What We Do
Know why the number moved,down to the driver that did it.
A dashboard tells you revenue fell. Working out why is the part firms charge a fortune for, and the technique behind it is not a secret: break the number into what drives it, then keep going until the cause is visible. We put that structure into a tool your team can run, without the jargon and without the engagement fee.
The technique
The method a strategy firm sellsis not actually a secret.
Take the number that moved. Break it into the things that produce it. Break those into their own inputs. Keep going until you reach something a person can act on. That is most of what a large engagement delivers, and the reason it works is that it forces the vague question, why is revenue down, into a specific one, which of these eleven things changed.
What firms actually charge for is doing that carefully, writing it down, and being consistent about it. The technique is teachable. Keeping it current after the consultants leave is the hard part, which is why the deck ends up in a drawer.
So we put the structure in a tool your own team runs, without the jargon or the engagement fee. The tree stays alive because using it is easier than rebuilding it.
What goes in it
How your business reasons, written down once.
Metric trees
Each headline number decomposed into the drivers underneath it, and those into segments and the levers you can actually pull. Recurring revenue broken into new, expansion, contraction and churn, each with its own sub-tree.
Reading rules
What counts as a real move versus normal noise. Which comparison is the right one for this metric. What you check first when it drops. The judgment your best analyst carries in their head, on the page instead.
Business context
The events that explain the chart. A price change in March, a system migration in June, three sites closed for refit. Dated, with the impact recorded, so next year nobody relitigates why that quarter looked strange.
Why this matters more now
Ask an AI why revenue felland it will guess plausibly.
Point an assistant at your warehouse with no context and it will produce a confident, well-written answer assembled from what generally causes revenue to fall in businesses that generally look like yours. Sometimes that is right. You will not be able to tell which times.
Give it your drivers, your definitions and your reading rules and the work changes. It walks the tree you built, checks the branches in the order you specified, and reports which one moved. The reasoning is yours and the answer is checkable.
Everything here is served to your warehouse over an API and to Claude over an MCP server, so the same definitions reach both. Most of what people call AI analysis today is the first version. This is what it takes to get the second.
And the filing cabinet
Half your business logicis in one person’s spreadsheet.
Region groupings. Cost centers. Product hierarchies. Which locations count as franchise. Which job codes are eligible for what. None of it comes out of a source system, all of it is needed by every report, and it usually lives in a workbook somebody emails around.
That is the reference data problem, and it is the reason two reports disagree even when the warehouse is fine. We host it properly, with versioning and validation, and serve it back as JSON, CSV or ready-to-run SQL for whichever warehouse you use.
It is the least exciting thing on this page. It is also the thing that stops being a problem permanently once it is done.
What you get
The tool this runs on.
Included with a retainer. Available on its own if the rest of your stack is already in good shape.
DimTable
Metric trees, prompt libraries and interpretation rules alongside your dimension tables and taxonomies. Serves the whole thing to your warehouse over an API and to Claude over an MCP server.
What changed
What it changes.
Same question, same answer
With the drivers and the reading rules written down, two analysts stop reaching two conclusions and an assistant stops inventing a third. The meeting argues about what to do rather than whose number is right.
AI that knows your business
Given your definitions, your drivers and your economics, an assistant reasons from how the company works instead of guessing from the shape of the data. Its answers stop needing to be checked line by line.
Out of the spreadsheet
Product hierarchies, cost centers and regions stop living in one analyst's workbook and start living where the warehouse can read them. The reporting no longer depends on that person being available.
What else
This one rarely shows up alone.
Can your team explain last month yet?
If the answer depends on who you ask, the tree is worth building.