Decompose the movement
A 14% increase broken into the vendors, customers, projects, departments, and one-off items responsible, each with its dollar contribution to the change.
AI capability
Every system can tell you an account moved 14% against last quarter. The work — the part that takes a controller two days a month — is decomposing that into the specific things responsible and writing down why. That decomposition is a traversal when every transaction carries its dimensions.
What it does
A 14% increase broken into the vendors, customers, projects, departments, and one-off items responsible, each with its dollar contribution to the change.
Ordered by effect on the variance rather than by size. A small account that doubled often explains more of the movement than a large one that drifted.
Did the cost rise or did you buy more? These have different causes and different fixes, and a single percentage conflates them.
A variance caused by an invoice landing on the first rather than the last of the month is not a real movement, and labelling it as timing prevents a pointless investigation.
A written explanation in the register your board pack uses, as a first draft. Editing beats composing from a blank page every month.
From the variance to the contributing account to the individual transactions and their source documents.
In most finance teams, variance analysis is a controller exporting two trial balances into a spreadsheet, computing differences, sorting by magnitude, and then investigating the top handful by opening transactions one at a time. It takes most of two days and it is re-derived from scratch every month.
It also stops early. The controller investigates the largest few movements because that is what fits in the time, which systematically misses the case where three medium movements in the same direction share one cause.
An expense line up 22% has two very different explanations. You paid more for the same thing, or you bought more of it at the same price. The first is a vendor conversation, the second is an operational one, and reporting the combined percentage tells you to have neither.
Separating them requires unit-level data, which exists because line items and quantities are captured during extraction rather than summarised at the header. It is one of the more useful outputs and one that spreadsheet-based analysis almost never produces.
A meaningful share of month-over-month movement is not movement at all — it is an invoice arriving on the 1st instead of the 31st, or a payroll period with three runs instead of two. Investigating those is wasted effort and, worse, explaining them in a board pack as though they were real makes the commentary less trustworthy.
Timing effects are identified and separated before ranking, so the list you review is actual movement.
Limits
Every capability page on this site carries one of these, because a feature described without its boundaries is a claim rather than a description.
It will tell you margin fell because one client shifted toward a lower-margin service line. Whether that was a pricing failure or a deliberate strategy is context it does not have.
A variance cannot be decomposed by department if the transactions carry no department. Where that is the case it says so rather than producing a partial answer that looks complete.
Variance against a budget nobody believed in is arithmetic against a fiction. The analysis is only as useful as the comparison basis.
Questions
Two periods and one puzzling line is enough to show you the decomposition and the transactions behind it.