Start with the Decision, Then Build the Dashboard

Codie Sanchez made the point on Modern Wisdom this month: owners need visibility into activity and results. We agree, with one addition — visibility starts with a decision, not a tool. We built ours the other way round, and it took under seven days.

The decision

At a multi-million-dollar e-commerce marketplace we help run, the weekly decision that mattered most was simple to state and hard to answer: which part of the funnel needs attention this week, and who owns the fix? Three systems held a piece of the answer — the web analytics, the shop platform, the email tool — and they rarely agreed. "How did the newsletter do?" used to take a person and the better part of a morning.

So the first document we wrote was not a data model. It was one page: the decision, the three numbers that inform it, the source that governs each number, and the person who acts when a number turns red.

The definitions

Every number on that page got four properties before anything was built:

• What it means, in one sentence a new hire could repeat.
• Which source wins when two systems disagree.
• The date range and the denominator — "conversion" with no denominator is an argument waiting to happen.
• When it refreshes, and how the reader can see that.

This is the unglamorous part, and it is where most dashboards fail. A screen can be beautiful and still be a place where two managers read the same number two ways. Definitions are what turn a chart into a decision. An example: "revenue" meant three different things across our three systems — gross at checkout, net of refunds, and attributed to a campaign. Until one of them was declared the governing number, every weekly review started with an argument about which chart was right.

What the screen shows

The first live view was the conversion funnel: traffic in, product views, add-to-cart, checkout, paid — split by device and by traffic source, for the last 30 days.

It shows where the drop is concentrated. It does not, on its own, say why, and it does not attribute a sale to a campaign — that view is still in progress, and the dashboard says so. A dashboard that pretends to know more than it does trains people to ignore it.

How we built it

The facts: a founder who does not write code, one junior developer, and Claude Code. Under seven days from that one-page spec to the first live view. Under $10,000 for the first deployment. The pipeline is ordinary — GA4, Shopify and Klaviyo data into BigQuery, transformations in dbt, an MCP layer so an AI assistant can query it, and the dashboard on top.

The AI wrote the code. The humans wrote the spec, the definitions and the checks, and read every output with the suspicion you would apply to a new analyst's first report. That division held. Where the model got a definition wrong, the one-page spec was what caught it — not a code review.

What it does not do (yet)

Attribution and lifetime value are in progress. Predictive churn is not built. Nothing on the dashboard acts on its own: it reports, and a person decides. We think that is the right order for a business of this size, and it is the order we install for clients.

What broke

Nothing in the system. What broke was the ability to hide. When activity and result sit on one screen with an owner beside them, "the numbers are unclear" stops being an available answer. That is uncomfortable for a week and healthier for a year.

The next decision, and who owns it

Every red number on the screen has an owner, a next action and a review date. That is the whole management method: see it, own it, act on it, look again on Friday.

Where to start

If you run a business with 5 to 100 people and one weekly decision still takes a person and a morning to prepare, start with the decision, not the tool. A free 15-minute audit maps what that decision is costing you. Applications for the Founder Clarity Sessions closed on 2 October; the five sessions run over the following two weeks, and we will show what came out of them here.