The challenge

A 200-store specialty retail chain was running its weekly business cadence on 12 separate revenue dashboards. Finance owned three of them, merchandising owned five, store operations owned the rest, and each had grown out of a different incident, a different consultant, or a different leader’s preference. Same week, same stores, three different net-sales numbers — sometimes off by seven figures.

The damage compounded downstream. Reorder decisions for fast-moving SKUs had to wait for someone — usually a senior merchandising analyst — to manually reconcile two or three sources before a buyer would commit. Average decision latency on inventory reorders had drifted to 11 days. On seasonal categories, that latency was the difference between recovering a markdown event and eating it.

The executive team didn’t lack data. They lacked a version of it everyone trusted enough to act on.

The approach

We started by killing the question “whose number is right?” That meant treating the semantic layer as the deliverable, not the dashboards. Three weeks of work upstream of any visual:

• Rebuilt the revenue model on a single grain (store-day-SKU), single calendar (NRF retail calendar), single store hierarchy reconciled against the master file in the ERP.
• Catalogued every metric the 12 legacy dashboards exposed (147 distinct definitions, many overlapping). Mapped each to one of 38 canonical metrics. Documented the deltas where legacy definitions diverged from ERP truth, and flagged which divergences were intentional versus accidental.
• Stood up a reconciliation harness that runs on every refresh. Any canonical metric whose dashboard value drifts more than 0.1% from the system of record blocks publish until a human acknowledges it.

Only then did we touch the dashboards. We replaced the 12 with 3, each shaped to a role rather than a department:

• An executive view (one screen, weekly cadence, comp-store and total-chain trended).
• A merchandising view (category- and SKU-level performance, sell-through, weeks-of-supply, on the cadence the buying team actually runs).
• A store operations view (store-level performance, labor-to-sales ratio, traffic conversion, refreshed twice daily).

Every legacy dashboard URL was redirected to the appropriate replacement, with a short banner explaining which canonical metric had replaced the legacy one, so users saw continuity rather than a wall.

The results

The numbers we set out to move all moved within the first quarter post-launch:

• Decision latency on inventory reorders dropped from 11 days to 2 days, because buyers stopped pre-reconciling and started acting on the canonical view directly.
• The finance team’s prep time for the weekly business review fell from roughly 8 hours to 45 minutes — most of that previously was reconciling between dashboards, and that work no longer existed.
• Reconciliation variance against the ERP system of record sits below 0.1%, with the harness catching the rare drift before any user sees a wrong number.
• Active dashboard users grew 62% in the 90 days after launch, almost entirely from store operations and merchandising teams who had previously been getting numbers third-hand from regional managers.

The harder-to-quantify result is what the VP of Finance flagged: weekly business reviews stopped opening with a debate about which dashboard to trust. That argument had been a fixture of the cadence for years. It is gone.

What we would do again

The order of operations mattered. The team’s instinct — and the one we pushed back on — was to start by redesigning the dashboards. The dashboards were the symptom. The semantic model was the disease, and fixing the model first made the dashboard work almost mechanical. We would also build the reconciliation harness on day one again; it is the single artifact that has kept trust intact past the launch.