Improving customer experience: from diagnosis to a verified action plan

Start from a lived difficulty to improve customer experience

When an executive committee asks to "improve customer experience", the request is legitimate but unusable as it stands. It says neither which customer, nor which moment, nor what should change.

This article is written for CX Directors, Heads of CX and VPs of CX, Voice of the Customer or Customer Insights, at large and mid-market B2C companies, who must turn that ambition into a measurable action plan. It does not revisit the fundamentals of customer experience: it offers an operational method, from diagnosis to follow-up after rollout.

The starting point is always a lived difficulty: a customer who has to call several times to change an address, a sign-up abandoned at the last step, a refund whose date nobody can announce. These situations describe a customer need, a precise moment in the journey and a gap between what was expected and what happened.

Write this difficulty as one sentence, from the customer's point of view: "I want to do X, and I can't because of Y." Also set the scope: which customer segment, which channel, which period. This sentence becomes your starting objective, the one you will check at the end.

Build a diagnosis that combines perception and behavior

A useful diagnosis answers two questions: what do customers experience, and what do they actually do? Behavioral data (drop-offs, repeat contacts, handling times) shows where the journey deteriorates, without saying why. Customer feedback explains what got in the way, without saying how many people are affected. Combining the two is what lets you understand a pain point. The Voice of the Customer is precisely what links what customers say to the moments when they say it.

In its publication on customer relations and indicators, the AFRC (the French customer relations association) points out that assessing performance properly depends on choosing the right tools and the right data, then studying correlations between indicators to inform decisions. There is no universal threshold: each diagnosis is built from your own baselines.

Understand the words customers use

Verbatims are the raw material of perception. Read them before you classify them. Customers rarely describe a "usability issue"; they write "I can't find where to confirm", "they keep asking me for the same papers" or "nobody calls me back".

These phrasings point to the exact moment of the difficulty, what the customer tried and how they feel. So group verbatims by situation rather than by keyword, noting the words used and the effort described. This fine-grained understanding keeps you from treating one symptom in place of another.

Check what the data actually allows you to conclude

A high volume of mentions does not mean a high impact. A survey sample can also overrepresent unhappy customers, or on the contrary those who completed their task.

Distinguish three levels: observation (what is noted), interpretation (what you infer from it) and evidence (what confirms the interpretation through an independent source). A pain point confirmed by verbatims, journey data and feedback from advisors is more solid than one flagged by a single source. Note the limits of each source: you will avoid overinterpreting a result later on.

Trace the pain point back to a cause you can act on

A pain point describes what the customer goes through. A cause describes what produces it inside your organization. As long as you stay at the level of the pain point, actions only fix the surface: an apology message, a response script, one more question in the online help.

Getting to the cause means looking behind the screen or the phone: a business rule, a tool that does not pass on a piece of information, a manual validation, a dependency between two teams. Projects for digitalizing customer relations often fail for this reason: they automate a journey without dealing with the cause of its difficulties.

Reconstruct the situation before changing the journey

Pick a few real cases that represent the pain point. For each one, reconstruct the full sequence: what the customer did, what they saw, who stepped in, what information was circulating and at what point everything got stuck.

Do it with the teams involved: advisors, back office, product, sometimes external partners. Each sees part of the situation, and bringing them together often reveals a cause nobody could see alone. Do not change the journey during this phase: a premature change may shift the difficulty elsewhere and blur the diagnosis.

Formulate a falsifiable hypothesis

Then formulate a hypothesis about the cause that facts could contradict. "The journey is too complicated" is not one. "Customers drop off because the proof document requested at the third step is not announced at the start" is.

A good hypothesis states the assumed cause, the expected effect on the customer, how to check it and what would refute it. Keep several hypotheses open as long as the diagnosis cannot settle the question: improvement then becomes a learning process, rather than a series of bets.

Prioritize actions with a transparent decision grid

Not every cause can be addressed at the same time. A transparent decision grid lets you choose and, above all, explain why one topic comes before another. Without it, priority often goes to the department that pushes hardest.

The grid combines criteria everyone knows: measured impact (NPS points, business effect), strength of the diagnosis, implementation effort, available resources and dependencies on other projects. Publish it with the scores given and their justifications.

Compare problems at the same level

A grid only compares meaningfully things that are comparable. First group topics by level: recurring contact reasons, journey steps, business rules.

For recurring contact reasons, the deciding criterion is their measured impact in NPS points, not how often they are mentioned. A reason cited less often can weigh far more heavily on recommendation than a very frequent but minor one.

Use a table that prepares the action

The prioritization table links each problem to the evidence that confirms it, the planned action and the way its effect will be checked.

Each row links a verification indicator to a specific action. That link is what will tell you, after rollout, whether the action had the expected effect.

Build an action sheet that engages the right teams

A prioritized action needs a short, shared sheet: the problem, the cause retained, the planned change, the teams involved, the success measure and the date of the first review. It fits on one page.

The sheet engages the right teams, not just the customer relations service. The causes of a pain point often lie in product, billing, logistics or IT. Without those teams, customer service ends up handling the consequences of decisions made elsewhere.

A VoC platform gives each sheet a common base: the verbatims involved and the impact of each pain point, measured in NPS points rather than in number of mentions.

Describe the change expected for the customer

Describe the result from the customer's point of view before describing the solution. "The customer knows from the first page which documents to prepare" is an expected change. "Add a banner on the home page" is only one solution among others.

Assign ownership of the result

Each sheet has a single owner, accountable for the result and not just for delivery. They coordinate contributors, track dependencies and raise the alarm when the project drifts.

In a B2C organization of this size, one action often involves several teams. Without a named owner, each team considers it has done its part once its task is finished, even if the customer sees no difference.

Test an improvement without overinterpreting the results

Before rolling an improvement out everywhere, test it on a limited scope: one channel, one region, one customer segment. The test checks the hypothesis and reveals side effects before they reach every customer.

A before/after comparison is intuitive, but misleading. Between the two periods, seasonality, a marketing campaign, a technical incident or the customer mix may change. An improvement in the indicator therefore does not prove that the action caused it.

Whenever possible, compare two groups exposed at the same time, one with the change and one without. This is the principle of A/B testing as described by Nielsen Norman Group: only the tested variation differs, which isolates its effect. On human-staffed channels, where a controlled test is harder, a comparable control group is still better than a before/after reading.

Set the duration, the main indicator and the result that would lead you to drop the action before the test starts. Also read the verbatims from the test group: an indicator can improve while customers describe a new difficulty.

Set up a follow-up loop after rollout

A deployed improvement is not a closed topic. Without follow-up, many actions lose their effect: a rule gets bypassed, a tool changes, a new team was never trained.

The follow-up loop links the indicators on the action sheet to customer feedback. Customer insights analysis lets you track how each theme evolves over time and spot a pain point that comes back.

Check that the improvement holds over time

Plan checkpoints on fixed dates. At each one, check the main indicator, the verbatims on the theme and feedback from the field. Watch the quality of the response, not just the resolution rate: a case can be closed without the customer getting what they expected.

If the effect wears off, find out why before adding an action. A targeted correction is often enough: train a team, adjust a rule, update some content. Record what you learn on the sheet. Fnac Darty has publicly reported, for its marketplace, a doubled NPS and complaints divided by four since its approach went into production (Fnac Darty case study).

Expand only when the need justifies it

When an improvement works within one scope, it is tempting to extend it everywhere. First check that the need is the same elsewhere: a journey suited to new customers does not necessarily meet the expectations of loyal ones.

Some improvements are better adapted to specific segments. The topic then belongs to personalized customer experience, with its own rules.

Example of a plan to clarify a sign-up

Here is a hypothetical example, built to illustrate the method and not taken from a real case: a B2C retailer offering online sign-up to a subscription delivery service.

Lived difficulty. Customers start signing up without finishing, and some call customer service to ask for help.

Diagnosis. Verbatims show that customers do not understand why proof of address is requested, or which documents are accepted. Journey data confirms that drop-offs are concentrated on this step.

Hypothesis. If the list of accepted documents and the reason for the request are announced on the first page, drop-offs and calls about this reason will decrease. It would be refuted if informed customers drop off as often as the others.

Prioritization. This reason is not the most cited, but it has the strongest impact on new customers' NPS.

Action sheet. The sign-up journey manager owns the result. Contributors are the product team, legal for the list of documents and customer service for its answers.

Test and follow-up. The new journey is shown to part of the visitors, while the rest keep the old one over the same period. No numerical target is set before the baseline measurement. The team then tracks drop-offs, calls and verbatims from new subscribers.

See Glanceable on your data.
A 30-minute conversation, on your own data.

Request a demo →