This is a presentation and workshop I shared during the launch of Intel Capital's EdTech Accelerator.
It covers two core topics:
1) Developing and prioritizing strong hypothesis statements
2) Testing hypotheses via qualitative and quantitative tools
My thinking draws draws on Steve Blank's customer development process and Javelin's Experiment Board and enhances it with the startup marketing process I've written about on my blog (http://www.cezary.co).
The workshop is relevant to both technology companies as well as any startups looking to validate their business.
3. AGENDA
WORKSHOP (1:30PM - 4:30PM)
Developing hypotheses (25min)
Exercise: Create your own (60min)
Break (10min)
Testing hypotheses (15min)
Exercise: Deploy a test (70min)
4. HYPOTHESIS STATEMENTS ARE DECLARATIVE
Definition
States a fact or an argument
Ends with a full stop
Example
I believe that…(X is true)
5. THEY BEGIN WITH THE CUSTOMER-PROBLEM
Customer-Problem
Customer X has Problem X
(that is worth solving)
I believe that…
High school students struggle with
basic data literacy
*Always phrase the problem from the customer’s perspective
*Add emotional language whenever possible
6. AND THEN ADDRESS THE PROBLEM-SOLUTION
Problem-Solution
Problem X can be solved
by Solution X
I believe that…
Bringing real-life data into the classroom
will make math and science more engaging
*Only relevant after you’ve validated that a problem exists
*Solution should be unique to alternatives
7. STRONG HYPOTHESES ARE DERIVED FROM
DATA & INSIGHTS
Customer-Problem
High school students struggle with
basic data literacy
Rationale
25M high school students are
falling behind in quantitative subjects
Problem-Solution
Bringing real-life data into the classroom
will make math and science more engaging
Rationale
The explosion of data today makes it
more accessible than ever before
8. AND ARE ALWAYS BASED ON ASSUMPTIONS
Customer-Problem
High school students struggle with
basic data literacy
Assumption
Parents and school districts consider
math and science a high priority
Problem-Solution
Bringing real-life data into the classroom
will make math and science more engaging
Assumption
Most students are genuinely interested
in math and science
*Focus on the riskiest assumption
10. BEFORE MOVING TO MORE GRANULAR TESTING
Customer
Overarching persona
Specific segment
Stage in lifecycle
Acquisition source
Geography/location
Users vs. influencers vs.
decision-makers
Problem / Challenge
Master problem
Sub-problem
Distribution/marketing channels
Buying process/sales funnel
Willingness to pay/pricing
Messaging
Loyalty/retention
Evangelism/virality
Solution
Existing product/service
New product/service
Specific feature/flow
Marketing tactic
*Keep one variable constant
11. YOU CAN SOURCE HYPOTHESES FROM MANY PLACES
Behavioral insights
Background research
Industry benchmarks
Company stage
Growth goals
Investor requirements
Previous hypotheses
12. Highest risk
Biggest potential impact
Start high-level, then get specific
Questions to ask
Is the hypothesis core to your value proposition?
Will validating the hypothesis significantly de-risk your business?
For each hypothesis, have you validated the higher-level one above it?
BUT MUST PRIORITIZE THEM TO CREATE VALUE
13. LAUNCH EXPERIMENTS IF THEY MEET 3 CRITERIA
Speed
“one-afternoon” rule
simple over elaborate
Cost
cheap tools
sweat equity
shared resources
Accuracy
right audience
right approach
14. SET GOALS AND SUCCESS METRICS
Goals should be quantitative when possible
Qualitative results can be measured
Examples
A minimum of X respondents will agree
The majority of respondents will exhibit positive language
User will convert at a 5% significance level
Overall retention will rise by X%
1 in 5 people will recommend a friend
15. FOLLOW THESE BEST PRACTICES
Baseline Better
Keep it simple
Customers understand the detailed
technology behind my product
Customers understand why my product
is faster than competition
Break big hypotheses
into smaller batches
I can acquire users via paid spend
I can acquire users via social ads
I can acquire users via paid content
I can acquire users via trade shows
Add details where
possible
I can acquire new users from Facebook
efficiently
I can acquire new users from Facebook
mobile ads at $1/click
Test one thing at a time
(avoid: and/either/or)
Customers prefer product X over
product Y or Z
Customers prefer product X over Y
Customers prefer product X over Z
16. BREAKOUTS
(60 minutes)
Define core customer-problem & problem-solution
Compile & prioritize list of key hypotheses
Share work with group
bitly.com/hypothtesting
18. START WITH QUALITATIVE TESTING TO GATHER
FEEDBACK
In-person conversations / live demos
Focus groups
Video conversations
User testing / video recording
Public forums / blogs
Surveys ($ or no $)
19. FOCUS THE CONVERSATION FOR BETTER RESULTS
Don’ts Do’s
Don’t rely on friends and family Do speak with outsiders who lack a filter
Don’t tell respondents you’re working on an
idea
Do tell respondents that you’re seeking an
opinion
Don’t ask leading questions
Isn’t it terrible when…?
Do let them describe their environment and
motivations
Don’t provide options unless necessary
Is your preference A, B or C?
Do let people decide on the answer
Don’t ask hypotheticals
What would you do if…?
Do ask about past behavior
How have you solved this problem in the past?
Don’t ask about intent
Would you tells others about this product?
Do test their willingness to take action
Can you share this product with a friend?
20. USE QUANTITATIVE TESTING TO CAPTURE KEY DATA
AND IDENTIFY ACTION
Product analytics
Heat maps
Social media ads
Paid search
Landing pages
Placeholders / waitlists
A/B testing
21. ENSURE THAT YOUR DATA IS RELIABLE
Questions to ask
Who is your audience?
How did they come to your app?
Are the results real or sampled?
Is the sample representative of your audience?
How is significance determined?
22. EVALUATE RESULTS
Significance of sample
Minimum 10 responses for qual, 30 for quant
Tangible outcome
Signed letter of intent, verbal agreement
Strong body language
Overall enthusiasm, leaning in
Correct metric hierarchy
Later funnel metrics trumps early funnel
If data is ambivalent, revisit hypothesis