Adrian DunkleyBlog

Opinion · Scientist, Entrepreneur, Businessman

Understand your customer before you optimise the campaign

A narrow amber path branches through several layers of dark blue paper before reaching a bright opening.
Illustrative conceptual diagram by AI Boss Limited.

I developed Understand Your Customer, or UYC, to connect what an organisation learns about its customers with what it actually changes. A campaign can be tuned to get more people to open a message while the reason they leave remains buried in a service queue. I wanted the decision about what to change to start with the customer’s difficulty.

A missed renewal can start with a failed payment

Take a hypothetical subscription business. The marketing team sees customers whose subscriptions expired after they opened the renewal email. Support can see that some asked how to replace expired payment cards. Opening the email tells the teams little about whether a subscriber intended to renew or could complete the payment.

Suppose there are customer reports of an expired card and a failed payment on the same account. Those records give the team a service problem it can test. A product manager could randomly assign eligible subscribers to two groups before the renewal reminder goes out. Both groups would get the same reminder. One would receive a direct way to update the card. The other would use the usual payment route. In each group, count completed renewals within the same window and divide by everyone assigned to that group. If more people reach the payment page and the renewal rates remain similar, the extra visits alone give the manager little reason to expand the change. Check how uncertain the difference is before deciding whether to roll it out.

What the six phases are for

Customer Discovery asks what people are trying to do and where they get stuck. Data Foundation checks whether the records can answer the question. If the team cannot connect a call for help with the payment attempt, the available data tell it less than the records appear to contain.

Insight Generation brings together what customers say and what the records show. Solution Design turns the most credible explanation into a proposed change. Implementation and Testing give the team a chance to check the change on a limited scale. Measurement and Optimisation uses the results to decide what to keep, what to change and which earlier assumption to revisit.

The Customer Equity Trees presentation places UYC alongside a separate research proposal about how the choices available to customers can change their financial paths. That work raises a question I think belongs in customer analytics from the start. What might improve for the person whose behaviour we are studying? In the renewal example, the payment change is worth testing because it could save a customer who wants to renew a call to support.

How a finding reaches another team

Diffusion begins with particles in random motion. Imagine a closed chamber with many labelled particles on one side. Particles can cross the line between the halves in either direction. Early on, more particles tend to cross from the crowded half because more are available to cross from that side. As the difference in concentration shrinks, the net transfer between the halves falls on average. Particles still cross in both directions when the average concentrations are equal.

Diffusion in a closed chamber

Left half 40

Right half 0

A dashed line divides the illustration into two equal counting areas. Forty markers begin on the left.
The dashed line is a counting aid. The model keeps 40 markers inside the chamber and assigns no physical time to a step. Compare the left and right counts at the start and in the later snapshot. At the start all 40 are on the left. Later, the counts are similar. Illustrative conceptual diagram by AI Boss Limited.

What gets lost in a handover

I use diffusion here as a qualitative analogy. An observation can begin in one team and become useful to several. Molecules need no meeting or permission to spread. Customer evidence needs somebody to share it and say what the records cover. A help desk can keep the source while a product manager works from an authorised summary. The number of people who hear about an issue says little about whether they heard it accurately. A physical diffusion law supplies no forecast for a person or a team.

An expired card is only one possible reason a payment fails. If callers are the team’s only source, customers who tried again without calling may tell a different story. The analyst can ask whether the same pattern appears in payment records before proposing any change. The two sources answer different questions. The account tells you that a payment failed. The customer can tell you where they stopped and why.

Choose one decision to revisit

In my People First, Problem Second, AI Last approach, I choose the tool after I can describe what the customer needs. At the next campaign review, pick one group of customers you had planned to contact. Ask the team that handles enquiries to bring a de-identified account of a recent problem in that group. Write down the service decision they think it should change. An AI tool can group feedback you have permission to use. An analyst should check the original reports, look for accounts that disagree and record who has not been heard. Before changing the offer, agree who will ask a customer whether the proposed service change addresses the problem.

Sources and scope

My UYC source summary identifies the procedure manual behind the six phases and explains how my Customer Equity Trees presentation relates to the framework. The renewal example is hypothetical and contains no customer records.

For random motion and net transfer, see OpenStax College Physics, section 12.7, Figures 12.20 and 12.21.

See what mentoring more than 30 founders taught me for more on testing an idea. Kill My Startup helps founders challenge the assumption behind a proposal. StarApple AI documents my broader applied AI work.