See an exampleGet a pilot dashboard

How it works

How we build a dashboard you can check.

Every Number starts from the decision you need to make, proposes a few metrics, and gives every number a receipt. During the pilot, a person does this work by the same rules.

Get a pilot dashboardSee the examples

Four steps

1. The decision

The 15-minute call · Sample
  1. Who reads this dashboard?The steering committee, every Monday.
  2. What do they decide with it?Which projects need help this week.
  3. How often?Weekly.
  4. What does good look like?Nine or more projects on track, no late milestones.

Goal Decide which projects need help, weekly.

Start from what you need to decide.

On a 15-minute call we ask who reads the dashboard, what they decide with it, how often, and what good looks like. The answers decide what goes on the page.

2. The metrics

What the AI sees: 12 column names and your goal
  • project
  • owner
  • status
  • update_date
  • milestone
  • due_date
  • done_date
  • budget
  • spend
  • risk
  • rating
  • closed
  1. Projects on trackReady
  2. Milestones hitReady
  3. Budget usedReady
  4. Open risks by ratingReady
  5. Missing updatesReady
  6. Forecast at completionNeeds a remaining-estimate column

No rows were sent. You approve every metric before we build · Sample

Five to seven metrics, not a wall of charts.

We match your goal to metrics that answer it, and check which ones your sheet can already calculate. The AI suggests from your column names and your goal. It never sees the rows.

3. The definitions

Active clients128
Definition

Clients with a paid invoice in the last 90 days

  1. v1Jul 2Created: any invoice counts
  2. v2Oct 9Paid invoices only · 14 fewer clients

Sample data

Every metric is written down once.

Each metric gets a plain-English definition you approve. When a definition changes, the change is dated and its effect on the number is shown.

4. The receipts

Active clientsOctober · Sample data
Definition
Clients with a paid invoice in the last 90 days
Source
invoices.xlsx · 1,284 rows
Filters
Status is not “Test”
Left out
14 rows: 3 subtotals, 9 blank amounts, 2 duplicates
Version
v2, since Oct 9
Data through
Oct 15

Every number shows its work.

Definition, filters, rows used, rows left out, source file, and the date the data runs through. One click on any number.

Anatomy of a receipt

Six facts stand behind every number on a dashboard. Click one.

Active clients, October128Sample data

Definition

What the number counts, in one sentence you approved.

Clients with a paid invoice in the last 90 days

Six rules we don't break.

They come from how we designed the product. In the pilot, the person building your dashboard follows them by hand.

  1. 1Same data, same number.Run it twice and you get the same result. The AI isn't involved when a dashboard calculates.
  2. 2No number without a receipt.If a number can't show where it came from, it doesn't go on the dashboard.
  3. 3The AI never types a number.Any sentence on the dashboard takes its numbers from the calculation, never from the AI.
  4. 4The AI never sees your rows.It gets column names, column types, and your goal. Nothing else.
  5. 5The AI never touches your data.It describes the dashboard. A database engine does the counting.
  6. 6Combining tables never inflates totals.When two tables are joined, each amount is still counted once.

What a person does in the pilot

The product will automate this after launch. For now, one person does all of it, in this order.

  1. 1

    Reads your sheet

    Finds the table, the header, totals rows, blanks, and date formats.

  2. 2

    Proposes metrics on the call

    Five to seven, matched to the decision you described.

  3. 3

    Writes the definitions

    One sentence each, which you approve before anything is built.

  4. 4

    Calculates with a database engine

    No number is typed by hand or by an AI.

  5. 5

    Checks each number by hand

    Against your sheet, before it reaches you.

  6. 6

    Delivers in 5 business days

    With the definitions page and every receipt.

Questions

More about your data on the security page

Does the AI make up numbers?

No. The AI suggests what to measure. A database engine calculates every number, and each one links to the rows behind it.

Why not let the AI write the queries?

On real, messy company data, AI-written queries fail far more often than on clean examples, and the worst failures still run and return a number. We let the AI describe the dashboard. A compiler and a database engine do the rest.

What exactly is a receipt?

The six facts behind a number: definition, filters, rows used, rows left out, source, and the date the data runs through.

What does a person do in the pilot?

Everything on this page, following the same rules. The product automates it after launch.

What happens when my sheet changes?

In the pilot we build from the sheet you send. After launch, new versions keep the same definitions, and any change is dated.

Do you train AI on my data?

No. Your spreadsheet is used only to build your dashboard. To suggest metrics, we send only your column names and your goal to Anthropic's API, never the rows. Anthropic doesn't use API data to train its models.