Hong Kong Restaurant Chain Supplier Invoice Extraction by Outlet

A Hong Kong F&B chain's consolidated AP run never tells you which kitchen is leaking margin. The number that matters lives one level down — prime cost per outlet, which only reads true when every supplier-invoice line is coded to the outlet that received the delivery and to the GL cost type that classifies it. Miss either axis and the per-outlet P&L is a fiction; the founder asks why the Causeway Bay store is dragging the group margin and finance has nothing to show beyond a consolidated food cost that looks fine.

That coding discipline is what Hong Kong restaurant chain supplier invoice extraction has to deliver. HK F&B chain operators process supplier invoices from a heterogeneous distributor mix — Angliss, DCH, Foodgears, Wismettac, beverage importers across beer, wine and coffee, fresh-produce wholesalers from the Vegetable Marketing Organisation down to the wet-market layer — and every line on every one of those invoices has to land against two axes: the outlet that received the goods, and the GL cost type that classifies them (food, beverage, packaging, supplies, services). On Xero, the canonical pattern uses two Tracking Categories — Outlet for location and either Department or Cost Type for GL classification. Inland Revenue Ordinance Section 51C requires the underlying records to be retained for at least seven years.

Sector scale gives the discipline its weight. The Census and Statistics Department puts HK$109.6 billion in 2025 restaurants sector receipts on the table, with non-Chinese restaurants up 4.0% in value over the year and Chinese restaurants down 2.9% — a market where the F&B percent on each individual outlet is the lever finance pulls and the consolidated number is the audience report.

The reader this piece is written for is the in-house finance manager, the outsourced HK bookkeeper handling multi-outlet F&B AP, or the founder still wearing the finance hat at a 2-to-30 outlet group. The framing assumes you already run this workflow and are looking for the article that names what you actually deal with.

---

The HK Supplier Mix and What Their Invoices Actually Look Like

Walk an outlet's weekly invoice stack and the supplier set is genuinely heterogeneous, in a way that US-shaped restaurant accounting content never names. Cluster it by category and the document patterns each cluster produces become the reason extraction routes succeed or fail.

Premium foodservice distributors. Angliss Hong Kong Food Service is the dominant name on the international side — meats, wines, spices, dairy, and dry goods at scale. DCH Food Trading, the food arm of Dah Chong Hong, runs meat, seafood, produce, and grocery sourced from thirty-plus countries. Foodgears Industrial covers a similar breadth across meat, dairy, beverage, grocery, and frozen. Wismettac Nippon Foods is the Asian-foods and frozen specialist most groups rely on for Japanese and pan-Asian inventory. Chef's Garden runs fresh, frozen, and dried produce. ETAK International is the long-running food importer most groups have at least one account with, and Eugina Limited adds broad wholesale coverage including Macau and Mainland reseller relationships. These are the names on the lion's share of every chain's weekly cost-of-goods spend.

Local fresh produce. The Vegetable Marketing Organisation sits at the regulated end; below it are the wet-market wholesalers each outlet's chef knows by name, and the local poultry and pork distributors who deliver several times a week.

Beverages. Beer comes from Asahi HK, San Miguel HK, Kirin HK, and the HK Brewing alliances on the local-craft side. Wine and spirits split across Watson's Wine, ASC Fine Wines, Altaya Wines, and Maxim's Wines. Coffee for cafe groups runs through Coffee Academics, Cupping Room wholesale, Pacific Coffee wholesale, and a handful of independent local roasters. Soft drinks come from Swire Coca-Cola HK and A.S. Watson.

Dry goods, packaging, and chemicals. Disposables suppliers, takeaway packaging vendors, and chemical specialists — Diversey and Ecolab HK at the cleaning end — show up monthly rather than weekly but each carries its own invoice format.

Specialty service contractors. Coffee-machine service, fryer-oil management, grease-trap pumping, knife-sharpening, and uniform laundry each invoice on their own cycle. None of this is glamorous; all of it has to be coded by outlet.

The bilingual content patterns track the supplier type. International foodservice arms — Angliss, Wismettac — typically issue English-primary headers with the Chinese supplier name (商業登記號碼 alongside the BRN) but English item descriptions. HK-headquartered distributors, DCH and Foodgears among them, run bilingual stacked formats with Chinese item names and English descriptions side-by-side on each line. Local fresh-produce wholesalers issue Chinese-primary documents with handwritten Chinese item names. Wet-market suppliers hand over carbon-paper books with handwritten Chinese only — no English at all, no machine-readable text, sometimes no clear line separation. Coffee, tea, and beverage importers split: the international brands run English-primary; HK-local roasters run bilingual.

For line-item extraction, the implication is concrete. Bilingual stacked formats from DCH and Foodgears are where parsing earns its keep — both scripts on the same line carry information the bookkeeper needs and either a route that drops the Chinese description loses the inventory-tracking signal the chef relies on, or the route that drops the English loses the GL coder's reference. Handwritten wet-market books are where every extraction route degrades; manual entry is the honest answer at the volumes those suppliers represent. The English-primary international invoices are the easy half — header capture is reasonable on any modern OCR and line-item depth depends only on how the route handles tabular structure. Routes that handle the bilingual stacked format well extract bilingual Chinese-English supplier invoices into Excel without losing Traditional Chinese fields and let the chain finance function keep both reading audiences served.

A note on cross-border sourcing, kept brief because the depth belongs elsewhere. Many HK F&B chains source produce, frozen items, and packaging from Mainland for cost reasons, which means fapiao 發票 arrive alongside HK commercial invoices — structurally different documents with the Unified Social Credit Code instead of a BRN, CNY instead of HKD, and Mainland province or city addresses on the supplier line. They sit under the same IRD retention framework even though the field set differs.

---

Two-Axis Allocation — Outlet by Cost Type, Not Just Department

Every supplier-invoice line in a HK F&B chain has to land on two genuine dimensions: the physical outlet that received the goods, and the GL cost classification that tells the books what the spend was for — food, beverage, packaging, supplies, services. Hold both axes and per-outlet management accounts work; collapse to one and they don't.

The canonical example is the weekly Angliss invoice that covers three outlets in a single delivery run. The first split is by outlet: how much of the meat, dairy, and dry goods landed at the Sheung Wan store, how much at the Causeway Bay store, how much at the Kwun Tong store. The second split happens inside each outlet's portion: the meat and the dairy roll up to food cost, the bottled mixers roll up to beverage cost, the takeaway containers roll up to packaging. Both splits feed both reports — the per-outlet P&L the founder reads monthly, and the consolidated category-spend reporting that tracks what the group is paying across all suppliers for, say, beverages.

US restaurant accounting content collapses this into one axis and calls it "department". A single department code per line — "kitchen" — can't simultaneously carry the location and the GL classification. The chain that runs a single-axis architecture either knows where the spend happened or what it was for, never both at once, and finance has to rebuild the missing axis from invoice metadata at month-end. That rebuild is where errors compound. The discipline scales to small chains too: a cafe group with two outlets and five cost types still produces ten distinct P&L cells from the same invoice line, and a five-outlet bubble-tea chain produces twenty-five. Dim sum chains, izakaya groups, and the F&B arms of small hotel groups all run the same pattern at whatever cardinality their footprint demands — though hotel finance functions typically swap the cost-type axis for USALI 12th edition departmental coding across Rooms, F&B, OOD, and EWW rather than the food-beverage-packaging-supplies-services split a pure restaurant chain runs.

The structural signals on the document itself drive both splits. The delivery address — usually printed on the invoice header or on the line-item block where the supplier's logistics system has stamped it — is the signal for outlet allocation. The line-item content (item description, sometimes a product code or SKU when the supplier provides one) is the signal for cost-type allocation: bottled water and craft beer are beverage; Diamond Crystal salt and AA flour are food; clamshell containers and serviette dispensers are packaging. The cleaner those signals are on the source document, the cleaner any downstream coding becomes. The deeper mechanic of how to split a single foodservice distributor invoice across multiple restaurant locations has its own workflow walk-through; the point here is that the two-axis discipline is what makes per-outlet GL coding work for cafe chain operators in the first place — without both axes captured at the line, the management accounts can't carry the analysis the founder is asking for.

---

Tracking Category Architecture for HK F&B Chains on Xero

Xero gives a chain finance function a fixed envelope to work inside: two active Tracking Categories per organisation, with up to 100 options per category. That envelope is what the two-axis allocation has to fit, and the canonical pattern for HK F&B chains uses it directly — Category 1 for Outlet (one option per restaurant or cafe location), Category 2 for either Department (kitchen, bar, floor, catering) or Cost Type (food, beverage, packaging, supplies, services).

The Department-versus-Cost-Type choice tracks how the chain runs management reporting. Operations-led groups, where the GM and head chef drive the monthly review, often choose Department because the conversation is about how each part of the operation is performing. Finance-led groups, where the controller drives the review, often choose Cost Type because the conversation is about where the spend is going. Both work. The decision has to be made once and held; switching mid-year creates a Tracking Category history split that nothing in Xero reconciles cleanly.

The 100-option ceiling on Category 1 caps practical chain size at 100 outlets per Xero org. For almost every HK SME chain in the brief's target band (2–30 outlets) that's not a constraint. For the larger boutique groups, the dim sum chains, and the F&B arms of hotel groups that run past 100 outlets, the working pattern is one Xero org per region or per brand, consolidated in spreadsheet at month-end. The multi-org pattern adds bookkeeping overhead but it is the only way to keep per-outlet coding clean inside Xero's structural envelope.

The sales-side mirror is what makes the architecture useful for per-outlet P&L rather than just per-outlet AP. Eats365 — the HK-built POS that dominates the chain F&B segment — integrates with Xero using the same Tracking Categories on the revenue side, so per-outlet sales and per-outlet supplier-invoice cost land against the same coding. Per-outlet P&L falls out of the standard Xero report once both feeds are running. Hong Kong restaurant supplier invoice consolidation Eats365 Xero is the published integration territory; the architectural fact for AP is that the cost-side Tracking Category set has to mirror whatever the POS side sets up. Chains running Affinity POS, Square HK, or Storehub face the same logic with different integration mechanics — the principle (mirror the categories across cost and revenue) holds across the stack.

The multi-site coding discipline is broader than F&B. Chains running this Tracking Category pattern for supplier invoices face essentially the same per-site coding workflow on utility bills and other recurring multi-site overhead — the same Outlet category accepts CLP electricity invoices, Towngas billings, and WSD water charges as cleanly as it accepts an Angliss food delivery, and the sibling multi-site allocation pattern for CLP, Towngas, and WSD bills walks the utilities side of the same architecture.

---

The Duplicate-Payment Trap in Chain Operations

Chain AP creates a structural opening for duplicate payment that single-site AP doesn't. The same supplier invoice can land at head office through multiple channels: paper from the delivery driver passed to the outlet GM and walked to the office on the next supply run; a PDF attached to the supplier's emailed dispatch confirmation; a JPG snapped on the outlet GM's phone and sent to the bookkeeper on WhatsApp because the paper copy got lost; a download from the supplier portal three days later when the AP clerk does the weekly portal sweep; and sometimes a copy on the supplier's posted month-end statement. Any pair of those channels can deliver the same document to the AP record on different days — and an AP clerk who doesn't catch it pays it twice.

The cause is structural, not a discipline failure on any one person's part. Invoice receipt is distributed across outlets by design — the goods land at the outlet, so the document does too — but payment is centralised at head office because that's where the bank authority and the bookkeeper sit. The handoff between outlet and head office is where duplicates are born, and the more channels the chain tolerates for that handoff, the more duplicates surface.

Three controls hold the line.

Invoice number as a hard uniqueness constraint at the AP record. Supplier name plus invoice number plus invoice date is the natural key; duplicates fail the constraint at entry rather than slipping through to payment. Whether that constraint is enforced by Xero's bill duplicate-detection, by a manual check the bookkeeper runs at posting, or by validation in a pre-accounting layer is a tooling choice — what matters is that the constraint runs at every entry point, including the supplier-portal download route most teams forget to police.

Supplier statement reconciliation at month-end. The 月結單 from each major distributor is the safety net under the entry-time control. Match every invoice on the statement to a posted bill before approving the statement-driven payment run; the unmatched invoice on the statement is either missing from the books (post it before paying) or already paid through another route (don't pay it again). This is what supplier statement reconciliation by location is for.

One designated route per outlet for invoice intake. Even if the channel mix varies — paper at the door, WhatsApp from the outlet GM, supplier portal at head office — each outlet's invoices funnel into a single inbox, not to multiple bookkeepers. A central inbox plus a clear ownership rule for the channel sweep means the duplicate is caught at entry, not at the bank reconciliation.

The cleanest of the three is supplier-statement reconciliation when the statement carries delivery addresses. Matching each line on the statement to the outlet that actually received the goods catches duplicate-payment risk and outlet mis-allocation in one pass — which is why month-end reconciliation only works when the AP records carry both the supplier invoice number and the receiving outlet, every time.

---

Rolling Up Prime Cost Per Outlet

Prime cost is food cost plus beverage cost plus labour cost as a percentage of revenue, and the operating benchmark HK F&B chain finance functions report against is under 65% at the outlet level. A chain running at 62% consolidated can still have an outlet at 73% — and that outlet is the one losing the group its monthly margin while the consolidated number reads fine.

The roll-up at outlet level uses two upstream feeds that have to land against the same outlet code. Food cost and beverage cost roll up from the AP side — the supplier-invoice extraction this article walks through, coded by outlet and by cost type. Labour cost rolls up from the POS time-attendance feed and the payroll output, with statutory MPF contributions (強積金) captured in the payroll line. Divide the sum by the outlet's revenue from the POS sales feed and the F&B percent for that outlet falls out — provided every input was outlet-coded the same way.

The dependency on cost-side coding is total. Miss the outlet allocation on a single Wismettac invoice that covered three outlets — credit it all to one outlet's food cost — and the F&B percent on each of those three outlets is wrong by exactly the misallocated amount. The outlet that received nothing reads cleaner than it actually is. The outlet that received the full credit reads worse. The consolidated number is unaffected and looks fine, which is why per-outlet prime cost only carries a useful signal when the underlying coding holds.

Supplier-invoice extraction does not provide the labour input. The MPF remittance statements that close out the labour-cost side of prime cost are sibling extraction work — the same per-outlet discipline applied to a different document type — and the practitioner walk-through of how to extract MPF remittance statements for the labour-cost side of prime cost covers that side of the calculation. Treating MPF as a separate workflow keeps the AP and payroll feeds independent in the books, which is also how the IRD expects to see them.

The same coding architecture supports more than the headline calculation once the per-outlet, per-cost-type AP data exists. Beverage gross margin per outlet, food-cost variance week-on-week against a rolling baseline, supplier price-creep tracking against the same SKU across months — all of them fall out of the structured AP feed without rework. Hong Kong F&B chain prime cost calculation per outlet is the headline; the line-item-level analyses are the dividend on the same upstream coding.

---

IRD Section 51C Retention as the Compliance Overlay

Section 51C of the Inland Revenue Ordinance (Cap. 112) requires every person carrying on a trade, profession, or business in Hong Kong to keep sufficient records of income and expenditure to enable the assessable profits to be readily ascertained, and to retain those records for at least seven years from the end of the transaction year. Supplier invoices are core records under this rule. Failure to comply without reasonable excuse carries a fine up to HK$100,000.

The operational pairing with extraction is direct: extract once, structure once, store the source digital. The structured spreadsheet (or the posted Xero bill, in the workflow this article walks) feeds the books and the management reporting. The source PDF or image is archived alongside, indexed by supplier, invoice number, and outlet, and held for the seven-year retention window. Both layers stay live for the same period because the IRD does not accept the structured output as a substitute for the source document — the records have to support reconstruction of the transaction, which in practice means the source artefact stays available to the auditor for the full window.

This is the workflow that satisfies Section 51C without drowning the chain in paper or unindexed Dropbox folders. The alternative — paper archives at head office, scanned and dumped into shared drives by date with no index — meets the letter of the rule but fails the practical test: when the IRD asks for the Foodgears invoices for FY2022, the bookkeeper has hours of unindexed PDFs to wade through, not minutes of indexed extraction records. The IRD's own guidance, set out in its information pamphlet "A Guide to Keeping Business Records", describes what a small business needs to retain and the form acceptable records can take; chains running structured extraction with source-archive retention sit comfortably inside that guidance. The deeper operational walk-through of the Section 51C record-keeping requirements for Hong Kong businesses is the compliance-side companion to this article.

The same supplier invoices feed adjacent compliance regimes. Under the Public Health and Municipal Services Ordinance (Cap. 132) and the Food Safety Ordinance (Cap. 612), licensed food premises have to retain records that support food traceability — the supplier name, the date of receipt, and what was received are common fields the FEHD inspector wants to see. Different rule set, different retention period, same source documents. The chain that builds the AP retention workflow correctly gets the food-traceability layer largely as a by-product, and the dedicated walk-through of FEHD food traceability records the same supplier invoices feed into covers what FEHD specifically expects beyond what IRD requires.

---

Three Posting Routes — From Manual Entry to Pre-Accounting AI Extraction

A HK F&B chain finance function has three concrete routes for getting supplier-invoice data into Xero, QuickBooks Online HK, or MYOB. Each route handles a different volume band and produces different quality on the bilingual line-item parsing the chain mix demands. Picking the right route at the right scale is the practical decision the in-house finance manager or the outsourced bookkeeping firm has to make.

Route 1: manual entry into Xero, QBO HK, or MYOB. The bookkeeper or AP clerk opens each invoice, types the header data and each line, and assigns the Tracking Category values for outlet and cost type. The route has zero tooling overhead and full human judgment on coding — the bookkeeper sees the document and codes it directly, with bilingual content handled by whoever is reading the invoice. The trade-off is volume. Manual entry is sustainable up to roughly 80 invoices per week before quality drops; above that, line-item depth gets compressed, Tracking Category values get defaulted, and bilingual handling becomes a function of the bookkeeper's Cantonese fluency on the day. Outlet allocation errors compound silently until the per-outlet P&L looks wrong at month-end and finance has to walk back which Wismettac invoice was credited to which kitchen.

Route 2: built-in OCR via Hubdoc, QuickBooks Online Receipt Capture, or MYOB In Tray. Header-level data — vendor name, invoice number, invoice date, total — captures reasonably across most invoice formats. Line-item depth is shallow on most HK supplier invoices because the parsers are built primarily on US and UK invoice patterns. Bilingual handling on the DCH and Foodgears stacked formats is weak — Chinese item names are typically dropped, mangled, or transliterated rather than retained alongside the English description. Tracking Category assignment is still manual; the OCR captures the invoice but the bookkeeper still has to code the outlet and cost type at the bill-review step. Useful as a digital filing layer that gets the source PDF attached to the Xero bill record; not a workflow that produces clean per-outlet, per-cost-type AP data on its own.

Route 3: pre-accounting AI extraction with full line-item parsing, bilingual handling, and Tracking Category coding in the output. Every line on every invoice is extracted into a structured Excel, CSV, or JSON file, with bilingual fields captured intact (English description and Traditional Chinese item name in their own columns), an Outlet column populated from the delivery address on the invoice, and a Cost Type column populated from the line-item content. The bookkeeper reviews the structured file, imports it into Xero or QBO HK or MYOB through the platform's supported bill-import path, and applies judgment at the review step rather than at the typing step. This is the route the AI-powered supplier invoice extraction for multi-outlet F&B operators workflow at invoicedataextraction.com is built to cover for HK F&B chains.

The chain finance function writes the prompt once — vendor and invoice number from the header, line items with the English description and the Traditional Chinese item name in separate columns, outlet derived from the delivery address, cost type derived from the line-item content with the food-versus-beverage-versus-packaging rule spelled out the way the chain runs it — and the same prompt runs over a weekly batch from every outlet. There is no pre-built F&B template; the allocation logic lives in the prompt the chain writes, in plain language, and the output reflects it.

The decision tends to settle by scale. Under roughly 80 invoices per week, manual entry holds and the tooling overhead doesn't earn its keep. Above that — and most multi-outlet chains are above it by the time they cross five locations, as is any hotel running three or more F&B outlets posting through a single AP function — the line-item bilingual workload outpaces what built-in OCR delivers on its own, and pre-accounting AI extraction starts paying for itself in bookkeeper hours saved at month-end close. That is the band where the Hong Kong restaurant group finance manager supplier invoice automation conversation moves from optional to operational.

---

What This Workflow Doesn't Solve — and Where the Adjacent Pieces Live

Supplier-invoice extraction with two-axis coding is one slice of the chain finance operating system. Several adjacent layers sit around it, and naming what this workflow does not handle is part of running it honestly.

Inventory management is downstream and out of scope. The supplier-invoice extraction tells the books what was bought; perpetual inventory, stock-on-hand counts, and consumption variance are a separate system — Marketman, MarketBoy, and Crunchtime are the names HK chains run when they decide they need that layer. Recipe costing sits one step further downstream. Outlet-level food cost from supplier invoices feeds the F&B percent denominator; mapping that food cost to per-dish theoretical cost (and the variance between theoretical and actual) lives in recipe-costing tooling and in the menu-engineering work the head chef does with the operations director.

POS sales-side data is upstream of the per-outlet P&L the AP coding feeds. Per-outlet revenue comes from whichever HK-built POS the chain runs — Eats365 across most chain F&B operators, Affinity POS in some boutique groups, Square HK in cafe-style operations, Storehub across smaller chains. Supplier-invoice extraction handles the cost-side. Per-outlet P&L only reads true when both sides land against the same Tracking Category coding, which is why the architecture section earlier treated the sales-side mirror as a structural requirement rather than a nice-to-have.

Payroll and MPF are sibling, not same. Labour cost — the third leg of prime cost — comes from payroll output and the monthly MPF remittance statements, which are their own document type with their own retention obligations. Treating payroll as a separate workflow keeps the AP and the payroll feeds independent in the books, which is the structure the IRD expects to see.

FEHD food-safety records sit alongside IRD retention. The same supplier invoices feed food-traceability records under the FEHD framework — supplier identity, date of receipt, what was received — but the rule set, the retention period, and the operational use are different from Section 51C, and the dedicated walk-through is the cleaner reference than reproducing it here.

Three-way matching of purchase order against delivery note against invoice is upstream procurement-side reconciliation. Chains that run three-way matching catch supplier price errors and short-deliveries before invoice posting, which keeps the AP record clean from the start — the practitioner walk-through of aligning PO, delivery note, and supplier invoice in a HK restaurant procurement workflow covers how the three-document reconciliation runs in practice. Chains that run AP-only catch the same errors at month-end statement reconciliation, if at all — which is the difference between a control that catches errors when they're cheap to fix and a control that surfaces them after the bank payment has already gone out.