AI for Financial Reporting in Consumer Goods: Gross to Net, Trade Accruals and a Close That Holds
The number a consumer goods business reports is not the number it invoiced. Between the list price on the order and the cash the business keeps sit trade investment, retrospective rebates, listing and promotional fees, penalties and deductions that arrive months late. The close is an exercise in estimating that gap and then defending the estimate.
That is a different problem from producing a ledger. The ledger is mostly right. What takes the days is the judgement on top of it: which accrual is still needed, which is stale, which claim belongs to which promotion, and what the business earned on a customer once trade spend is allocated honestly.
One boundary governs everything below. A properly built agent does not set an accounting judgement, release a provision or post an unreviewed journal. It reconciles, drafts, evidences and flags. Every estimate carries a named accountant.
Gross to net is a bridge with many owners
The bridge from list to invoiced to net revenue is the most important schedule in the business and rarely the best maintained. Invoice discounts sit with the account manager, retrospective rebates with commercial finance, listing fees with the category team, deductions with whoever is closest to the remittance advice.
An agent builds the first draft from the ledger and the trade agreements instead, mapping every reduction from gross back to the customer, the product and the agreement that authorised it. What it cannot map is the useful output: the part of net revenue that no agreement explains.
Trade promotion accruals and the plan that was not the promotion
A promotion is accrued when it is planned and settled when the customer claims. Between those events it changes. The mechanic moves, the period extends, volume lands well above or below plan, and a second activity runs over the top of it in a region nobody told finance about.
An agent reconciles the accrual against what shipped, reading the promotion record, the volume in the promoted period, the price invoiced and the claim evidence so far. An accrual with no activity behind it is money the business may already have earned. Activity with no accrual is a claim that will land in a period not expecting it.
Settlement lag and the accruals that will not close
Claims arrive long after the activity ended, so every close carries a tail of prior period accruals waiting on settlement. An agent ages that tail by promotion, by customer and by the period the spend belongs to, separating balances still supported by expected claims from balances never touched since they were raised.
The deduction reserve and what is genuinely open at close
Customers deduct for promotions, shortages, late delivery, damaged stock and reasons that resolve into a reference number and nothing else. The reserve held against that population is a judgement about how much will eventually be conceded.
The accounting question is not how to work the queue. It is what the open population supports. An agent classifies open items by type, customer, age and the outcome of similar items already settled, producing the reserve with the population attached rather than a rate applied to a total. Where the settlement pattern for a type has shifted, it says so.
Release discipline and the entry nobody wants to sign
Releasing a stale accrual feels like inventing profit, so stale accruals sit. An agent lists every balance with no supporting claim and no settlement movement for longer than policy allows, with the evidence for each, at the same point in every close. It cannot make the decision, only stop it being quietly avoidable.
Retrospective rebates and growth incentives that resolve at year end
Volume rebates, growth incentives and tiered discounts are earned progressively and confirmed only when the measurement period ends. Until then the accrual rests on a forecast of where the customer lands against the threshold, and a customer sitting just below a tier boundary is the most expensive uncertainty in the account.
An agent tracks each agreement against volume to date, projects the landing position on the run rate, and reports the customers whose projected tier has moved since the last close. It also flags terms that are ambiguous about measurement: which products count, whether returns are deducted, whether the period is calendar or trading.
Listing fees, marketing contributions and where the cost belongs
A payment for shelf space, a leaflet slot, a display or a joint marketing activity has to be tested before it is booked. The accounting standard on revenue treats consideration paid to a customer as a reduction of revenue unless the business receives a distinct good or service at fair value. Many such payments fail that test. Some pass.
An agent reads the agreement, extracts what the customer committed to deliver, and compares the treatment applied against similar agreements in prior periods. Where a payment sits in marketing cost, it assembles the evidence for a distinct service or reports that none exists.
Customer and channel profitability once trade spend is allocated honestly
Most consumer goods businesses know gross margin by product and guess at profit by customer. A discount channel that looks thin on gross margin can be the healthiest account in the book, and a grocery multiple can be the opposite.
An agent allocates spend to the customer and product it was committed against rather than into a pool, using the promotion records and the agreements, and shows the allocation line by line. The basis stays a finance decision, but an explicit and challengeable one.
Freight, warehousing and the cost of serving
Full pallets into a central depot and mixed cases into a store do not cost the same to serve, and a single rate per case hides that entirely. An agent builds the allocation from shipment records, the drop profile and the actual carrier invoices, showing what each customer costs to serve given how it genuinely orders.
Contract manufacturing and the cost that arrives late
Where production sits with a third party, cost reporting comes back on somebody else timetable and format. Usage statements, yield losses and waste arrive later than the close would like. Finance accrues, trues up, then explains the true up.
An agent reconciles production declared by the partner against materials issued and finished goods received, prices the variance before the statement lands, and compares what was charged against the rates in the supply agreement. That belongs inside the close, because it changes the value of stock on the balance sheet and not only the payable.
Stock with a date on it, and the write off that follows
Short shelf life stock loses value on a clock. Remaining life determines whether a line can be sold normally, sold into a secondary channel, marked down or destroyed, and policy usually expresses that as bands. A line with short remaining life and a confirmed promotion behind it is not the same risk as the identical line with nothing behind it.
An agent prepares that judgement rather than making it, assembling remaining life by batch, the demand actually committed, the promotional activity planned and the outcome of recent similar decisions. The provision is then argued on evidence rather than defended on policy.
Volume, price and mix, and the commentary commercial will dispute
Every pack carries a bridge explaining the movement in revenue and margin, and every such bridge is contested, because mix can be made to carry almost anything. Whether a change is priced or mixed depends on the level the bridge is calculated at, and moving between levels changes the story without changing the result.
An agent computes the bridge at a defined level, holds that definition constant between periods, and drafts the explanation with the customers and products driving each component named and the transactions cited.
Consolidating markets that sell on different trade terms
The same brand is sold on invoice discount in one country, on retrospective rebate in another and through a distributor in a third, so the same commercial reality lands in different lines of the group result. Consolidated net revenue is comparable. The bridge above it frequently is not.
An agent maps each market trade spend categories to the group definitions, reports what does not map cleanly, and holds that mapping as a maintained artefact rather than institutional memory. The working is stored, versioned and reproducible.
The audit question that arrives after filing
The questions come months later and always in the same form. How was this accrual calculated. Show the agreement behind this rebate. Assembling that evidence afterwards costs weeks, and the person who made the judgement has often moved on.
An agent working inside the close records evidence as a by product: the calculation, its inputs, the source extract and when it was taken, the documents relied on, the approver and the date. Where a link between a reported figure and its support does not exist in the systems, it has to say so rather than construct a plausible path. Inferred evidence is worse than none, because it will be believed.
Where AI fails here
They are confident about promotions that were never recorded properly. Where a mechanic was agreed by email and executed on trust, an agent reconciling claims to plans produces a clean looking answer built on a record that does not describe what happened.
They cannot judge whether a customer will claim. Settlement history is a guide until a buyer changes or a dispute becomes strategic.
They will reduce judgement to a rate if you let them. The moment the output is accepted without the population being examined, the estimate has become a habit with arithmetic on top.
They do not know what the commercial team has committed verbally. Trade investment exists as an understanding long before it exists as a document, and no system reads an understanding.
They degrade quietly when terms change. A new agreement structure or a restructured incentive, and the logic keeps producing output that looks much like last month. Somebody has to own the mappings and review them.
One line of the bridge, at one close
Do not start with the whole bridge. Start with one line of it, at one close, for one group of customers. Trade promotion accruals for your largest few accounts is a sound choice: the population is finite and you already know roughly what the answer should be.
Run it alongside the close you are running anyway, and pay attention to the disagreements rather than the matches. Then ask the harder question: how long did the evidence pack take to assemble, and would it have answered the audit question without a person rebuilding the working from memory.
Adopt AI Partners