Expert portal rebuild: preparing a strategic decision through proof

A fully used tool, a program changing scale

The partner agencies of the Expert program help PrestaShop merchants build, migrate and maintain their stores. To run this network, the company relied on a portal built on an off-the-shelf PRM tool: agency registration, certification management, lead distribution, a public directory for merchants.

Two years after deployment, the tool was used to its full potential by every team: workflows, synchronizations, directory, deal management, scoring. At the same time, the program was changing scale. A new certification model, a rework of the customer data logic, the ambition to turn the ecosystem into a measurable business lever rather than an opaque cost center. The portal was supposed to carry that transformation. It was holding it back.

Three blockers beneath the repair request

On the surface, an improvement request: broken filters on the directory, dated pages, frustrated agencies. Underneath, three blockers of a different nature.

The structural ceiling first. Every team was using the tool correctly, and the tool could not do better. A locked front end where every change went through the vendor’s support, one-way CRM synchronization, multilingual content re-entered language by language. A usage gap can be fixed with training. A structural ceiling cannot.

The absence of measurement next. No reliable data on searches, clicks, contact requests. The directory was a black hole: unable to prove its value, unable to prove its failure, so impossible to arbitrate on anything but intuition.

Sunk cost bias last. Years of investment, processes built around the tool, an ongoing contract. Everything pushed the organization toward yet another repair effort, more comfortable to defend than a reassessment.

The situation called for something other than a redesign project: creating the conditions for an investment decision, with no data to ground it and against the organization’s natural slope.

Preparing the decision without being the decision maker

Head of Product Design of the effort, within a collective bringing together product, CRM, marketing, channel, tech and data, carried day to day by a product-design duo.

The mandate: lead the process meant to inform the choice between optimizing the existing tool, buying another market tool, or building. Concretely, frame and write the design brief, structure the two successive discoveries (the public directory, then the portal as a whole), guarantee that every user’s voice weighed in the case, agencies and merchants as well as internal teams, and produce the artifacts on which the decision would be made.

The limits were clear: neither the budget nor the final decision, which belonged to the sponsors. The responsibility was the quality of the case, with a precise expectation: a defensible decision file, within a bounded timeframe.

Establish the facts, requalify, build conviction

Establish the facts before proposing. Facing a contested but entrenched tool, the first battle is factual: a dozen agency interviews, six workshops with every team operating the portal, five key journeys replayed step by step, an external audit of the experience and the architecture, a benchmark of the alternatives. This work established the fact that changed everything: the problems did not come from a lack of usage but from the limits of the tool itself. Naming that difference in front of the sponsors was the first deliverable of the effort, before a single screen.

Requalify the request. The directory discovery surfaced two blind spots. No usage measurement anywhere: without data, every improvement discussion was played on intuition. And an SEO cannibalization no one had qualified: the most natural organic search for finding a partner led to another internal directory, the modules one. Fixing filters could do nothing about that. The question was therefore reformulated, and owned as such in the brief: do we want to build a measurable product that creates value for merchants, agencies and the company? Framed that way, it called for a decision, not a facelift.

Design the backstage along with the storefront. The agencies’ frictions (opaque scoring, leads with no follow-up, scattered information) were partly the symptom of an internal problem: the Channel Managers, responsible for agency portfolios, had no consolidated view and pieced reality together from emails and exports. The design brief set the principle: a smooth internal experience is a condition for a successful partner experience. Every agency journey in the target was designed as the mirror of the internal journey that makes it possible.

Build conviction along the way. Leaving a tool involves contracts, migrations and teams: it is not won in a readout meeting. The setup was designed against the tunnel effect: short, regular check-ins with the sponsors, learnings and doubts shared as they came, a discovery board open to everyone. The decision meeting was only ever meant to formalize a conviction already built.

Two prototypes side by side rather than one more report

That left the choice of how to bring the final decision. Three options.

A classic comparison file, weighted matrices and scores included. Rigorous, but with no grip on the real adversary: against sunk cost, every number in a table can be argued, and the status quo option always walks out alive.

A user test of the target prototype. Useful to validate an experience, but off topic for the decision: the sponsors’ doubt was not about the desirability of a better portal, it was about the necessity of leaving the tool to get it.

The third option was chosen: put two artifacts side by side. On one side, the prototype of the rebuilt experience, with a trust journey for the merchant, verified proof on agency profiles, a brief form that qualifies the request, a feedback loop that feeds the system. On the other, a simulation of the best the existing tool could do, built honestly, patches and broken journeys included. That honesty was the credibility condition of the case: a caricatured strawman would have destroyed trust.

The validation criterion was written before the presentation: not a score, not a test, but the sponsors’ reaction on seeing the two scenarios side by side. A value gap described in a report gets debated. Made navigable, it gets observed.

AI-assisted prototyping made this choice viable: two comparable, manipulable experiences produced in a few days, where weeks would have been needed. This practice, learned by doing, has since served as a model for other decision cases.

One honest note on tempo, because a story too smooth would serve no one: the case moved faster than its formalization. Collective conviction consolidated artifact after artifact, up to a decision that surprised no one. That was the point of the setup.

A decision made, lasting capabilities

The decision first. The organization left repair mode and made a structural choice: leave the market tool at the end of the contract and rebuild. The documented file is its record: diagnosis, options, proof, recommendation.

The capabilities next. The public directory, the most exposed piece, was rebuilt on prestashop.com, measurable by design: tracking plan, SEO and CRM integration specified before the first mockup. The in-house Expert portal followed, developed internally, now in production. Teams now change a partner journey without opening a ticket with a vendor.

The data black hole from the start received a structural answer, delivered with the production release. A North Star first: qualified connections per month, the value of the ecosystem reduced to a single indicator. Then the KPIs that break it down, measurable from V1: portal logins, with first logins isolated to track the post-migration window; completed merchant briefs, which materialize the lead at the source; contact requests sent to agencies; and the fate of each lead, tracked by status from first contact to won or lost. The whole fits in a tracking plan of about thirty events covering eleven business use cases, from agency activation to lead handling, and in a steering dashboard specified batch by batch, from vital to comfort. With an assumed discipline: a baseline period before any alert threshold, because fresh tracking is not interpreted, it is observed. The program that could prove nothing now knows exactly what it has to measure.

The practice itself, finally, and it is measured. The portal rebuild was the opportunity to transform the way of designing: documentation and the design and front-end evolutions driven end to end by directing AI rather than coding in its place. The orchestration report kept over the rebuild window gives the measure: 21,458 AI actions directed, spread across five verbs, explore, act, structure, document, delegate. The split tells the posture: execution is massively delegated to AI, structure and documentation stay human, so the product remains steerable once you are no longer in the loop. And the tool changed along the way, from one coding agent to another, without changing the method: the direction does not depend on the tool.

The organization, finally. The Build vs Buy vs Optimize framework serves as a reference for the other pieces of the ecosystem. The mirrored-journeys principle, external and internal designed together, applies to the following efforts. And preparing decisions with compared prototypes has become a practice that outgrows this case.

The next effort is open: turning the directory, now instrumented, into an engine of qualified connections between merchants and agencies. The value of an ecosystem is not measured by the profiles it displays, but by the connections it creates.

The mandate ended in August 2026, and the effort was handed over equipped to continue without me: the portal in production, the North Star and its measurement plan, the specified dashboard, the decision log. That was the design criterion from day one, and it is the real test of everything above.