5. On Experimentation
In a lot of ways building a company is
like following the scientific method. You
try a bunch of different hypotheses, and
if you set up the experiments well, then
you kind of learn what to do…
We invest in this huge testing
framework. At any given point in time,
there’s not just one version of Facebook
running in the world. There’re probably
tens of thousands of versions running
because engineers here have the power
to try out an idea and ship it to maybe
10,000 people or 100,000 people. And
then they get a readout.
6. As humans, we:
3. When successful, assume we know
what made it successful
2. Inflate the impact of a success
1. Overestimate the probability of success
8. Validated
learning…
… is a process by which we learn by
experimenting with an idea and
measuring it to validate its effect.
9. By using validated learning
we can create the best
business recommendation
based on how it touches:
• the consumer
• the business
10. A point about learning…
Unconscious
Incompetence
Conscious
Incompetence
Conscious
Competence
Unconscious
Competence
Ignorance is
bliss
I’m not
doing it
right!
I can do
it if I try
12. …is about finding what
works best, and quickly.
Experimenting…
…creates the
opportunity to
explore multiple
directions,
choose the best
and refine it.
17. Each experiment
tests a falsifiable
hypothesis.
Focus on speed
for faster
validated learning.
Experiment Design
18. Why we need to experiment
• AS humans, we:
• 1. Overestimate probability of success
2. Inflate the impact of a success
3. When successful, assume we know what made it successful
• Changing environment
19. Typical graph for development
Options for
Success
Cost of
change
Time Launch
Amount
Planning
Costs increase, options decrease
20. The first principle is that you
must not fool yourself
- and you are the easiest
person to fool.
Richard Feynman
23. hypothesis
Pronunciation:
/hīˈpäTHəsəs/
NOUN
A supposition or proposed
explanation made on the
basis of limited evidence as
a starting point for further
investigation
assumption
Pronunciation:
/əˈsəm(p)SH(ə)n/
NOUN
A thing that is accepted
as true or as certain to
happen, without proof.
28. Specify
hypothesis
Design
experiment
& actions
to be taken
with output
Run test
and take
actions
The Experimental Design iterative learning loop
State what we believe
to be true and required
for the offer’s success
Choose a methodology
define the measure and
set a target and actions
Run test, capture the
outcomes and follow
actions to be taken
Prioritise!
If hypothesis is
confirmed, we persevere
and GO.
If falsified, we pivot and
change the proposition.
What will kill the
proposition first?
2
34
1
30. Exp 1 Exp 2 Exp 3 Exp 4
Experiment along the journey
Check if the performance matches the promise
Idea BusinessMany, many iterations
Video
Advert
(A/B)
Web Page
(A/B)
Paper
Prototype
31. 1 Placed Facebook ads 2 Clicked through to Landing Page 3 Sign-up
Sequence experiments to test the future experience
4 Recruited for
beta product test
32. Prototypes are the engine of experimental design
Making stuff
deepens our
thinking and
drives ACTION
If you want to make something
great, start making.
Tom Kelley, IDEO
Having stuff means
we can share with
people who will “get
it” more easily
We can iterate and
make better
development
decisions
1 2 3
‘‘
33. The purpose of a
Minimum Viable
Proposition (MVP) is to
maximize learning per
euro we spend*.
* using customers and not testers
34. Real“Test” products
Consumer
perception
Our perception
MVP 1.0
Prototypes
€/real & LCM
1.0
Propositions
on the market
MVP MVP
Learning tools
MVPs are prototypes that customers believe to be real
Proto A Proto B Proto C Proto DProto AProto C
39. When we involve
real customers, we
need to engage
them through
storytelling
Add pic
40. 1. Create a story
2. Make the video
3. Fix the video
41.
42. Too
complex?
A simple structure is your friend:
Confrontation/tension
Enlightenment/Change
Exploration
Closure }e.g. try a four
act structure
43. Exp 1 Exp 2 Exp 3 Exp 4
Experiment along the journey
Check if the performance matches the promise
Idea BusinessMany, many iterations
Video
Advert
(A/B)
Web Page
(A/B)
44. A/B test
Big directions
In the early stages, it’s not a small variable
vs. small variable (aka optimisation)
We have to answer and find the big
directions before we can optimise.
45. Choose the best channel for your communication
http://www.growthtribe.io/blog/brass-method-your-ultimate-guide-to-prioritising-which-customer-acquisition-channel-to-test-first/
Facebook isn’t always the right channel to reach
your audience.
46. Analytics
How many & when?
- e.g. Google Analytics
- e.g. Advert analytics
What did they do?
- E.g. Hotjar
n.b. The numbers never match
47. Exp 1 Exp 2 Exp 3 Exp 4
Experiment along the journey
Check if the performance matches the promise
Idea BusinessMany, many iterations
Video
Advert
(A/B)
Web Page
(A/B)
Paper
Prototype
48. 1st consult
skills
2nd consult
habit
3rd consult
Last resort
basic instruction one remedy one Talcum powder (messy but it
works)
basic instruction two remedy two
gauge skill level advanced instruction
introduction to new skill error strategy
terminology understanding confront habit
surprising instruction
Concierge MVP
Team identified an advice model
which appears to very effective
50. GoodBad
While the one on the left looks nicer, it is not usable.
Iterate through your sketches, throw lots away, keep the best stuff.
51. How useful?
Try http://popapp.in and see how quickly
you can get a prototype running.
Use this to test navigation & meaning very
early on. Use the results to iterate through
a web app with full analytics. Use this to
understand if a feature is worth building.
53. Draw on expertise around you to:
Structure Effective Hypotheses
Identify the right Risks in the Idea - Elephant in the room
Ensure Stakeholders stay accountable to the Results
Create the ability to Run Quick, Iterative Experiments
Stop the experiment when you’ve collected the full data
Hire statistical knowledge (sample # you must use to find a significant result)
Run at same times - compare apples with apples
Create control variables across experiments
Integrate traditional testing where relevant
Build your individual expertise
Use clear assessment criteria (Desirability, Feasibility, Viability)
56. After each experiment, there are always 4
possible outcomes
1. GO You are confident that your hypothesis is
valid and have the evidence to show it
2. Confirm
3. Pivot
4. STOP
You are confident that your hypothesis is
correct but you need more evidence to
support it
You no longer believe that this offer is a
right for you to pursue at this time.
You believe in the vision but you need to
re-visit the means & path to get to the
vision