Your team already has the cashflow-model and budget-vs-actuals playbooks — the worked recipes for projecting six months from a starting balance, and for lining a budget up against what actually happened, step by step. Each is a good move, done once. This module is the layer above the recipe. It’s where you master the forward-looking discipline — a model you can defend line by line and flex on demand, and a variance story leadership can act on — build them for your own books, and prove, against a real rubric, that you can.
It’s Module 3 of the certifiable Finance & Ops track. It inherits the three foundation files you built in Module 1 — the categorization-rules.md that makes every cost land in the same bucket, the data-checklist.md that makes the inputs trustworthy, and the finance-controls.md that names who signs off — and it follows Module 2, where you turned those files into a reconciled, reported month. M1 made the books trustworthy; M2 reported the month that closed; M3 is where finance stops reporting the past and starts shaping the decision — the forecast leadership plans against and the variance story it acts on.
A model you can’t explain is a model you can’t defend — and the first time that matters is the worst possible time to learn it, in a board meeting, with a runway number on the screen you can’t trace to its inputs. Planning is where finance stops reporting the past and starts shaping the decision: the forecast that sizes the hire, the variance that says which gap to chase and which to ignore. Master it and your numbers stop describing what happened and start changing what happens next.
A model you can defend beats a model that’s “right”
The default failure isn’t a wrong number — it’s a right one nobody can explain. A model lives in a spreadsheet someone inherited, half its cells reference cells nobody remembers, and the runway number it spits out is trusted because it looks confident and it’s always been there. That works until the month it doesn’t, and then a broken carry-forward that’s been quietly off for a year surfaces in front of the board. The discipline that prevents it is one rule applied without exception: every formula is explained in one plain sentence, or it doesn’t go in the model.
- A black box gets trusted until it’s catastrophically wrong. The danger isn’t that an unexplained model is obviously broken — it’s that it looks fine. Nobody questions a number they can’t see inside, so the error compounds silently across six months of formulas until the stakes are highest. The fix is to refuse the black box: if a line can’t be said in a sentence — “closing equals opening plus net” — you don’t understand it yet, and a number you don’t understand is one you can’t defend to an auditor, a board member, or the FTA.
- You own the model; Claude builds it. This is a hard line. Claude assembles the table fast, carries the balances forward, runs the arithmetic without a slip — but the assumptions are yours, the logic is yours, and the verification is yours. The verify gate is concrete: spot-check one month by hand against the formulas. If month two’s closing matches your own arithmetic, the method holds across all six; if it doesn’t, you’ve caught the broken link before the board did.
- Show the work on every line. The same discipline M1 set for a total applies to a forecast: Claude states the formula or the steps beside every figure — “net equals inflows minus the recurring costs from
categorization-rules.md.” A model where you can read the method is one a teammate can re-run and a reviewer can trust. A model where you can’t is a guess with a chart on top. - A model that drives a decision gets a second pair of eyes. This is the controls gate, and it’s load-bearing here. The moment a forecast sizes a hire or justifies a spend, it stops being a spreadsheet and becomes a decision — and the preparer-≠-approver line from M1 applies to the decision, not just to a payment. A wrong runway number that greenlights three hires is the expensive kind of wrong. The person who built the model is never the only person who signs off on what it recommends.
The formula in one sentence
Here’s the mechanics of a model you can stand behind — and the altitude that turns one model into an answer for every “what if.” The point isn’t the spreadsheet; it’s that the spreadsheet is legible, so changing one input gives you a new decision you can trust as much as the first.
- The model is its assumptions — state them first. A forecast is only as good as what enters it, so the assumptions go on top, in writing: the opening balance, the recurring costs (pulled from
categorization-rules.mdso they bucket the way every other report buckets them), the revenue line, the churn or growth assumption, and any one-offs by the month they hit. Make Claude read them back to you before any math — a wrong input caught here is one number; caught in month six it’s a model rebuilt. Garbage in, confident garbage out. - Build one month fully, then roll it forward. Build month one as a table — opening plus inflows minus outflows equals net; closing equals opening plus net — and explain each line in a sentence before extending. If you understand month one, the rest is repetition, and you’ve verified the logic before it copied itself six times. Then carry each month’s closing into the next month’s opening; the first month the balance goes below zero, if any, is the runway.
- Model once, scenario many. This is the leverage. One model’s assumptions hold every what-if inside them — you don’t rebuild to answer a new question, you flex an input. The distinction that matters is scenario vs. rebuild: a scenario toggles a single assumption on the same model and watches it respond — churn +2pt, hire 3, revenue flat — with every other line still tracing to the inputs you already verified. A rebuild is asking Claude for “a new model where we’ve hired three people,” which quietly changes more than the one variable and breaks the carry-forward you just spot-checked. Flex the assumption; don’t rebuild the sheet.
- The scenario is where the model earns its keep. A single projection is a guess dressed as a plan; a model you can flex is a decision tool. “Churn +2pt → revenue erodes → net flips negative by month four” and “hire three → +AED 90,000/mo → the cushion that was growing now drains” are the answers leadership actually asked for — and because every line traces, you can defend the scenario with the same confidence as the base case. On Desktop you edit the assumptions in the file pane and re-run; next month’s forecast is changing three numbers and accepting the diff, not rebuilding from zero.
Variance is a story, not a number
Budget vs. actuals is the other half of planning: not what you think will happen, but what the plan got wrong and why. The craft people skip is that a variance report’s job isn’t to restate the gap — leadership can read a red cell — it’s to explain it. “Software AED 8,000 over” is useless; “the analytics tool’s silent price hike plus a prepaid annual” is a decision.
- Don’t restate the delta — name the cause. A table of red and green cells tells a reader that something diverged, never why, and the why is the entire value. “Marketing 30% over” could be a problem, a one-off you approved, or a timing shift that reverses next week — and those three call for three completely different responses. A variance you only restate is a variance you’ve made leadership investigate themselves.
- Price, volume, or timing — the three things a variance can be. Every gap is one of these: price (the same thing at a higher rate), volume (more of it than planned), or timing (it landed in a different month than budgeted). Forcing each flagged gap into one of the three is the whole trick, because the classification is the action: a timing variance reverses next month and needs nothing; a price or volume variance is a real trend that might need a decision.
- The context is yours; the classification is shared. Claude can classify a gap as price, volume, or timing from the numbers, but the real why — the campaign you approved mid-month, the renewal pulled forward, the vendor that raised its rate without telling you — is context only you hold. Hand it that context so the narrative explains causes; where it’s guessing, it writes “cause unclear, investigating” rather than invent one. A plausible-but-wrong cause sends someone chasing the wrong fix, which is worse than no cause at all.
- End in a narrative, not a spreadsheet. The deliverable leadership reads is a few sentences — the total over or under, the two or three gaps worth explaining with their cause, and which are one-off versus ongoing — with the full table as a verifiable appendix underneath. The narrative is what separates “here’s a variance report” from “here’s what changed and what to do about it.”
The threshold that makes variance useful
A variance report that flags everything flags nothing. Every line varies by a few dirhams; surface all of them and the three that matter drown in the rounding. The fix is a single decision made up front and written down: the materiality threshold — the size of gap worth a sentence.
- Set the floor to what leadership would actually act on. Flag only variances over a dirham floor or a percentage, whichever is larger — say, over AED 2,000 or over 10%. Set it too low and the report is noise; too high and a real creep hides beneath it. Pitch it at the gap your leadership would genuinely want a decision on, and write it in the file so it’s identical every month rather than re-argued each close.
- Threshold and classification are the two-stage filter. The threshold decides which gaps get explained; price/volume/timing decides what the explanation is. Together they turn a wall of red cells into a short, ranked list of things that actually matter — the three or four rows that need a sentence, each already carrying its cause. One without the other half-works: a threshold with no classification surfaces gaps but doesn’t explain them; classification with no threshold explains everything and so highlights nothing.
- The threshold is respect for the reader. It isn’t laziness — it’s the difference between handing leadership a spreadsheet to interpret and handing them the decisions to make. A report that surfaces only signal earns the reader’s attention the next month; one that cries wolf on every wiggle gets skimmed, and the one variance that mattered gets skimmed with it.
The Arabic plan narrative — authored, a peer lane
For a MENA team the Arabic forecast and variance story isn’t a translation pass bolted on after the English ships. It’s authored, as the M1 bilingual standard requires — and a plan has its own edges a translation gets wrong.
- The narrative is authored, not translated. A forecast or variance story for an Arabic-reading owner is written in Arabic from the figures, not the English run through a translator. In finance, where the whole product is trust in the numbers, a narrative that reads translated reads untrustworthy — and a board that doesn’t trust the narrative doesn’t trust the forecast under it.
- Localize the framing, not just the words. What a Gulf owner reads a forecast to learn — the cash position, the runway, the decision it forces — and how directly you frame a “we’re starting to burn” scenario, differ from the Western order. Lead the Arabic narrative with the decision, not the table; re-shape what it surfaces first. You adapt the framing; you never adapt the finished English sentence.
- Numbers Western, currency AED, layout RTL. The conventions from M1’s bilingual standard hold: numerals stay Western even in Arabic copy, currency is AED, and the variance table runs right-to-left with the columns of numbers still legible and aligned the way a finance reader expects. A forecast that switches numeral systems mid-table reads amateur to the person signing off.
- Same verify gate, both languages. The Arabic forecast’s every formula is explained and every variance traces, exactly as the English. The failure mode is the approver who can’t read the Arabic narrative waving it through — so the gate is cleared by someone who actually reads it. That’s the credibility line in this region, not a nicety.
Your assignment
Build the forward-looking pair for one set of books — your own (recommended: the output is a real model and variance report your team plans from) or the sample company Mizan, the GCC bookkeeping SaaS whose own finance team runs throughout this track. Everything inherits the M1 foundation files and, ideally, the M2 reconciled actuals. Open the folder with your costs file and last month’s close in Claude Desktop, approve each read in the “Ask permissions” prompt, and work in the chat — no terminal needed.
Module 3 deliverable — plan & forecast
Inherits from M1: categorization-rules.md + data-checklist.md +
finance-controls.md
Inherits from M2: the reconciled actuals + the categorized month
1. cashflow-model.md (a 6-month projection you can defend)
- an assumptions block on top: opening balance, the recurring costs
(from categorization-rules.md), the revenue line, the churn/growth
assumption, one-offs by the month they hit
- the 6-month table: opening + inflows − outflows = net; each closing
carried into the next opening — EVERY formula explained in one
plain sentence
- at least ONE scenario toggle: flex a SINGLE assumption (churn +2pt,
hire 3, revenue flat) and report the effect on net and the runway
- spot-check one month by hand against the formulas (the verify gate)
2. budget-variance.md (the variance story leadership acts on)
- budget vs actual per category: planned, actual, variance (AED + %),
buckets aligned to the SAME categories first
- a materiality threshold, stated and applied (e.g. > AED 2,000 or
> 10%) so only the gaps worth a sentence are flagged
- every flagged gap explained as PRICE / VOLUME / TIMING — the cause,
not the delta restated; "cause unclear" where you're guessing
- a narrative under 200 words: total over/under, the gaps worth
explaining, one-off vs ongoing, the line that needs an owner
Bilingual teams: author the Arabic forecast + variance narrative from the
figures — lead with the decision, Western numerals, AED, RTL, same verify
gate. Don't translate the English at the end.The toolkit blocks below give you a fill-in template for each, so you’re filling in structure, not staring at a blank sheet.
How it’s graded — the rubric
This is the part the free playbooks don’t have, and the part that makes the credential mean something. Your two files are scored against five criteria. Each is meets / nearly / not yet — and “nearly” on any one is a revise, not a pass.
Plan & forecast rubric
1. Every formula is explained No line of the model is a black box. Each
formula is stated in one plain sentence; a
stranger could re-run the whole projection.
2. The model is scenario-able A SINGLE assumption flexes — churn, a hire,
flat revenue — and the model responds; the
effect on net and runway is reported, not
guessed, with the carry-forward intact.
3. Variance is a story Every flagged gap is classified PRICE /
VOLUME / TIMING — the cause, not the delta
restated. A guess is marked, never invented.
4. The threshold is set + applied A materiality floor is stated and used, so
the report surfaces the gaps worth acting
on — signal, not every wiggle.
5. The Arabic narrative is The forecast + variance story is authored in
authored Arabic, conventions right (Western numerals,
AED, RTL), same verify gate. (Bilingual.)The discipline is deliberately what a senior controller would demand: a model nobody can explain, a scenario that’s really a fresh guess, or a variance report that restates deltas without a cause fails quietly — in the board meeting, in the hiring decision it mis-sized — so it has to be caught here, against the rubric.
The bar, shown — a worked model answer (Mizan)
You don’t have to guess what “meets” looks like. Here are passing excerpts for the sample company — yours doesn’t need to match these, it needs to clear the same bar. Mizan’s figures are illustrative; your own books will differ.
cashflow-model.md — Mizan (excerpt)
Assumptions (stated first — the model IS these)
Opening cash AED 2,400,000 (from the reconciled May close)
Revenue (recurring) AED 512,000/mo (held flat — conservative, no growth)
Recurring costs (from categorization-rules.md — same buckets every month):
Payroll 312,000 · Hosting/infra 48,000 · Software & tools 26,000 ·
Marketing 64,000 · Office & admin 22,000 → 472,000/mo
Net +40,000/mo
Month 1 — every line explained in one sentence
Opening 2,400,000 = last month's closing (carried forward)
Inflows +512,000 = recurring MRR, held flat in the base case
Outflows −472,000 = the five recurring lines above, one bucket each
Net +40,000 = inflows − outflows
Closing 2,440,000 = opening + net
6-month base case (the cushion GROWS — cash is not the near-term risk)
M1 2,440,000 · M2 2,480,000 · M3 2,520,000 ·
M4 2,560,000 · M5 2,600,000 · M6 2,640,000
→ No month goes below zero. For Mizan the headline isn't runway, it's
the TREND — so we flex the levers that could change it.
Scenario A — churn +2pt (the revenue lever)
Toggle: monthly revenue churn +2pt → revenue erodes ≈ AED 10,000/mo.
Net by month: +30k · +20k · +10k · 0 · −10k · −20k
→ Net flips NEGATIVE at month 4; the cushion peaks (~2,460,000) then
turns down. Cash doesn't run out in the window, but the direction
reverses — churn, not spend, is Mizan's cash lever.
Scenario B — hire 3 analysts (the cost lever)
Toggle: +3 hires × AED 30,000/mo loaded → Payroll 312k → 402k.
Net = 512,000 − 562,000 = −50,000/mo from month 1.
Runway = 2,400,000 ÷ 50,000 = ~48 months (cushion ÷ steady burn)
→ Affordable only if the 3 hires lift revenue by ≥ AED 50,000/mo to
close the gap. The model turns "can we hire?" into a number — and
a number that sizes a hire goes past a second approver (controls).
Spot-check (the verify gate)
Re-added M2 by hand: 2,440,000 + 40,000 = 2,480,000 ✓ the method holds.budget-variance.md — Mizan (May, excerpt)
Threshold: flag any variance > AED 2,000 OR > 10% (whichever is larger).
Category Budget Actual Var % Flag
Payroll 312,000 312,000 0 0% — on plan
Hosting/infra 45,000 48,000 +3,000 +6.7% ● volume
Software&tools 17,000 26,000 +9,000 +53% ● price+volume+timing
Marketing 60,000 64,000 +4,000 +6.7% ● volume
Office & admin 22,000 22,000 0 0% — on plan
─────────────────────────────────────────────────────────
Total 456,000 472,000 +16,000
The flagged gaps, explained as price / volume / timing (the cause, not the delta)
Hosting +3,000 VOLUME — AWS usage up with customer growth. Ongoing,
small, expected; no action.
Software +9,000 the line leadership asked about — decomposed:
+6,000 TIMING Figma annual plan prepaid in May, not
monthly → reverses next month; not an overrun
+2,000 VOLUME one new tool added this month (a real choice)
+1,000 PRICE the analytics tool's silent price hike
(same tool, higher rate) → chase the vendor
Marketing +4,000 VOLUME — one extra campaign approved mid-month. One-off.
The read leadership can act on (< 200 words)
+16,000 over plan for May — but +6,000 is timing (the Figma annual prepay
reverses) and +4,000 is a one-off campaign, so +10,000 is non-recurring. The
STRUCTURAL increase is +6,000: +3,000 of software (a +1,000 price hike + a
+2,000 deliberate new tool) and +3,000 hosting on customer growth. The
underlying run-rate is ~AED 6,000 over plan, and the single line with an
owner is the analytics tool's silent hike — renegotiate or cut it. (Feeds
M4 — vendor spend.)plan-narrative-ar — Mizan (note)
Authored, not translated. The Arabic forecast LEADS with the decision, not
the table: «المسار النقدي يبقى موجبًا، والخطر ليس السيولة بل معدّل التسرّب» —
the cash path stays positive; the risk isn't liquidity, it's churn — because
that's what a Gulf owner reads a forecast to learn. The variance story leads
with the one line that needs a decision (the analytics price hike), not the
+16,000 headline. Numbers Western, currency AED, the variance table RTL with
the columns still aligned. Same verify gate: every formula explained, every
variance traced, cleared by an approver who reads Arabic.What you keep — toolkit additions
These extend the Foundation toolkit you installed in M1 — copy them into the same folder your team works in, replace the Mizan placeholders with your own, and every forecast and every variance starts from structure instead of a blank sheet.
The cashflow-model.md template — the assumptions block, the projection with the “explain every formula” rule baked in, and a scenario row.
# cashflow-model.md — [Company]
## Assumptions (the model IS these — state them first)
Opening cash: [a balance you trust — ideally a reconciled close]
Revenue: [recurring/mo + the growth OR churn assumption]
Recurring costs: [pull from categorization-rules.md — the same buckets]
[line] [amount] · [line] [amount] · ...
One-offs: [a tax bill, an annual renewal — with the month it hits]
## The projection (explain EVERY formula in one plain sentence)
Opening = last month's closing (carried forward)
Inflows = [the revenue line]
Outflows = [the recurring lines above]
Net = inflows − outflows
Closing = opening + net
→ roll forward 6 months; report the first month (if any) below zero = runway.
## Scenario row (flex ONE assumption; SAME model)
[ churn +Xpt | hire N | revenue flat | cut [line] ] → new net/mo, new runway
(Toggle one variable only. A rebuild changes more than you meant — don't.)
## Verify
Spot-check one month by hand against the formulas. Matches → method holds.
A model that sizes a hire or a spend → a second person owns the decision.The budget-variance.md template — the table, the price/volume/timing column, and the threshold that surfaces signal.
# budget-variance.md — [Company] · [month]
## Threshold (state it; identical every month)
Flag any variance > [AED 2,000] OR > [10%], whichever is larger.
## The table (align budget + actuals to the SAME buckets FIRST)
Category | Budget | Actual | Variance (AED) | Variance (%) | Flag
## Every flagged gap → price / volume / timing (the cause, not the delta)
PRICE = same thing, higher rate (a hike — usually ongoing; chase it)
VOLUME = more of it than planned (a choice — was it deliberate?)
TIMING = landed in a different month (reverses next month; no action)
Where you're guessing a cause: "unclear, investigating" — never invented.
## The narrative (< 200 words — what leadership acts on)
Total over/under · the gaps worth explaining (with cause) · one-off vs ongoing
→ the underlying run-rate, and the ONE line that needs an owner.The saved prompts — paste them in the chat with the right files open. (On the optional Power Track these become slash commands; you need none of that to use them today.)
The saved-prompt library (Plan & forecast) — add to the Foundation set
Build the model and explain every formula
"Read recurring-costs.csv and categorization-rules.md. First read my
assumptions back as a list — opening balance, every recurring cost by its
M1 category, the revenue line and the growth/churn assumption, any one-offs
by month — and flag anything missing or ambiguous. Then build month 1 as a
table and explain EACH formula in one plain sentence. Don't extend to 6
months until I've confirmed month 1."
Roll it forward and find the runway
"Extend the model to 6 months, carrying each closing balance into the next
opening. Give me the full table, the closing each month, and the first
month (if any) the balance goes below zero. Show the carry-forward so I can
spot-check that each opening equals last month's closing."
Run a scenario off the SAME model
"Off the same model and assumptions, flex ONE variable: [churn +2pt / hire 3
at AED 30k/mo / revenue flat]. Change only that input, re-run, and tell me
the new net per month and the new runway. Don't rebuild the model or touch
any other assumption — I want the effect of this one lever, traceable."
Explain every variance as price / volume / timing
"Read budget-2026.csv and this month's actuals (same buckets — align them
first and flag any mismatch). Compute variance per category in AED and %,
flag only gaps over AED 2,000 or 10%, and classify EACH flagged gap as
price, volume, or timing. Context I know: [the campaign, the renewal, the
hike]. Where a cause isn't in my context, say 'unclear, investigating' —
don't invent one. Then a narrative under 200 words: one-off vs ongoing, and
the line that needs an owner."
Author the Arabic plan narrative (don't translate)
"From these figures, author the Arabic forecast + variance narrative for an
Arabic-reading owner: lead with the decision (the runway / the gap that
needs action), not the table. Numbers Western, currency AED, the table RTL.
Same verify gate — every formula explained, every variance traced. This is
authored from the numbers, not the English run through translation."What you’ve proven — and what’s next
Clear the rubric and you’ve proven something the free playbooks can’t certify: that you can build a forward-looking model you defend line by line and flex on demand, and turn a wall of red cells into a variance story leadership can act on. That’s the Plan & forecast stage of “Certified Finance & Ops with Claude.”
From here the track turns a plan into the levers that move the cash and the close that puts it in front of the board, each module assessed the same way:
- Module 4 — the cash levers: the two places the cash actually moves — AR collections (the money owed you, aged and chased) and vendor spend (the renewals, the duplicate tools, the silent price hike you just flagged) — turned from a line in the forecast into real dirhams pulled forward and saved.
- Module 5 — the close & the board, then the capstone — one company’s month closed, planned, and presented end to end, graded into the certificate.
First, make what you built reusable. The toolkit additions above — the cashflow-model.md and budget-variance.md templates plus the saved prompts — extend the Foundation toolkit you installed in M1, so every forecast and every variance starts from structure instead of a blank sheet. And because this module lives or dies on a number a person can defend, pair it with the operating guide — the data, controls, and sign-off layer that belongs underneath every model that sizes a decision.