Consulting brief and journey diagnosis · Lesson 2

Journey mapping and hypothesis testing

9 min read

Lesson objectives

  • Frame the decision, scope the engagement, and separate evidence from hypotheses across channels.
Reading tools

A journey map is not a sequence of screens owned by the organization. It reconstructs what a customer does from the emergence of a need until the outcome is obtained or abandoned. A person may start with an advertisement, move into an app, wait for a message, call, and then visit a branch. If the organization maps only the app, it calls the move to phone an ‘exit’ even when it is a necessary step toward completion. Build the map from specific events in interviews, logs and observation. Beside each event, record the customer's expectation, what actually happened and the owner of that step. GOV.UK's experience-map guidance considers what users do, think and feel over time and how service parts depend on one another; use it as a research principle, not a mandatory template for every organization. Do not invent an emotion from a click log. An unspoken or unobserved feeling is an interpretation to test.

Begin with cases that have a clear start and end. Reconstruct a timeline for each case before merging them into a general map. An event card records approximate time, channel, customer action, service response, step outcome, evidence source, and confidence. Do not merge distinct journeys merely because their stages have similar names. A customer tracking a late parcel and another disputing a delivery charge may both begin on a shipping page, but they require different decisions. Make separate maps when paths or barriers differ materially. A polished visual is not needed first; an auditable event table is more useful than a diagram that hides exceptions. After comparing cases, mark team handoffs, repetition, silent waiting, and any step after which the customer cannot tell what happens next. Treat these as hypotheses, not settled causes.

Distinguish outcome evidence from mechanism evidence. Contact logs may establish that 28% of return-request owners contacted again within a week, but they do not explain why. Interviews with five users may reveal confusing status messages, but five accounts do not establish prevalence across the population. Agents may report insufficient settlement authority; this is evidence of an operating constraint to compare with case-transfer logs. For each hypothesis, write a prediction that could be falsified. If the message drives repeat contact, repeats should cluster after failed notifications and users without messages should describe an information gap. If authority is the main cause, repeats should rise after transfer to a decision-owning team even when the message arrives. Deliberately seek cases that contradict the preferred explanation. Ignoring them turns the map into a compelling story rather than a diagnosis.

Case selection is part of inference quality. If you study completed cases only, people who stopped before reaching an agent disappear. If you study complaints only, rare but vocal problems may dominate. Deliberately include successful and failed cases, different channels, and users needing language support or an alternative access route where relevant. Document why each group was chosen and what it will not represent. Do not request more personal data than needed to connect events; a case pseudonym and aggregated event analysis may suffice. In interviews, separate ‘What happened?’ from ‘Why do you think it happened?’ The latter is the customer's interpretation and may or may not describe the internal system accurately. Respect that interpretation when designing communication, but test the system mechanism with independent evidence. This keeps silent digital traces from overwhelming experience while avoiding promotion of one person's account to a statistic.

A journey map needs a rule for unknowns as much as one for events. On each case sheet, record the time source: system timestamp, customer recall or agent estimate. A customer may say ‘I waited three days’ while the system shows 48 hours; do not accuse either side before learning whether workdays or a silent interval between steps is meant. Record the gap as a research question and inspect actual message and event times. When merging cases, do not copy one starting stage to all: one customer entered from a message link, another at a branch and a third called before having an account. Show branching paths and the shared decision point, including people who never complete. A ‘silent period’ may lack a system event yet matter deeply to the customer; substantiate it through interview and approximate dates without inventing minute precision. What customers do, think and feel can guide understandable communication, but an emotion field should not be filled from the researcher's guess. Use a consented account or mark ‘not asked’ rather than infer feeling from a red box in the map. Source labels on every line make the map correctable when new logs appear rather than a fixed picture defending the team's first hypothesis.

A polished journey map can hide important differences. Suppose the mean time from refund request to receipt is five days. It does not show whether everyone waits five days or most wait two while a minority waits a month. Plot the distribution and exception reasons, then trace long-tail cases step by step. They may cluster in one product, payment channel or branch, calling for a different intervention than a message to everyone. If counts are tiny, describe cases without claiming a stable rate. For each stage, record what the customer knows, what the organization knows, and what either can do. A gap between knowledge and action explains much contact: the customer may see ‘processing’ without a date or review path, while an agent knows the delay cause but cannot change the promise. Separate information repair from authority repair. Test a third explanation when needed: perhaps the underlying rule is confusing or unfair and needs policy-owner review, not merely a better interface. When the sponsor asks for one map, explain when multiple group maps or layers are necessary while retaining a concise shared decision model. A ‘typical persona’ is not an observed person; it summarizes assumptions that need connection to actual events and interviews. Finally label each finding as log-established, participant-reported or consultant-interpreted, and name evidence that could refute it. The map then remains a learning tool that changes with stronger evidence rather than freezing the first appealing story.

Finish diagnosis with an evidence ledger rather than an impressions board. For each finding record its exact claim, supporting event or number, source, sample, alternative explanation, confidence, and the next observation that could change the decision. ‘Some replacement cases switching to phone coincide with no confirmed notification in our log’ is more accurate than ‘bad notifications cause every call.’ Association in a sample is not general causation. Keep access constraints and participant consent in the internal work record, and do not expose identifiable quotes in the leadership recommendation. Rank hypotheses by decision consequence rather than ease of proof: which mistake would cost most if ignored, and which small check separates two executable options? At this point, the map moves from experience narration to directing research and option appraisal.

Lesson summary

Frame the decision, scope the engagement, and separate evidence from hypotheses across channels.