Lab: The 13-week cash forecast

The signature FP&A lab. Below is the data summary Priya's team assembled (in reality: exports from the AR/AP systems — the summary is the realistic input, because assembling it is the part AI can't do for you). Build the forecast in your spreadsheet with AI as decomposition analyst, architect, assumption drafter, and challenger. You own it at the end — the readout has your name on it.

OPENING CASH: $1.62M    CREDIT LINE: $2.0M undrawn (review due Q4)
INFLOWS
  booked AR due wk1-2: $1.94M / wk3-4: $1.71M (collection: 92% on-time,
    6% +1wk, 2% +2wk, trailing 6mo)
  revenue run-rate beyond booked: $2.05M/wk invoiced; DSO 34 days
    (range 31-38, trailing 6mo); wk 38-48 seasonal uplift +18-24%
    (this forecast covers wks 27-39 - uplift enters at wk 38)
  Marlowe Retail = 11% of revenue, pays net-45, historically on time
OUTFLOWS
  carrier settlements: ~$1.31M/wk (tracks revenue w/ 1-wk lag, ~62% of rev)
  payroll: $410K biweekly (lands wks 1,3,5,7,9,11,13)
  fuel: $185K/wk base, indexed (diesel currently +4% vs quarter avg,
    trailing 12mo range -9% to +16%)
  fixed (rent, leases, insurance, debt svc): $290K monthly on the 1st;
    PLUS quarterly insurance $240K in wk 9 (a payroll week); SaaS/overhead
    $55K/wk
  capex committed: $150K wk 5 (trailer refurb, already approved)

1. 1Draft the driver tree with AI (architect prompt: describe the business + paste the summary; ask for the weekly driver structure and which mechanism drives which weeks — known/pattern/assumption). Edit it until you'd defend it: collections off booked AR for weeks 1-4, DSO-lagged run-rate beyond, settlements as %-of-revenue lagged one week, the fixed calendar as given.
2. 2Build the grid in your spreadsheet — 13 weekly columns, inflow/outflow rows from the tree, closing cash per week. The spreadsheet does every sum. Color-code cells known/pattern/assumption. Sanity checks before proceeding: payroll lands 7 times ($2.87M total), and the wk-9 stack (payroll $410K + $240K quarterly insurance — plus monthly fixed on the 1st if your calendar puts it in wk 9; check that assumption and write it down) should make wk 9 visibly the tight week.
3. 3Write the assumptions log with AI drafting, you adopting: DSO 34 (range 31-38), collection curve, diesel +4% held (range −9/+16), settlement ratio 62%, no seasonal uplift until wk 38 (one week of tailwind at the horizon's edge). Every row: value, basis, range, owner (you). One assumption the summary doesn't hand you: opening AR / pre-window collections beyond the booked wk1-4 receipts — you have to state it (e.g. 'run-rate collections begin landing at DSO lag from wk 5') and label it in the log, because your exact trough moves with it. Expect the base-case low point in the tight week (wk 9) and in the low-seven-figure range — quote it as a stated-assumption estimate, not a precise figure, and if yours differs wildly, diagnose which driver did it before reading on; that diagnosis is the skill.
4. 4Run two scenarios with decision thresholds: (a) Marlowe churns at wk 4 (−11% revenue, settlements follow at 62% one week later — note how much of the hit the cost structure absorbs, and whether wk-9 goes negative or just uncomfortable; state the credit-line implication and the 'extend the facility review now' recommendation if warranted); (b) diesel hits +16% (the top of its trailing range) from wk 2 — against your +4% base case that's 12 points on the $185K index base, ≈ +$22.2K/wk; recompute it from your own grid, then say whether it's material or absorbable and defend it. For each: the threshold ('if X by week Y') and the action it triggers.
5. 5Run the challenger pass, then write the one-slide readout: base-case shape in one sentence, the tight week named with its cause, the two watch items with early-warning signals, scenario thresholds, and the naive-baseline comparison you'll score yourself against next cycle (seasonal-naive weekly cash from last year's pattern). Verify every number AI drafted into the narrative against your grid. Sign it.