORKFLOW EXPERTS
Security Operations, Analytics, and Reporting (SOAR) tools automatically run playbooks for common security
workflows, freeing up limited analyst bandwidth to handle the more niche cases. It’s still to be decided whether
they meaningfully penetrate the enterprise market directly or power the next generation of managed
security service providers as CISOs increasingly outsource analyst work.
Example: Demisto is developing a workflow tool that
integrates existing security tools
Sources: Demisto
108. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 108
CYBERSECURITY
THE SORS MUST EMBRACE CREATIVE DISRUPTION AND SHIFT INTO SOI
Step 2: de-couple analytics and
policy control capabilities into a
marketplace of “callable” security
function modules that leverage a
broad swathe of SOR data.
Step 1: commoditize SORs into
backend “data feeds”
Step 3: develop an application
platform for SOEs to build on top
of the SOIs. Splunk’s Splunkbase is
an early illustration of this concept.
SOR landscape getting complicated and competitive. New SOIs are coming in. SORs
must move “up the stack” and embrace new operating models that commoditize their
very crown jewels.
109. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC
$0
$20,000
$40,000
$60,000
$80,000
$100,000
$120,000
$140,000
0 50000 100000 150000 200000 250000 300000 350000 400000
Averageordersize
Customer count
CHKP
FEYE
IMPV
PANW
SYMCFTNT
PFPT
Blue Coat
MIME
CYBR
109
CYBERSECURITY
SOI WILL EXPAND MARKET OF LOW COST SECURITY SERVICES
But what if we could deliver SOI functions as callable microservices on serverless
backends to lower the cost of delivering security services?
Cutting edge security technology has historically been prohibitively expensive
More Expensive /
Narrow Customer Set
Less Expensive /
Broader Customer Set
Source: Company data, Goldman Sachs Global Investment Research
110. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 110
CYBERSECURITY
WHAT WILL HAPPEN TO LEGACY SECURITY COMPANIES?
•Legacy security companies need to shift from peripheral to disruptive
acquisitions. Ex. CASB acquisitions from 2016 do not place incumbents directly
into the heart of the cloud, but buying a System of Record platform startup will.
•Most likely outcome this year and and next is legacy security companies buy
Security Operations, Analytics, and Reporting (SOAR) startups to put
themselves closer, but not fully embedded in the cloud IT stack.
•Paradox of legacy companies and PE firms buying and integrating security
products is that it only brightens the spotlight on independent SOAR and
security analytics vendors who differentiate by casting a wider net than
any multi-product suite.
•Democratization of machine learning may swing the pendulum in favor of SIEM
vendors who can build an intelligence moat around their legacy SORs.
Security incumbents have been busy buying cybersecurity startups — is M&A really a silver bullet?
111. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 111
Vertical Theme #4
Internet of Things (IoT)
112. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 112
TL;DR
INDUSTRIAL IOT PREDICTIONS
1
Distributed analytics will be critical for remote/low-bandwidth industrial IoT operations.
2
3
IoT is potent for competitive advantage amongst industrials like gunpowder was for kingdoms of the
1200s.
Industrial IoT = earlier than most of us think because distributed infrastructure remains in its infancy.
Security for IoT will spawn directly from distributed analytics architectures.4
The next frontier is systems management software bridging disparate IoT software systems.5
114. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 114
INTERNET OF THINGS
WELL, EVERYONE TALKS ABOUT IT…
Predictive maintenance of equipment can save massive amount of time and cost
63% reduction
in maintenance time on site
$340K-$1.7M loss per day
due to shutdown
$11M loss per day of
unplanned downtime
Oil & gas refinery Natural gas drillerBuilding security systems
115. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 115
INTERNET OF THINGS
…NOBODY REALLY KNOWS HOW TO DO IT
IoT product platforms have been in market for years
with an established approach…
…the industry just assumed Industrial IoT would work similarly.
Closed system software stack with broad protocol support and prepackaged apps for asset
management, alert management, product relationship management, and workflow management
116. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 116
Ample in home environment Low-bandwidth in remote environment
Homogenous time-series data streams
from a couple sensors
Heterogeneous data from the drones,
drills, and operational databases
Business rules, ML-based anomoly
detection
Complex, event-driven analytics
INTERNET OF THINGS
PRODUCT VS. INDUSTRIAL IOT
The connected washing machine The autonomous drone in a sensor-laden oil field
Connectivity
Data
Analytics
Product Industrial IoT
117. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 117
INTERNET OF THINGS
…EVERYONE THINKS EVERYONE ELSE IS DOING IT
John Deere has mastered connected farming
operations
Sure dude…
What does the connected cow have to say?
118. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 118
INTERNET OF THINGS
SO EVERYONE CLAIMS THEY ARE DOING IT…
• Forrester: 25% of enterprise IT claims to be using IoT software platforms, 33% of
global enterprise developers reported building IoT applications in 2017.
• BCG/IDC: Insights from clients points to €250B in annual spend by 2020.
Sources: Forrester Research, “Winning in IoT: It’s All About the Business Processes” by BCG
119. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 119
Putting hype aside… IoT = new era of
industrial competitive advantage
120. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 120
MACRO PERSPECTIVE
GE IS ALREADY LEADING PACK OF INDUSTRIALS BECOMING TECH COMPANIES
In 2012, GE and followers set out to
create software systems to improve
internal operational efficiency
121. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 121
INTERNET OF THINGS
THEY SOON REALIZED WHAT THEY WERE BUILDING WAS UNIQUE
Megaclouds will scale your application
up, down, and side-ways…
… but have no idea how to bring software over here…
Industrials are building IoT platforms — highly specialized PaaS with modules for industrial processes
such as asset productivity, operations scheduling, maintenance, and product delivery for their clients
122. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 122
INTERNET OF THINGS
INDUSTRIAL CHARTER = DEVELOP SYSTEMS OF INTELLIGENCE
Domain expertise
Data
AI
Data-driven product design
Industrials have the opportunity to evolve their software platforms into powerfully defensible systems
of intelligence
Industrial systems of intelligence
?
123. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 123
INTERNET OF THINGS
MORE IMMEDIATELY, RACE IS ON FOR SOFTWARE PLATFORM DOMINANCE
The Palantir of Industrial IoT
• Services company helping industrials like
John Deere develop their own systems of
intelligence for IoT
Developing portfolio of IoT software
• Differentiating with distributed analytics capabilities
and blockchain for P2P transactions.
• Bought Tririga for facilities management applications
Parlaying Azure portfolio towards IoT
• Microsoft combined an IoT device
management platform with its robust portfolio
of streaming analytics, and easy-to-use machine
learning services to develop IoT software for
predictive maintenance and remote monitoring
• Recently announced edge analytics for
distributing analytics processing across devices,
gateways, and the cloud
Strongest in distributed computing with
Greengrass and Lambda
• Greengrass adds a smart app server to IoT
gateways to enable distributed computing
• Amazon’s Lambda functions govern business logic
and manage device state across distributed systems
Analytics chops and industrial customer base
• Streaming analytics capabilities via SAP HANA
• Developing modules for predictive maintenance and
asset management
• It has one of the strongest industrial customer bases
in the tech sphere with its ERP heritage, but with a
very different type of buyer it is not clear this will give
them an advantage in IoT
Strong player for remote, low-bandwidth scenarios
relying on cellular connections
• With Jasper Technologies acquisition, strongest
network of telcos to better manage cellular data fees
in remote locations
• Acquisition of ParStream is a catalyst for Cisco to
develop edge analytics capabilities needed for
remote area IoT
Most mature industrial IoT software platform
• Built on Cloud Foundry with easy migration between on-premise and cloud
• Developing analytics capabilities with acquisitions of wise.io for machine learning and Bit Stew for data ingestion and transfer
• Planning to better automate services with Servicemax acquisition
• Strong network of SIs and partners
• Needs to developed more pre-packaged software modules
124. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 124
INTERNET OF THINGS
SKILL SETS ARE GOING TO COME TOGETHER
Can these guys learn to run technology businesses?
i.e. developer evangelism, partnerships and integrations
Can these guys learn industrial processes?
Seems doubtful…
So we’re starting to see healthy signs of cooperation
GE for application PaaS Microsoft for IaaS
126. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 126
Endpoints as massive P2P data centersData centers as central control
Edge servers as bridges
Mini data centers in cell towers
IoT gateways
INTERNET OF THINGS
TIDES ARE TURNING AWAY FROM CLOUD AND BACK TO THE EDGE
Compute resources will slowly shift from the cloud to devices
127. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 127
INTERNET OF THINGS
CONSUMER TECH TRENDS + NEW EDGE PROCESSES = READY FOR EDGE ML
New NVIDIA Jetson TK1 is
built for computer vision, ML,
NLP on a variety of small
devices
Nest proved out local ML data
processing…
…as did Google Now
+
128. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 128
INTERNET OF THINGS
GATEWAYS ARE LOCAL FUELING STATIONS BRIDGING DEVICES AND CLOUD
Bridges gaps in:
•Networking throughput that render real-time data processing too slow for the cloud
•Compute for local data preprocessing that may be too resource intensive for endpoints
•Intermediary data store for efficient, spoke-hub distribution of sensor data
1010101010101010101010101010 1010101010
1010101010101010101010101010 1010101010
1010101010101010101010101010 1010101010
1010101010101010101010101010 1010101010
Gateway
50B devices in 2020
Source: Cisco IoT Report
Spewing data
streams >>
Centralized cloud
129. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 129
INTERNET OF THINGS
GATEWAYS ARE BEING INFUSED WITH SOFTWARE TO ENABLE EDGE COMPUTING
Cisco’s Parstream allows for
efficient, spoke-hub distribution
of sensor data at IoT gateways
Amazon’s Greengrass sends functions
with complex event processing rules
for data filtration and synchronization of
digital shadows for managing asset
state across low network environments
Vapor.io retrofits cell towers with
mini data centers for local data
preprocessing that may be too
resource intensive for endpoints and
too time sensitive or prohibitively
expensive to send to the cloud
130. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 130
Analytics at the edge to make
instantaneous decisions. Speed is mission
critical in the case of brake failure detection
on a speeding train, where symptoms show
up in data just minutes before a disaster.
Large wireless data fees to send
to cloud. Worth the cost? Only if
useful for historical analysis.
Utilize gateways when you
can to save on device
battery power drain.
Amazon AWS
Greengrass
Putting it all together — highly distributed IoT operations
Scenario: Brake failure preventative maintenance on remote supply transport train
131. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 131
… Governing data flow will be a new analytics architecture
Scenario: Brake failure preventative maintenance on remote supply transport train
132. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 132
INTERNET OF THINGS
ANALYTICAL TOOLS BUILT FOR THE INTERNET DON’T WORK FOR IOT
Resource intensive
Try putting heterogeneous industrial data streams into traditional big data pipeline…
ETL
Learn
Build
Data lake
Extract value
High latency Loss of critical real-time insights
XXX
133. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 133
INTERNET OF THINGS
ANATOMY OF A DISTRIBUTED IOT ANALYTICS ARCHITECTURE
Industrial sensors
Machine processes
Distributed
instrumentation
Data
labelling
Industrial scale ETL
Stream
processing
Messaging
system
Predictive
maintenance
Inventory mgmt
Asset productivity
Gateway Cloud
Machine
learning
Instant response
Additional context,
data filtering
Deep insights,
model updates
Device
Expert operator
feedback
Data historians
System
architecture
Software modules
134. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 134
INTERNET OF THINGS
CONTENDERS FOR DISTRIBUTED IOT ANALYTICS IN THREE CAMPS
IoT application PaaSPure play startupsMegaclouds
(recently acquired by Greenwave Systems)
* Work-Bench portfolio company
*
135. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 135
INTERNET OF THINGS
SECURITY FLOWS FROM DISTRIBUTED ANALYTICS
Armis
•Distributed analytics architectures instrument deeply into endpoints in the gateway,
and thus will be the providing data to security solutions focused on device anomaly
detection and distributed policy-based prevention.
•Traditional security vendors talk a big game about IoT but they are going to struggle
to get into the industrial space because operators aren’t going to want to instrument
connected assets 10 ways like IT does in the data center.
•Because of this dynamic, distributed analytics vendors have an opportunity to
become security vendors themselves.
• Outside of endpoint and network security, the radio frequency spectrum is a new
topology that the most discerning government agencies and financial institutions
will protect against external IoT “intruders.”
Example startups
136. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 136
INTERNET OF THINGS
NEXT UP: OPPORTUNITY TO BUILD SYSTEMS MGMT LAYER OF INDUSTRIAL IOT
Industrials connecting asset in their supply chain must do the same for software shipped with these assets.
Much like with the rise of systems management software (Tivoli, BMC) in the 90s to help IT more efficiently manage and get value
out of disparate appliances in the data center, a management layer to integrate disparate IoT software stacks will likely emerge.
Connected asset #1 Connected asset #2 Connected asset #3 Connected asset #4 Connected asset #5
OEM/IoT
platform
vendor
Management layer
137. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 137
INTERNET OF THINGS
LASTLY, IOT SOFTWARE STARTUPS SHOULD AVOID DEVELOPING HARDWARE
TL;DR: IoT software startups should focus on use cases in which the underlying
physical assets are already IoT-enabled.
• Vertical AI software is highly specialized, and creating a full stack solution tuned to a
particular use case often means developing proprietary hardware to obtain data from
older, non-IoT enabled physical assets.
• Besides the operational challenge for a startup to set up hardware manufacturing, many
startups we meet are incurring heavier losses than typical vertical SaaS companies at the
same stage because they absorb the hardware cost and just sell the software.
• These startups intend to convince OEMs to manufacture the devices on their behalf. We
believe this wishful thinking because OEMs will not be able to extract enough value from
hardware purpose-built to serve even the largest of vertical application markets.
138. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 138
Lastly, A Few Tips for
Entrepreneurs Going Forward
139. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 139
TL;DR
TIPS FOR ENTREPRENEURS GOING FORWARD
1
2
Systems of Intelligence are the long run combative play against the megaclouds, but there are still ways to
build value in cloud ecosystem.
Enterprises still want licenses. Thwart their demand by undercutting SaaS pricing sooner rather than later.
Systems of Intelligence companies will need a thin-edge of the wedge market entry strategy, for which
there are several models emerging.
3
140. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 140
Megaclouds will be busy duking it out this year…
141. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 141
TIPS FOR ENTREPRENEURS
WHEN MEGACLOUDS HOST OSS PROJECTS, FOCUS ON ‘NICHE’ FEATURES
Google releases Kubernetes to the OSS community…
…Amazon reacts by integrating with Mesos
Offensive play to render
Amazon lock-in obsolete.
Defensive play aligning with
more mature (although less
progressive) alternative
Neither of them are focusing on enterprise
features and support, leaving room for these
pure plays to shine:
* Work-Bench portfolio company
*
142. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 142
TIPS FOR ENTREPRENEURS
WHEN MEGACLOUDS GO COMMERCIAL SERVICE, PLAY AGAINST LOCK-IN…
Google tries the Amazon lock-in approach thinking it
can lure new GCP customers with its innovative new
database product.
Google releases Spanner as a managed service for GCP
Microsoft releases me-too competitor called Cosmos
Private/virtual
private cloud
Pure play Cockroach Labs is well positioned to power
Amazon’s combative strategy and address the massive
landscape of enterprise managed databases.
When megaclouds go play the lock-in game… …Go multi-cloud, multi-region as a differentiator
* Work-Bench portfolio company
*
143. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 143
TIPS FOR ENTREPRENEURS
MEGACLOUD AGNOSTIC PLAY = PERTINENT WHEN SENSITIVE DATA INVOLVED
Google, Microsoft, and Amazon want to enforce vendor lock-in
by developing one-click ML deployment services on their
functional backends…
… But enterprises need flexibility to move ML workloads to
where the data is and not vice versa as megaclouds hope.
Why? Because machine learning and data need to sit together,
often on the same GPU server, and sensitive customer records
can’t just instantaneously be moved to the cloud for a data
science project.
Hence a big opportunity for pure play ML platform startups:
* Work-Bench portfolio company
*
144. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 144
TIPS FOR ENTREPRENEURS
THIN-EDGE OF WEDGE STRATEGIES CRITICAL FOR SYSTEMS OF INTELLIGENCE
Systems of Intelligence have a chicken and egg problem: Customers want proof the power of automation can
help their business and startups need the data to train the system so it can actually deliver on that promise.
• The DVR player: Lightweight version of the product that takes historical data from a customer and delivers
insights in retrospect. This approach provides the necessary training data and proof points to convince the
customer to deploy the solution for real-time analysis.
• Vertical AI masquerading as invisible software: Although example in the market are less obvious today, some
enterprise chatbot startups take this approach where they sell automation bots bottoms up to employees with the
intention of using the data the bots integrate with to gathering insights into how businesses operate. This can be
used to build a system of intelligence for optimizing business functions and operations to be sold more formally to
senior management as a next evolution of the company.
• Single pane of glass: A prevalent approach is to integrate disparate data and provide unified visibility across
databases. In this respect, the thin edge strategy is data middleware, with applications that enable business
process transformation upsold on top of this core functionality.
145. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 145
Megaclouds aren’t the only bullies,
don’t forget the SaaS-holes
146. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 146
TIPS FOR ENTREPRENEURS
SAAS-HOLES ARE TO BLAME FOR ‘LICENSES IN THE CLOUD’ PHENOMENON
•SaaS vendors are making enterprises run back to licenses.
SaaS SOR vendors are becoming mighty and taking advantage
of it — using aggressive tactics to expand dollar share within
existing accounts, often by shoving excessive features and
extensive contract terms down customers’ throats
•Enterprises are push back by opting for licenses to run in
their own virtual private cloud… which may ruin things for the
rest of the industry should the trend continue to persist.
Why this matters:
More license revenue = less deferred
revenue = higher capital
requirements
147. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 147
TIPS FOR ENTREPRENEURS
UNDERCUT SAAS PRICING TO AVOID LICENSES IN THE CLOUD RUT
Most enterprise infrastructure startups
make the mistake of not undercutting
SaaS prices right away, and only do so
after customers start opting for licenses.
0%
20%
40%
60%
80%
100%
Year 1 Year 2 Year 3 Year 4 Year 5 Year 6 Year 7 Year 8 Year 9 Year 10 Year 11
First-time customers by deployment
model (%)
Subscription-based License-based
0
50
100
150
200
250
300
350
400
450
Year 1 Year 2 Year 3 Year 4 Year 5 Year 6 Year 7 Year 8 Year 9 Year 10 Year 11
3-year TCO by year of sales contract ($USD, 000s)
Annual subscription 3-year license
This leads to ‘licenses in the cloud.’
Long protracted path to get out of
‘licenses in the cloud’ rut, discounting
SaaS prices helps.
Theoretical growth metrics for high growth enterprise infrastructure startup
148. 2017 ENTERPRISE ALMANAC // @WORK_BENCH // #2017ALMANAC 148
Join our Work-Bench extended community
Sign up for our Enterprise Weekly newsletter, a weekly digest of all things enterprise with 10K+ subscribers.
WWW.WORK-BENCH.COM @WORK_BENCH HELLO@WORK-BENCH.COM
thank youTHANK YOU
MICHAEL YAMNITSKY
Venture Partner, Work-Bench
@ITSYAMNITSKY MICHAEL@WORK-BENCH.COM
Please reach out to say hello!