Numbers to Judgment
F&PAnalystAIly · Discovery map

Numbers to Judgment

How an FP&A analyst gets from a count to a call, where the hours go, and why an AIly makes that work faster and more accurate.

Erix Pizano, XireTechOctober 2026Hypothesis · validate in discovery

These are the workflows most mid-size FP&A teams run. They're what we walk into discovery with, so the first conversation is about their work, not about AI. Every box below gets replaced by what Meridian's team actually does.

1The basic loop

Look, organize, form a recommendation is the right shape. Two steps hide inside it, and they're where most of the hours and most of the risk sit:

Verify, inside "Organize"Before anyone compares anything, the numbers have to tie to the closed books. Every later step inherits an error here.
Explain, inside "Recommend"Finance can see that a line moved. Only the person who runs that budget knows why. The analyst proposes a cause and the owner confirms it. That's the first place judgment is required.

Most months the output isn't a recommendation. It's an explanation. A recommendation comes when the explanation changes the outlook.

1 GetPull actuals fromthe books and theplan from themodelAutonomy2 VerifyDo the numbers tieto the closedbooks? Nothingmoves until theydoAutonomy3 OrganizeMap accounts toreport lines,compare to plan,rank bymaterialityAutonomy4 ExplainWhy did it move?Analyst proposes;the line ownerconfirmsAssist + Aware5 RecommendWhat it means forthe year, and whatto do about itAssistance6 DecideDirector or CFO:hold, reforecast,escalateYOUR THREE STEPSLookOrganizeForm recommendationThe decision becomes next period's plan, and the loop starts againAssembly: takes the hoursJudgment: needs a personGate: a person signs offWhich A the AIly brings

Figure 1. The loop every FP&A workflow repeats. Four of six steps are assembly. Judgment sits at the gates.

2Where the numbers come from

Nobody makes them up, and the analyst doesn't count them. The business counts: every time a client is billed, a payroll runs or a vendor invoice is approved, a system of record captures it. Those systems post to the general ledger. At month-end the controller's team closes the books: they reconcile, accrue and lock the period. The usual benchmark puts that at about six days.

Is their repository populated already? Half of it is. Actuals arrive finished, from the analyst's point of view, and they re-pull them every month. The other half the analyst builds and maintains:

The planBudget and forecast. FP&A builds it with the budget owners, line by line, driver by driver.
The mappingWhich of hundreds of ledger accounts roll into each line of the report. Usually a tab in a workbook that one person understands.
What they knowWhy that line is always high in March. What the CFO doesn't want to read. Last quarter's explanation. It lives in old commentary, email threads, cell comments and people's heads. No system stores it.

How do they get insight out? Mostly in Excel. They export from the ledger, paste into last month's workbook, re-point the formulas, re-check them, and pivot. Larger teams add a planning tool or a BI layer on a warehouse. In the most recent FP&A Trends survey, 52% of teams still plan in Excel.

1 · THE BUSINESS COUNTSRecorded where it happens2 · SYSTEMS OF RECORDEach team's own system3 · ACCOUNTING CLOSESNot FP&A: the controller's team4 · FP&A'S REPOSITORYWhat the analyst actually works from5 · ANALYSISClient contracts, feesbilledBilling / CRMRevenue, receivablesHires, payroll runs,bonusesHRIS and payrollPeople cost, the largestlineVendor invoices,purchasesAP / procurementOperating spendCash in and outBank feedsCash, reconciliationsGeneral ledger (ERP)Journal entries from everysubledger; one chart ofaccountsMonth-end closeReconcile, accrue, adjust,then lock the period.Median about 6 daysTrial balanceClosed actuals by account,by monthActuals extractReport export, savedsearch, or a query on thewarehouse. Re-pulled everymonthThe planBudget and forecast. FP&Abuilds this one, with thebudget ownersThe mappingWhich accounts roll intowhich report line. Usuallya tab in a workbookWhat they knowPrior commentary, emails,cell comments, what's inpeople's heads. StorednowhereThe workbook chainLast month's file,copied, re-pointed, re-checked. Actuals, planand mapping meet hereCommentary and thedecisionAwareness lives hereThe business populates actuals. Accounting makes them final. FP&A builds the plan and keeps the judgment, mostly in files and heads.

Figure 2. From a business event to a number on the analyst's screen. The dashed box is the part the AIly makes permanent.

3The monthly cycle, by role and day

This is the loop from Figure 1 laid across a real month. It's the workflow Meridian's eight analysts run for monthly close, budget-vs-actuals, variance analysis and management reporting.

AccountingFP&A analystBudget ownersDirector / CFOWorkdays 1–5Day 5Day 6Days 6–7Day 7Day 8Days 9–10Close the booksReconcile,accrue, lock theperiodPull and tieoutExport actuals,map accounts, tieto the trialbalanceAutonomyCompute andflagBudget vs actual,every line; flagwhat's materialAutonomyDrill and askGL detail,invoices,headcount; writethe ownersAssistanceAnswer whyOften late, oftenpartial. The waitis the hiddencostWritecommentaryCause, timing orstructural, whatit impliesAssist + AwareReview and editEvery edit is alesson that'susually lostPackage andcheckEvery numbermatcheseverywhere itappearsAutonomyApprove andsendTo the CFO, thenleadershipcorrectionsTypical month for a mid-size team. The days and the people are what discovery confirms.

Figure 3. Close to approved pack. The yellow arrows are the two loops that cost the most: waiting on owners, and review edits that are never kept.

StepWhere the hours goWhere it breaksAIly
Pull and tie outExporting, pasting, re-pointing formulasA broken link or a missed account; the base is wrong and nobody knowsAutonomy
Compute and flagEvery line, every month, against every thresholdA material line slips through in a busy weekAutonomy
Drill and askDigging through ledger detail; writing owner emailsVague questions get vague answers, lateAssistance
Write commentaryWriting from blank; re-explaining recurring itemsAn unconfirmed cause reads as fact; this month contradicts last monthAssist + Aware
Review and editThe Director rewrites the same things each monthThe lesson in the edit is lost by next monthAwareness
Package and checkMatching numbers across slides, tables and textA figure in the summary doesn't match the tableAutonomy

4The forecast loop

Monthly actuals feed the forecast. The forecast is what next month is measured against, and each quarter the approved forecast becomes the board narrative.

Actuals landClosed monthreplaces itsforecastAutonomyRefresh driversHeadcount, feerates, volumes,run-ratesAwarenessOwners updateHiring plans,deals, known one-offsAssistanceRoll forwardRe-run the modelto year endAutonomyExplain the moveVersus lastforecast: whatchanged and whyAssist + AwareHold orreforecastDirector / CFOBoard narrativeQuarterly, fromthe approvedforecasteach quarterThe approved forecast is what next month is measured against

Figure 4. The reforecast cycle. The judgment is whether the year has changed.

5Why the AIly makes it faster and more accurate

"I just don't see how it applies to what we do. Our work is numbers-based and requires judgment."

They're right, and that's exactly the case. Judgment happens at a few gates. Everything around those gates is assembly, and assembly is where the month goes. In the 2024 FP&A Trends survey, teams spent only 35% of their time on high-value work like generating insight. An AIly takes on the assembly and prepares the evidence for each gate. The judgment stays with the analyst.

Why David's prompts gave mixed results. A prompt arrives with nothing: no account mapping, no thresholds, no last month, no house voice, no checks. It's the same model with nothing to stand on. The ContAIner is what was missing.

Autonomy

The checks a reviewer does by eye, run every time
Speed
Tie-out, budget-vs-actuals, the materiality scan and the cross-document number match run in minutes, the same way every month.
Accuracy
Field audits found errors in at least 86% of the spreadsheets they examined, with 1–2.5% of cells wrong. A check that never gets tired catches what a reviewer misses in close week. It flags; it never fixes.

Awareness

What the team knows, brought to every task
Speed
A recurring item is explained once. Prior commentary, corrections and house voice load with every task. A new analyst starts with the team's memory on day one.
Accuracy
This month's story is checked against last month's. A Director's edit becomes a standing rule, so the same correction isn't made twice.

Assistance

First drafts of the writing around the numbers
Speed
Owner questions written so they can be answered yes or no, which shortens the wait. Commentary and narrative in the house voice. The analyst edits instead of starting from blank.
Accuracy
Every line is labeled as a fact or a hypothesis. An unconfirmed cause can't slip into the pack as fact.

What we measure instead of promising a percentage. Discovery sets the baseline; the pilot measures against it:

  1. Hours per step, per close
  2. Days from close to an approved pack
  3. Owner response time on variance questions
  4. Director edits per draft, and how many repeat
  5. Errors caught before review versus after

6How discovery gets the real diagrams

Follow one number. Take one line from last month's management report. Trace it backward to the transactions it came from, then forward to the decision it informed. Ask to watch it happen on screen, during a real close if possible. People describe their work as it should be; their screen shows what it is.

Questions, in the order the number travels

StageAsk
SourcesWhich systems hold revenue, people cost and vendor spend? Which ones does FP&A pull from directly?
Close handoffOn what day are the books final? How do you know? What changes after that?
RepositoryWhere do actuals land when you pull them: a workbook, a planning tool, a warehouse? Who maintains it?
MappingHow do ledger accounts become report lines? Who changes the mapping, and how often does it break?
Tie-outHow do you prove the numbers tie? What did the last mismatch cost you?
MaterialityWhich variances get explained? Is the threshold written down, or known?
OwnersWho do you ask why? How do you ask? How long do answers take?
CommentaryWalk me through last month's explanation of one line: every step, every click.
ReviewWhat does David change in a draft? Does the same edit come back next month?
PackageHow many places does one number appear in the pack? How do you check they match?
ForecastWhat drives the forecast? How often does it move, and who decides to reforecast?
JudgmentWhich calls do you make alone? Which ones need David? Which would you hand off tomorrow?

Sources

F&PAnalystAIly is built on the Cognitive Inheritance Framework. © 2026 XireTech.