know.2nth.ai Finance fin planning
fin/planning · Reference leaf

Planning & FP&A,
on ERP data.

Reporting looks back; planning looks forward. FP&A — budgets, rolling forecasts, scenario models — runs on the same ERP-fed warehouse, just pointed at the future. This leaf stays ERP-agnostic, leans on driver-based models instead of a thousand-tab spreadsheet, and keeps the human owning the assumptions.

Reference Budgets & forecasts Driver-based Same warehouse as reporting

Forecasting the business off the numbers it already records.

Financial planning and analysis is the forward half of finance: the annual budget, the rolling forecast that updates it, and the scenario models that ask "what if input costs rise 8%?". Where reporting tells you what happened, FP&A tells you what's likely to — and what to do about it.

The crucial idea: planning isn't a separate dataset. A good forecast is built on actuals, and the actuals already live in the ERP and the warehouse the reporting leaf feeds. FP&A reuses that foundation and layers drivers on top — volume, price, headcount, capacity — rather than re-keying history into a spreadsheet that drifts from the books within a month.

The plan is a model, not a system of record

A forecast is an opinion about the future, owned by finance. The agent and the tooling assemble it from history and assumptions; they never post it to the ledger. When a plan becomes a commitment (a budget the business is held to), it's a human decision, versioned alongside the reports.

Two ingredients: what happened, and what moves it.

Every forecast is history extended by assumptions. The history comes from the ERP; the assumptions are the drivers a finance team actually argues about.

IngredientSourceExample
ActualsERP / warehouse martsLast 24 months of revenue, COGS, opex by cost centre
Volume driversSales pipeline, ops capacityUnits sold, machine hours, billable headcount
Price driversAssumptionSelling-price increases, input-cost inflation
Cost driversHR + ERPHeadcount plan, salary review %, supplier terms
Cash driversAR/AP ageingCollection days, payment terms, capex timing

Driver-based planning is the difference between a forecast you can defend and a number you typed. Change one assumption — "we win the big contract, volume +20%" — and the model recomputes margin and cash without anyone editing 40 cells. The drivers are the conversation; the maths is the tooling's job.

Three artefacts, three cadences, one model.

FP&A produces three related things that teams often blur together. Keeping them distinct is half the discipline.

ArtefactQuestion it answersCadence
BudgetWhat did we commit to this year?Annual, then fixed
ForecastWhat do we now expect, given actuals?Rolling — monthly or quarterly
ScenarioWhat if a key assumption changes?Ad-hoc, on demand

The plan-vs-actual view is where this connects back to reporting: every month, the management pack shows the forecast next to what really happened, the variance, and the narrative. That loop — forecast, measure, re-forecast — is the entire point of FP&A, and it only works when the actuals and the plan sit in the same warehouse.

Planning as code, on the warehouse you already built.

If you've stood up the warehouse + Evidence stack for reporting (see the reporting leaf), FP&A is mostly an extension of it — not a new tool to buy. ERPNext's open data makes the actuals free to pull; the planning logic lives as version-controlled SQL alongside the reports.

Actuals — the reporting marts (GL, costing) already in the warehouse
Drivers (assumptions as a small, editable, version-controlled table)
Model (dbt: forecast = actuals trend x drivers)
Plan-vs-actual (Evidence: variance pack next to the report pack)

Driver tables in Git mean a forecast is auditable: you can see exactly which assumption changed between the March and April forecasts, and who changed it. That's something a sprawling spreadsheet can never give you. Heavier dedicated FP&A platforms (Pigment, Anaplan, Planful) exist for large, complex planning estates — but for a mid-sized business already on an open ERP, the warehouse-plus-code approach is far cheaper and reproducible.

The agent proposes; finance owns the assumptions.

Forecasting is judgement plus arithmetic. The agent is excellent at the arithmetic and the first-draft narrative, and must stay out of the judgement.

Baseline forecast

Extend the trend from actuals, seasonalise it, and present a baseline the team can argue with. A starting point beats a blank sheet.

Scenario generation

"Show me downside, base and upside on input-cost inflation." The agent spins the driver dials and renders the three P&Ls; finance picks which to plan against.

Variance commentary

Draft the plan-vs-actual narrative each month from the numbers. Finance edits the story; the figures come from SQL.

Assumptions stay human

The agent never decides that volume will grow 20% or that the contract lands. It models the consequence of an assumption a person owns and can defend to a board.

Warehouse-backed planning, or a spreadsheet?

Build it on the warehouse whenA spreadsheet is fine when
You re-forecast monthly and want plan-vs-actual automaticallyYou budget once a year and rarely revisit it
Multiple cost centres / entities to roll upOne simple P&L, a few lines
You need to defend which assumption changed whenThe plan is directional and low-stakes
You already run the reporting warehouseThere's no warehouse and the ROI isn't there yet

The honest line: a spreadsheet is a perfectly good FP&A tool for a small, stable business. The warehouse approach earns its keep when the forecast is frequent, multi-dimensional, and has to reconcile to the actuals — which is exactly when spreadsheets quietly start lying.

Plan in rand, with the rand's volatility in the model.

Two local realities belong in the drivers, not the footnotes. Forex: if you import inputs or export goods, the ZAR/USD rate is a planning assumption with real margin consequences — model a range, not a point. Interest rates: SARB's repo rate feeds finance costs and the cost of working capital, so a rate-sensitivity scenario is worth keeping warm.

B-BBEE spend targets are themselves a plan — procurement and skills-development budgets that the same model can track against actuals. And the discipline holds: an agent can model the rand at R19 or R21, but the assumption it plans on is a human's call.

Planning shares reporting's foundation.

FP&A and reporting are two views on one warehouse — backward and forward.

Primary sources.

The open tooling the warehouse-backed planning approach is built on.