Architecture and consulting
Three systems agreed on the size of the book and disagreed on how much of it was done. The gap was work physically finished and never billed.
01Key numbers
02The situation
The controller rebuilt the forecast by hand in Excel every quarter. Three systems disagreed on every number, and three conflicting definitions of remaining work were live in production at the same time.
Salesforce invoicing had only started in January 2026 and held 208 records. Every earlier invoice existed only in the accounting system, so no single query could see the whole history.
The first useful output was not a forecast. It was an arithmetic statement: all three systems put the book of work at the same size and differed only on how much of it was finished.
Recognise this in your own org?
03What we built
A nightly Apex engine that writes one forecast row per order per month, and makes every row explain itself.
A recompute joins 16 archive objects and the live invoice records on a legacy order key, then writes three financial fields per order. Everything downstream reads those three fields, so there is one definition of remaining work rather than three.
Finished work is dropped. Work with a billing window spreads across that quarter on an 80/20 billing-to-closeout split. Recurring work spreads evenly across twelve months. Undated work lands in the current month and is flagged. Legacy dates fall back to straight-line, with stale dates collapsing to now.
Each forecast row stores a method string saying exactly how that dollar got into that month. Revenue never lands in the past. Rounding drift is corrected on the last row rather than distributed silently.
An invoice is a duplicate if it falls inside the system transition window and matches another to the dollar within seven days. Outliers above a stated ceiling are excluded as a data error. Both thresholds are published, so a reader can disagree with the number by disagreeing with the rule.
Backlog, less what is marked finished and never to be invoiced, gives what the engine actually spreads. Add open opportunities at 75% or better and the board headline follows. The arithmetic is on the page.
We will demonstrate any of these live, on the real org,.
115 of 154 open orders carried an end date that was in the past or blank. A structured refresh came back with all 153 answered. The stale-date bucket fell from 99 rows to zero, and the projected total fell with it. A forecast that only ever moves up is not a forecast.
04Outcome
The engine reproduces the hand-built book to within rounding and publishes the arithmetic that gets it there.
Reconciled against the accounting ledger across the live and completed eras. The unkeyed residue is listed by transaction.
Keep reading
Deposits, the standard 10% construction holdback and statutory declarations had no home in the org. The ledger that fixed it stores every amount as a positive number and puts the sign in the type.
The client told us the tax was calculating by product instead of by province. We tested that against all 2,335 order line items and it was not true. Six other defects were, and we priced each one.
Tell us what is broken. You get a written read on it within one business day, before any money changes hands.