Alternative Hypothesis

A hypothesis is an assumption about our product or service that we can validate. Therefore, we frame the hypothesis in a way that the outcome becomes transparent and can be measured.

We speak of an alternative hypothesis if the hypothesis is phrased in an affirmative way. At the end of this article we will also look into null hypotheses (H0).

Using data to validate our hypotheses is data-driven product development.

Testing our hypothesis is important, because

• we validate with data if our assumptions were correct or wrong,
• we tame our own bias
• and look out for potential risk factors

How to write an hypothesis?

Blueprint:

if we do (ACTION),
we believe (FEATURE or SUBJECT)
will INCREASE, DECREASE, BE FASTER, BE SLOWER, IMPROVE … something,
because of (REASON / PROBLEM WE SOLVE)

As you can see, the expected outcome is specified with a modifier such as

• more / less,
• increase / decrease,
• slower / faster
• improve / optimize (…)

This will help to measure the outcome before and after the change.

if we OFFER TO FILTER USERS
we believe THAT OUR MID OFFICE EMPLOYEES
will BE FASTER DOING THEIR WORK
because THEY CAN FOCUS ON THEIR OWN TASKS ONLY.

if we SIMPLIFY OUR REGISTRATION FLOW
we beliebe that POTENTIAL PERSONAS
will SIGN UP FOR AN ACCOUNT MORE OFTEN
because THEY DON'T HAVE TO BOTHER ABOUT issue.

Variations

Below some alternative wordings to express a data driven hypothesis:

we believe (USER)
has (PROBLEM)
because of (REASON).

if we (ACTION),
this will (MEASURABLE OUTCOME).

we believe (USER) 
has (PROBLEM). 

If we (ACTION), 
she will (BENEFIT), 
resulting in (MEASURABLE OUTCOME).

we believe (MEASURABLE OUTCOME)
will be achieved 
if (USER) 
attain (BENEFIT) 
with (SOLUTION).

we believe (SOLUTION)
will create (MEASURE OUTCOME)
for (USER)
because of (REASON)

Problem statement

If the problem itself is not understood yet, you phrase a problem statement that focuses on what we need to learn (research).

• Keep it broad!
• Phrase it positively!
• Doesn't include the solution!

GIVEN THAT (context, situation)
HOW MIGHT WE (question)
SO THAT (goal, objective)
BECAUSE (need)

Quantifier

When you are searching for a solution and you can not predict the usage, you use generic validations such as increase / decrease.

In case your product is already on the market and you can already make more granular predictions, then you can use percentages to track success in more details:

we believe SENDING AN EMAIL AT 7 PM
will create AN INCREASE IN SALES OF 2%
for PERSONA
because THEY WILL BE REMINDED WHEN THEY HAVE TIME TO FINISH THEIR SHOPPING.

Null hypothesis (H0)

A null hypothesis expresses that there is no correlation between the observed data. It is used as test scenario to disprove and rule out assumptions.

In general it is a statement that the thing you are testing will not lead to an effect or a difference.

Think about a spurious correlation: “When people eat ice cream on a sunny day more people are born.“ I made this one up. If you want to check real data examples, you can visit for example Tyler Vigens website: https://tylervigen.com/spurious-correlations.

To illustrate the concept:

There is no relationship
between variable A and B.

A more realistic example (as it can also turn out that there is a relationship):

There is no relationship
between the new sales strategy
and the revenue.

There is no difference in task completion
between interface A and interface B.

Sending an email reminder 
does not affect 
the rate of user re-engagement.

In this article we were focusing on how to write an hypothesis. A hypothesis is translated to concrete metrics: