Why this exists

Their assistant helped with everything. Except the numbers.

We have spent ten years building the planning and reporting systems where companies keep their numbers. In 2026 the people we built them for started asking us the same thing. Their assistant could draft the email, summarise the meeting and fix the formula. So why could it not tell them what moved in last month's forecast?

The figures sat in a planning model, a CRM and a document library, and the assistant could reach none of them. Getting an answer still meant a report request and a wait.

The assistant was already on every desk. The numbers were not in it.

What they asked for

Three requests, every time.

Nobody asked whether the model was clever. They asked for three things, and none of them was another tool to learn.

Let me ask in my own words

No report builder and no query language. The question a planner would put to a colleague, answered from the model.

Tell me where the number came from

Which system, which environment, read when. A figure nobody can trace is a figure nobody uses in a meeting.

Let me act on it

Find what moved, then load the corrected budget without keying it into another screen by hand.

What we built

One connector between their assistant and the systems they already run.

We built it for the clients who were asking, and it became the product. It works in Claude and in Microsoft Copilot.

  1. 01

    Ask in plain language, get the figure

    The planning model, the CRM, the documents and the task tracker, read live at the moment somebody asks. Every answer names the system and the environment it came from.

    Every tool, by what it does, is on the tools page.
  2. 02

    Find what moved

    Ask which cost centres moved and the answer ranks them and names the lines behind each one, so month end starts at the cause.

    How finance teams use it: finance and planning.
  3. 03

    Load a change, with a preview first

    Hand the assistant a workbook. It checks every line against the model, shows what will change, and posts only when a person confirms, with a receipt to check the figures against.

    The whole path is on how it works.
  4. 04

    Words mean what finance says they mean

    Approved, versioned metric definitions, so "revenue" and "contribution" mean the same thing in every answer as they do in your finance team's own reports.

What changed

The question gets asked and answered the same day.

A budget holder asks why a cost centre is over. The answer arrives in the chat they already have open, with the lines behind it, before anybody has opened a report.

Before

  • A report request, and a wait of days for the one person with the login.
  • A figure pasted into a spreadsheet, with nothing to say where it came from.
  • A budget keyed in by hand, line by line, into the planning tool.

Now

  • The question asked in Claude or Copilot, and the answer back in seconds.
  • The figure with its source attached: the system, the environment and when it was read.
  • A workbook previewed, confirmed and posted, with a receipt to check it against.

What stays the same

Your data stays where it is.

Every figure is read from your own system when somebody asks, and nothing is copied into a second store. Each person signs in with their work account and reaches only what you give them. Writing is off unless you switch it on for a project, and then a person confirms every change. We run no AI model: the assistant, and the bill for it, stay yours.

Ready?

Bring the question your assistant cannot answer yet.

Pick one your people ask every month and wait days for. We connect the system that holds the answer, and you watch it come back in Claude or Copilot.

Why this exists: our clients rolled out AI assistants that were useful everywhere except where their numbers lived. This connects those systems, so the assistant can finally answer the questions people care about.