IBM, Walmart and the Food Safety Blockchain That Failed

Starting in 2016, Walmart and IBM set out to answer a simple food safety question: when there is an outbreak, how fast can you find where a product came from? In a test with sliced mangoes, tracing a package back to its farm took Walmart’s team nearly seven days. Using a blockchain network built on Hyperledger Fabric, the same trace took 2.2 seconds.¹

The result made headlines. In 2018, after a series of E. coli outbreaks linked to romaine lettuce, Walmart asked its leafy greens suppliers to join the network.¹ Blockchain looked like the future of food safety.

Why it failed: the technology worked, the input did not

A blockchain is very good at one thing: once a record is written, it cannot be quietly changed. But it has no way of knowing whether that record was right in the first place. The World Economic Forum put it bluntly in its blockchain toolkit: “If someone inputs garbage data onto a blockchain, that garbage is recorded forever and can inadvertently become a flawed source of truth.”²

In the mango and leafy greens pilots, suppliers uploaded data through web interfaces and new labels.¹ Someone still had to type in the lot, the date, the temperature and the inspection result. If that person wrote it on paper first, entered it at the end of the shift, or guessed, the blockchain faithfully locked in the guess.

That is why food safety on blockchain failed to become the industry standard it promised to be. The wider wave of enterprise blockchain faded with it. In 2022, Maersk and IBM shut down TradeLens, their blockchain platform for global shipping, after it failed to reach commercial viability.³ The lesson for food and manufacturing is not that traceability does not matter. The problem was never putting data on a blockchain. It was always the data input: the first record, captured by a person, on the floor. That is the problem Cosito solves.

Traceability rules raise the stakes

Regulators are moving in the same direction. The FDA’s Food Traceability Rule requires companies handling certain foods to keep records of key events such as receiving, shipping and transforming products. Its compliance date was recently extended to July 20, 2028.⁴ Whatever system a company chooses, the rule depends on the same thing the blockchain pilots did: accurate data entered at the moment the event happens.

Where quality data gets lost today

On most plant floors, a quality check is still reduced to a checkmark on paper: pass or fail, often without the reading, the detail or the reason. When there is a reading, it may be written minutes or hours later, from memory. It rarely reaches the quality module of the ERP, where it could drive a decision. Tablets and ERP screens were supposed to fix this, but they ask an inspector wearing gloves, next to a line, to stop, log in and type field by field. So the entry gets postponed.

Fix the input, and everything downstream improves

With Cosito, the inspector says what they measured and saw, at the moment they see it: “Lot 12, batch 5, internal temperature 30 degrees Fahrenheit. Crust color normal. Two units with torn edges, set aside.”

Cosito ties that sentence to the right lot and batch, the right inspection point and the right specification. Each reading is checked against its range. The record carries a time and the inspector’s name. The original audio, the transcript and the structured result stay together, so anyone can verify what was said.

Because Cosito mirrors the customer’s SAP or Microsoft ERP one to one, that result can become the record the plant already works with: an inspection result on the lot, or a quality notification when something is out of specification. A failed check becomes an action with an owner, not a note in a binder. And whether the data later feeds an ERP, a traceability report or even a blockchain, it starts accurate.

Details that matter on the floor

• Bilingual crews: many plants mix Spanish and English in the same sentence. Cosito is built for that.
• Photos with findings: inspectors can attach a photo to an inspection point, so a defect is documented with evidence.
• Traceable corrections: if a value is corrected later, the change is recorded with its own time.
• Typing is always an option: voice is the fastest path, not the only one.