AR Collections Forecast

The baseline

Forward billing forecast

PAPM recharge data only captures shared-services billing, not every AR line. Where a customer's matched PAPM figure looks like a small slice of their known current billing pace, their own "Future Due" balance is used as the monthly forecast instead — see Methodology.

Collections behaviour by segment

Days Sales Outstanding (DSO) on each customer's "active" receivables (everything except debt 720+ days old), calculated per customer where the book is material and PAPM billing is credible, and blended by segment otherwise — the direct, customer-specific read of how fast each relationship actually pays against 60-day terms.

Scenario forecast, Aug–Dec 2026

Each customer's receivables are rolled forward month by month on 60-day terms: the trailing two months of billing stay "not yet due," older un-collected balance rolls into overdue, and it runs off at that customer's (or segment's) own collection speed. Three paths, same mechanics, different assumptions.

Sensitivity: what it takes to hit 20%

Dec-26 overdue % under combinations of (a) how much faster collections run versus today, and (b) how fast the book keeps growing. Billing growth alone is not a fix — it dilutes the ratio without changing behaviour; boxed cell shows the base case.

Top customers

The 25 largest receivables balances — 71.7% of total AR sits in the top 20 — with their own DSO and billing basis where individually modelled.

Methodology & assumptions

What "overdue" means

Overdue % = (Total AR − "Future Due" bucket) / Total AR, evaluated at each month's close. This matches the collections team's own "Landing Overdue %" convention in the source workbook and ties out to 33.0% as of 28 Jul 2026.

Forward billing

PAPM's ICSB Flow Map lists 32,694 initiator→receiver recharge lines; 12,146 are explicitly tagged /MONTH and 245 /YEAR (converted ÷12) — these are treated as recurring. ~20,300 untagged lines (recruitment fees, one-time project/setup charges, ad-hoc services) are excluded from the recurring forecast since they don't repeat predictably. Pluto Flow Map (external, non-Oracle customers) has real invoice dates Jan–May 2026; each customer's forecast is their 5-month average. Customer names were matched between the AR file and PAPM by normalised exact match, then a conservative fuzzy match (guarding against brand-suffixed sub-accounts matching a parent's total). Where PAPM billing for a matched customer is under half of their current Future-Due balance, their Future-Due balance is used instead, since that PAPM line evidently isn't capturing their whole recurring bill.

Collections behaviour model

Each customer's book is split into an Active pool (Future Due through 361–720 days) and a Legacy pool (720+ days, treated as slow-moving/largely stuck debt, not normal payment behaviour). Active-pool DSO = Active AR ÷ monthly billing × 30, giving a monthly collection rate of 30/DSO. Customers with AR ≥ $250k and credible billing get their own rate; everyone else uses their segment's $-weighted blend. Collections terms are modelled uniformly at 60 days: each forward month, the trailing two months of gross billing are treated as "not yet due"; anything older in the Active pool is overdue and decays at that customer's collection rate; Legacy debt decays at a slow 2%/month baseline recovery rate.

Scenarios

Realistic — today's collection speed and billing pace, unchanged. The book naturally re-weights toward each segment's typical behaviour as older, non-representative balances roll off.
Downside — collection speed 10% slower, billing 2%/month growth, a seasonal summer slowdown in Aug–Sep.
Upside — collection speed 15% faster with an added 2pp/month legacy-debt recovery, flat billing, and a seasonal year-end collections push in Nov–Dec.

Known limitations

Only a single AR snapshot was available (no true multi-month history), so customer collection speed is inferred from the current aging mix and a DSO turnover model rather than observed payment timing — a modelling proxy, not measured behaviour.
PAPM billing coverage is incomplete at the individual-customer level (covers ~47% of AR $ directly); the remainder falls back to a Future-Due-based proxy.
76 small/parent-level customer name variants could not be confidently matched between the two files and were left out of PAPM matching (their AR is still included in the baseline and forecast via the Future-Due proxy).
The model doesn't distinguish credit notes/disputes from genuine slow-pay; large negative balances (credit positions) are floored at zero when rolled forward.