The number on the shelf and the number in the ledger
Stock is the balance-sheet figure most businesses are least certain about. Cash can be proved against a bank statement. Receivables can be aged and chased. Stock is a physical thing in a building, valued using a method, fed by documents that arrive from dozens of suppliers in dozens of formats — and the gap between what it should be and what it is tends to be discovered once a year, in a count, long after anything can be done about it.
Some of that gap is physical: breakage, wastage, things that walked. But a surprising share of it is documentary, and the documentary part is the only part you can fix at a desk. A unit price that went up in March and was never noticed. A delivery charge added to the goods total by one supplier and shown separately by another. Twelve cases invoiced and eleven received. A credit note that arrived three weeks later and was filed rather than applied.
None of those are dramatic on their own. All of them are invisible if the only thing you record from a supplier invoice is the total, which is what most bookkeeping does — because the total is what you pay, and paying is the urgent part. This page is about the other part.
This is not accounts payable matching
Worth separating early, because the two get confused and the confusion is why the second one rarely gets done. Three-way matching asks: should this invoice be paid? It compares the invoice to the purchase order and to the receipt, and when the three agree, payment is approved. It is a control over money going out, and it is usually well established.
Stock purchase reconciliation asks a different question: what is the stock worth now that it has arrived? An invoice can pass three-way matching perfectly and still leave your stock value wrong — because the quantity was recorded in the wrong unit, because freight was capitalised on one delivery and expensed on the next, or because the price change was correct and simply never reached the valuation.
Both questions are fed by the same document. The first is answered by the total and the approval; the second needs the lines. If you already run three-way matching, this is the layer underneath it — our three-way match page covers the payment-control side, and the two are meant to be read together rather than instead of each other.
Three numbers that should agree
For any purchase there are three records, produced at different moments by different people, and reconciliation is the act of putting them next to each other.
| Record | Says | Produced by | Typical failure |
|---|---|---|---|
| Supplier invoice | What you were charged for | The supplier | Price changed without notice |
| Goods received note | What physically arrived | Your warehouse | Signed without counting |
| Stock ledger | What you believe you hold | Your system | Never updated for the credit note |
The invoice is the only one of the three produced outside your business, which makes it the natural anchor — and also the one most likely to arrive as a PDF nobody has time to key in line by line. That is the bottleneck this removes.
Why totals cannot do this job
An invoice total is a single number that answers a single question: how much do I owe. It cannot answer any stock question, because two invoices with the same total can describe completely different deliveries — a hundred units at four pounds and eighty units at five pounds both come to four hundred, and only one of them matches what is on the shelf.
Line-level extraction produces a row per product: description, quantity, unit of measure, unit price, line total, and any discount or charge stated. That is the granularity at which stock actually exists, and it is the granularity at which every useful question can be asked. What did we pay for this product across the year? When did the price move? Which supplier is cheapest once freight is in?
It also enables a check that totals cannot support: the sum of the lines should equal the invoice total. When it does not, either a line was misread or the invoice contains something the lines do not explain — and both are worth knowing before the figure enters a valuation. Our line item extraction page covers the mechanics.
Quantities, and the unit-of-measure trap
Quantity looks like the simplest field on an invoice and is the one that causes the most silent damage, because the number is meaningless without its unit. Twelve can mean twelve singles, twelve cases of six, or twelve kilograms, and the same supplier will sometimes use more than one convention across their catalogue.
The stated unit is extracted alongside the quantity, so the ambiguity is at least visible rather than assumed. What no tool should do — and this one deliberately does not — is guess the conversion into your own stock unit. A wrong multiplier is the worst kind of error: it produces a plausible number, it applies consistently, and it misstates both quantity and value in the same direction for as long as nobody checks.
The practical answer is a conversion recorded once per product, kept as your own column, and applied deliberately. It is ten minutes of work per supplier and it removes a category of error that otherwise surfaces only in a stock count, where it is indistinguishable from theft.
Price changes you were never told about
Suppliers raise prices. Most do it politely, in a letter nobody circulated, or silently, on the next invoice. Either way the change reaches your business as a slightly different number on a document that is approved because the arithmetic is right — because it is right. The invoice is correct; the assumption underneath your pricing is not.
With a year of invoices extracted at line level, this stops being a matter of noticing. Sort by product and date, and every unit price change is visible with the date it took effect and the invoice it appeared on. What was an unexplained margin drop in a management account becomes a specific conversation with a specific supplier about a specific date.
The same view answers the question that follows immediately: did our selling price move too? A cost increase that was passed on is a non-event. A cost increase absorbed for eight months is a real amount of money, and it is usually discovered by accident.
Freight, handling and the landed cost question
Delivery charges are where stock valuation quietly goes wrong, because different suppliers present them differently and the difference is easy to miss. One includes carriage in the unit price. One shows it as a separate line. One invoices it a week later on its own document, referencing nothing.
Extracted as their own lines, these charges become something you can apply a policy to. Whether they are allocated across the goods as landed cost or expensed as a period cost is an accounting decision with real consequences for both stock value and reported margin — but it is a decision you can only make consistently about a figure you can see. When freight is invisible, the policy is whatever each supplier happened to do.
The inconsistency matters more than the amount. Capitalising carriage on the deliveries where it is itemised and expensing it where it is buried means the same product carries different costs depending on which supplier sent it — which makes cost comparisons between suppliers meaningless in exactly the situation where you most want them.
Short deliveries, over deliveries and split shipments
The gap between invoiced and received quantity is the most direct stock difference there is, and it is also the one most often absorbed without record — because the warehouse signed for what arrived, the office paid what was invoiced, and neither compared the two.
With both numbers in one table the comparison is mechanical: invoice quantity, received quantity, difference. What the difference means is not mechanical, and that distinction matters. A shortfall might be a genuine short delivery worth a credit note, a split shipment where the rest arrives next week, or a counting error at the door. Each has a different resolution and only a person can tell them apart.
Over-deliveries deserve a mention because they are treated as good news and are not. Stock you hold and were not invoiced for is stock with no cost attached, which overstates margin now and produces a surprise invoice later — often in a different period, which is the part that makes it awkward.
Credit notes and returns
A credit note is a purchase in reverse: it should reduce both the quantity held and the value carried. It is also the document most likely to be missed, because it arrives detached from the invoice it corrects, sometimes weeks later, and it never demands attention the way an invoice does — nobody is chasing you for money.
Extracted with the same fields and negative quantities, credits sit in the same dataset as the invoices and net correctly when totalled by product. That is enough to catch the common failure: the credit that was received, filed, and never applied, so the stock system still shows goods that went back to the supplier months ago.
Worth checking specifically at period end. A missing credit overstates both stock and cost, and unlike most stock errors it has a document sitting in a folder that proves it — which makes it the cheapest correction available.
How it works
1 — Upload supplier invoices
Any supplier, any layout, PDF or scan. A month or a year at a time; nothing needs sorting first.
2 — Lines are read, not just totals
Description, quantity, unit, unit price and line total per row — recognised by meaning, so new suppliers need no setup.
3 — Checked and flagged
Line sums are compared to the document total, and low-confidence fields are marked for a short review.
4 — Export and compare
Excel or CSV with the document recorded per row — ready to sit next to your receipts and stock ledger.
When the physical count disagrees
A count difference is a total, and totals do not explain themselves. The question that follows — where did it come from — has four possible sources: purchases, sales, wastage, and loss. Most businesses investigate the last two, because those are the interesting ones, and assume the first is fine.
It frequently is not. Reconciling purchases first is worth doing precisely because it is the side nobody suspects, and because it is the only one that can be settled with documents rather than with judgement. If the buying side accounts for a third of the difference, that is a third you no longer need to explain any other way.
It also changes the tone of the conversation. Telling a warehouse manager that stock is short by an unexplained amount invites defensiveness; telling them that the documented purchases account for most of it and here is the residual is a different meeting entirely.
What it does to cost of goods sold
Every stock valuation error has a matching error in cost of sales, and therefore in gross margin. That is the part which makes this more than a housekeeping exercise: an overstated stock value understates cost of sales and flatters margin, and the correction arrives later as a write-down nobody budgeted for.
The errors described on this page all push in that direction. Missed credit notes overstate stock. Unrecorded price rises understate the cost of what was sold. Freight inconsistently applied makes product-level margin unreliable in both directions at once. Individually small, collectively enough to make a monthly gross margin figure something people quietly stop trusting.
Reconciling purchases does not fix valuation policy or physical control. What it does is remove the documentary noise, so the margin you report reflects decisions you made rather than documents you did not read.
Imported stock and foreign currency
Stock bought abroad adds two layers. The invoice is in the supplier's currency, so the cost depends on the rate applied and when. And landed cost grows: freight, duty, customs clearance and handling arrive as separate documents from separate parties, often weeks apart, all relating to goods that are already on the shelf.
Currency is captured as its own field, so nothing is added across currencies by accident and the conversion stays an explicit step at a rate you choose. The related documents extract the same way as the invoice itself, which at least gets them into one place — but connecting a customs entry to the shipment it belongs to is a matching job that needs the reference, and references are exactly what these documents are worst at carrying.
The practical advice for importers is to keep a shipment identifier of your own on every related document as it arrives. It costs seconds and it is the only thing that makes landed cost reconstructable three months later.
How often this is worth doing
Monthly, and the reason is not thoroughness — it is that the evidence decays. A short delivery queried in the same month can be checked against a delivery note someone still remembers signing. The same query eleven months later is a negotiation with a supplier who has no reason to agree with you.
| When | What | Why then |
|---|---|---|
| As invoices arrive | Extract lines, file with the receipt | The delivery is still remembered |
| Monthly | Invoiced vs received by product | Credits can still be claimed |
| Monthly | Unit price changes since last month | Selling prices can still be adjusted |
| Quarterly | Freight treatment consistency | Before it distorts a reported margin |
| Before a count | Purchases reconciled and closed | So the count difference means something |
The last row is the one that changes a stock count from an ordeal into a measurement. A count run against unreconciled purchases produces a number nobody can act on, which is why so many counts end with an adjustment and a shrug.
Six mistakes
Recording only invoice totals. It is enough to pay and useless for stock. Everything on this page depends on the lines existing somewhere.
Assuming the unit of measure. Cases versus singles is the single most damaging silent error, because the result looks entirely normal.
Treating freight differently by supplier. Not because either treatment is wrong, but because inconsistency makes product costs incomparable.
Filing credit notes instead of applying them. They arrive detached from what they correct and nobody chases you for them, so they are missed by default rather than by accident.
Signing for deliveries without counting. A signature that means «a lorry came» rather than «this arrived» removes the second of the three records entirely.
Reconciling once a year. By then the credits are unclaimable and the price changes are history rather than a decision.
Who this is for
Wholesalers and distributors
High volume, thin margins, and cost accuracy that decides profitability.
Retail and e-commerce
Many suppliers, many formats, stock value that moves weekly.
Manufacturers
Raw material cost feeding directly into what a finished unit is worth.
Accountants with stock clients
A defensible purchase side before anyone argues about the count.
A full monthly cycle for a distributor is worked through on the wholesale stock control page.
What this is not
It is not a stock system. It does not hold quantities, run valuations or manage locations — your inventory software does that, and this produces data it can import.
It does not decide your valuation method. FIFO, weighted average, standard costing and the treatment of landed cost are accounting policies with real consequences, and a tool that quietly picked one would be making a decision it has no business making.
And it does not count anything. A physical count is a physical activity; what this changes is how much of the difference you can explain from documents before you start arguing about the rest.
Processing runs on EU-hosted infrastructure over TLS, the original file is deleted immediately after extraction, and documents are never used to train AI models — the security page has the detail.
Try it on one supplier
Extract a real supplier invoice free — no registration — and look at the lines before committing a month to it.