Reconstruct the customer journey
6 min read
Lesson objectives
- Reconcile interview, analytics and sales records; case: high traffic and stalled first purchase.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
A customer journey models how a person moves from need to decision and use; it is not a set of arrows copied from an analytics tool. Diagnose where a customer loses the ability or willingness to proceed, and what evidence describes that experience. A fictional short-course service has twenty thousand program-page visits and two hundred purchases. The overall conversion rate alone cannot tell whether people struggle to discover the right course, understand the price, create an account, or complete payment. Divide the journey into observable customer tasks: find, compare, build confidence, register, pay, and begin learning. For each transition, define a precise event and a second, reasonably independent record that could challenge it.
Do not confuse a stage with a channel. Someone may arrive from search at a comparison page, return through email, and pay after a telephone conversation. Crediting the whole journey to the last click hides the comparison page; assigning the full revenue to each touchpoint counts the same sale repeatedly. Specify the permitted session or user identifier, cross-device linkage limits, and observation window, then identify interactions you cannot see. An absent recorded path is not proof that the customer had no experience outside measured channels. Use the map to choose an improvement decision, not to claim a complete history for every individual.
Triangulate three kinds of evidence. Event data show what the system recorded under its tracking definitions; operating records show what was created, paid, or cancelled; interviews and usability sessions show how a small group understood the interface and why they struggled. If five participants say the certificate fee surprised them, do not convert five people into a population percentage. If the financial ledger records completed payments absent from the purchase event, do not infer a sales fall from the analytics dashboard. Build a reconciliation table with transition name, definition, numerator, denominator, period, source, possible missingness, and qualitative evidence that may explain behavior without replacing the count.
There is conflicting testimony. Sales staff say calls explain the price difference well, while learner interviews suggest that certificate conditions appear only after registration. Separate callers from self-service visitors. Examine the share of mobile visitors who see the price, their traffic sources, and whether conditions are clear before account creation. Calls may work for those who request them but represent a small minority. Options include exposing price and components earlier, simplifying registration, and repairing payment. Compare the alternatives by verified size of loss, team capacity, trust effect, and testability. A button color is not the answer if the offer itself is obscure.
A journey map must show people missing from measurement. Someone declining tracking or switching from phone to computer may have an incomplete recorded path; some platform reports model events that are not directly observed. Compare trends only after checking whether event definitions stayed stable, and inspect the change log around a launch. Use paid orders in the commercial system as a relatively independent anchor. When platform totals differ, record plausible reasons instead of forcing equality. You need not match every click to identify a step worth investigating, but you must understand the denominator before quoting a rate.
For a journey spanning a website and a call, define a privacy-conscious join rule: an order or referral code recorded when the customer agrees to its use, rather than exporting a name list automatically to every platform. Report the share of transactions that can be linked and the characteristics of those that cannot. If high-value calls are rarely linked, a mathematically correct “return by channel” table may still be biased. Test the possible direction through sensitivity analysis: assign half of unlinked orders to search in one scenario and all to direct visits in another. Does the preferred channel change? If so, defer a large budget transfer until linkage improves or an independent experiment is run. If it does not change, say so and state the assumptions.
The consulting deliverable is a ranked opportunity list, not an attractive journey poster. For each opportunity, name a specific point of friction, supporting and conflicting evidence, a short test, and a guardrail metric. For example: “Show price and certificate conditions before registration to a randomized sample; monitor paid completion, net refunds, and complaints about misleading information. Do not call a rise in registration-button clicks success by itself.” More clicks accompanied by fewer paid orders may simply attract poorly matched visitors. An improvement limited to mobile does not automatically transfer to desktop. The reconstruction ends with a falsifiable hypothesis, not a fixed portrait of an imagined customer.
Lesson summary
A customer journey models how a person moves from need to decision and use; it is not a set of arrows copied from an analytics tool. Diagnose where a