A well-known scenario: when a product or service is renewed, the assumption is made that an improvement has been achieved. Think, for example, of an adjustment to your website, the implementation of a new product feature or the digital transformation of an organisation to work more efficiently. But how do you know whether the situation is really better after the innovation? Start by making the right hypotheses.
What are hypotheses?
Hypotheses are statements about reality that can be tested on the basis of data. The purpose of hypotheses is to elaborate in a structured and explicit way:
- what you are going to change
- what you expect will result from this change
- how you expect the change to affect the value a product or service delivers.
Making a good hypothesis in 4 steps
Step 1: Make sure the hypothesis is relevant
A hypothesis must contribute to the goals, and thus the KPIs, of your organisation. Whether testing your hypothesis leads to a positive or negative result, the subsequent insights will more than likely help your organisation further. Moreover, projects with relevant hypotheses are more likely to receive support from within the organisation than irrelevant hypotheses.
Example: a webshop with a KPI 'conversion rate'
Relevant hypothesis, contributes to a KPI: By conducting a price promotion on all mobile phones, the conversion of mobile phones will increase.
Irrelevant hypothesis, irrelevant metric: By conducting a price promotion on all mobile phones, there will be more page views on the mobile phone product page.
Step 2: Determine your independent and dependent variables
In the hypothesis, specifically name what is being deliberately changed (independent variable), and on which KPI this change is expected to have an impact (dependent variable). Also describe the direction of the effect: do you expect an increase or decrease in the KPI? If you expect to influence several KPIs, draw up a separate hypothesis for each KPI.
Example: delivery service tests track-and-trace codes
Current situation: After ordering a product, customers will not receive a track-and-trace code.
New situation: After ordering a product, customers will receive a track-and-trace code.
Dependent variable: The number of customer service calls about the status of an order Independent variable: Receiving or not receiving a track-and-trace code.
Hypothesis: Sending a track-and-trace code with the order confirmation to customers who have placed an order will reduce the number of calls on this issue compared to customers without such a code.
Step 3: Determine your target audience
Consider who the target audience is to whom your hypothesis applies. This determines to which population the finding is translatable from your hypothesis.
Sample hypothesis with target audience: Offering an online order discount will lead to more conversions from online customers than not being offered this discount.
Example without target group: Offering an online order discount will lead to more conversions than not being offered this discount.
The target group of example two is unclear. Who does the discount apply to? Which group will have a higher conversion? When can this hypothesis be accepted or rejected?
Step 4: Compare at least two measurements
When testing a hypothesis, you compare at least two different measurements on which to base your conclusions. This allows you, for example, to compare a measurement before optimisation with a measurement after optimisation, or, as in A/B testing, both variants at the same time.
You can also create a group where you apply optimisation and compare it with a group where you do not apply optimisation, such as with an A/B test.
Example hypothesis: Working with digitised tools leads to fewer errors among employees than with analogue work tools.
The employer randomly gives digital work tools to 50% of its employees, and the other 50% keeps the current analogue tools. She compares the error rate of the digital group with the non-digital group.
A good hypothesis is half the battle
With these 4 steps you will come a long way. In addition to these steps, there are many things to keep in mind when working with hypotheses. Think of designing the measurements, analysing the results and setting up an optimisation process.
Besides your outcomes being able to show that the desired change has taken place, it is good to look a little further. Perhaps other measurements have been influenced that are very important for your organisation. In any case, one thing is certain, a good hypothesis is half the battle.
A handy hypothesis format
Do you also want to change your assumptions in A/B testing, product improvement or optimisation projects into demonstrable success? Follow the steps above and download our handy hypothesis format!
Need help formulating good hypotheses? Our Customer Experience Consultants are happy to think along with you.
This is an article by Maks Keppel, Web Analyst at Elsevier
Former colleague Maks helped organisations improve their products and services through data-driven optimisation during his time at Digital Power. He is now a Web Analyst working for Elsevier Life Science to optimise digital products.
Receive data insights, use cases and behind-the-scenes peeks once a month?
Sign up for our email list and stay 'up to data':