Why Food Traceability Breaks When You Need It Most

In most industries, disconnected data costs you time. In food, it costs you a recall — and you usually find out at the worst possible moment, from the worst possible source.

Here's the pattern that should worry every food manufacturer. In Australia, the single most common way an undeclared-allergen recall gets detected is a customer complaint — not internal testing, not a QA system, but a member of the public who had a reaction. By the time the problem surfaces, the product is already on shelves and in homes, and the clock on tracing it has been running for days.

That's the real stakes of disconnected data in food. It turns food traceability from a background admin task into the thing your entire recall response stands on. For most businesses, siloed systems are an efficiency tax. For a food manufacturer, they're the difference between a two-hour precise recall and a two-day scramble that pulls half your range off the shelf.

A recall isn't an inventory question. It's a genealogy question.

This is where a lot of food businesses quietly overestimate themselves. Ask any modern ERP or warehouse system "where is batch 4471 right now?" and it answers instantly. That's inventory. It's solved. Every decent system does it.

A recall asks something completely different. It asks for genealogy: what went into a batch, what that batch became, and where every unit of it ended up. That's not a location lookup — it's a family tree. And a family tree in food manufacturing runs in two directions, both of which matter:

• Backward — from a finished product back through its batch, its ingredients, and the specific supplier lots those ingredients came from.
• Forward — from a single suspect ingredient lot forward to every finished product it touched, every SKU, every production run, and every customer who received them.

Any one system can usually manage a slice of that tree. No single system holds the whole thing. And a recall needs the whole thing.

Where food traceability actually breaks: the forward trace

Backward tracing is the easier half — you start from one product and work back. The half that breaks food businesses is the forward trace, and it happens to be exactly the scenario Australia recalls for most.

Undeclared allergens are the leading cause of food recalls in Australia year after year — 38% of all recalls in 2025, and 197 over the past five years. And one of the most common root causes isn't a mistake you made — it's supplier verification: a raw ingredient carried an allergen, and that information never reached you.

Play that forward. Your supplier calls: a batch of an ingredient they shipped you three weeks ago was cross-contaminated with milk, or mislabelled. Now you have to answer, fast — every finished product that used that specific lot. Not the ingredient in general. That lot.

And that ingredient didn't sit still. It was split across multiple production runs — some into Tuesday's batch, some into Thursday's. Leftover was reworked into a later run. The same line ran three SKUs that week. One incoming lot has now fanned out across dozens of finished batches, several products, multiple shifts, and hundreds of pallets.

Picture the quality manager who takes that call on a Friday afternoon. Supplier lot number in hand, they open the ERP to find the matching goods-in receipt, cross-check batch records for which production orders drew from it, pull the rework log to catch the leftover blended into Monday's run, move to the warehouse system to find where those finished pallets went, then into dispatch to map pallets to customers. Five systems, one lot, every link matched by hand. That's not a query. That's an afternoon — or two.

That fan-out — commingling and rework — is what makes forward tracing a genuinely hard problem. It's not that the data doesn't exist. It's that the links between each step live in different systems, and reconstructing the chain means stitching them together manually.

The data exists. It just doesn't line up.

There's a second, quieter problem underneath all of this: the same physical thing has a different name in every system it passes through.

The supplier calls it lot L-2291. Your goods-in team books it against an internal batch number. Your warehouse system tags the pallets with their own licence-plate IDs. Dispatch groups it into consignment numbers tied to customer orders. Four identities for one lot of one ingredient — and nothing automatically connects them.

So when a trace has to cross from the ERP to the warehouse system to dispatch records to the supplier's certificate of analysis sitting in someone's inbox, a human reconciles those identities at every hop. Industry write-ups on disjointed systems in food manufacturing describe a full trace taking days of manual cross-referencing when it should take minutes. Under recall pressure — product in market, a regulator waiting — that manual matching is slow and error-prone. And an error here doesn't just delay the recall. It means you either miss affected product or condemn product that was never at risk.

This is the "human glue" holding food traceability together: people manually matching IDs across screens. It's invisible right up until the moment it's load-bearing — and a recall is exactly that moment.

When you can't trace precisely, you recall too widely

Here's the cost most people don't price in. When you can't reconstruct the exact genealogy quickly, you don't just lose time — you lose precision. And imprecision has a direct dollar cost.

If you can't prove exactly which finished lots contain the suspect ingredient, the only safe move is to recall broadly. Pull the whole SKU. Pull the whole week. Recall product that was never at risk, because you can't prove it wasn't. Every unit you pull unnecessarily is good product destroyed, and every extra SKU on the notice widens the brand damage.

The numbers make the stakes clear. An often-cited Food Marketing Institute and Grocery Manufacturers Association study puts the average direct cost of a food recall at around US$10 million — retrieval, disposal, and the like — with total impact typically running three to five times higher once you add lost contracts, business interruption and litigation. Nearly a quarter of recalls in that data exceeded US$30 million. FSANZ's own 2017 estimate put the average direct cost of an Australian food recall at roughly AUD $10 million. And this isn't rare: FSANZ has coordinated 866 recalls in the past decade, 92 in 2025 alone.

Consumer trust compounds it. In the research behind those figures, around 15% of shoppers said they'd never buy the recalled product again, and 21% said they'd avoid the brand entirely. Precision is what separates a $10 million event from a contained one — and precision comes from connected data. Nothing else.

The bar is rising — and it assumes your data is connected

Regulators are moving in one clear direction: show me, electronically, fast. In the US, FSMA 204 — the FDA's Food Traceability Rule — will require covered businesses to produce an electronic, sortable traceability record within 24 hours of a request. The compliance date recently moved to July 2028, and the reason for the delay is the tell: the FDA acknowledged that even well-prepared companies depend on accurate data from supply chain partners who aren't ready. The rule didn't get easier. The data problem got officially named.

If you export to the US, that 24-hour, electronic, sortable standard is your future baseline. You cannot hand-reconcile four systems and an email inbox in 24 hours. The only way to hit that bar is to have the genealogy already connected before the request lands.

The fix isn't a better system. It's connected systems.

Notice what the problem is not. It's not that your ERP is weak, or your warehouse system lacks features, or you need to rip everything out. Each system works. The failure lives in the gaps between them — and in the people manually bridging those gaps.

So the fix is to close the gaps. Connect the tables that hold each piece of the genealogy — supplier lots, ingredients, batches, rework, finished lots, pallets, consignments, customers — into one linked model, so the chain of custody exists as connected data instead of four disconnected fragments. That's what we build with Company Brain: a unified data layer across your existing systems that you can question in plain English. A forward trace stops being a two-day reconstruction and becomes a single question — which customers received product containing supplier lot L-2291? — answered from combined, live data.

And the order matters. We say "AI can't fix chaos" for a reason: you can't automate a trace across systems that were never connected. Unify the data first, and only then does a compliance worker like LOUIS or an inventory worker like IRIS have a reliable foundation to reason over — instead of guessing across fragments.

The recall is coming. The cost is your choice.

The recall you'll eventually face probably won't be exotic. On a decade of Australian data, it'll most likely be an undeclared allergen — milk, wheat or a tree nut — in a mixed or processed product, triggered by a supplier change or a packaging error nobody caught. Ordinary. Predictable. The kind of thing that happens to good manufacturers every week.

What won't be ordinary is the cost — and that's decided long before the call comes in. Whether you run a precise, contained recall or a two-day scramble that pulls half your range comes down to one question you can only answer in advance: is your data connected, or is it still held together by people matching IDs across four screens?

If you're not sure how many systems a full forward trace would touch in your business today, that's worth finding out before a supplier forces the question. A free AI readiness assessment is a straightforward place to start.