Anyone who buys raw materials or works in quality management in the food industry knows the effort involved. Every single raw material has its own document: one for cocoa, one for sugar, one for vanilla. It defines what the supplier delivers. Ingredients in percentages, allergens, nutritional values, origin, shelf life, packaging, certifications such as organic or vegan. It is called a product specification, often simply "spec" internally.
It sounds like paperwork - and it is. But without a current, approved spec, the raw material cannot be ordered. Without raw material, there is no product.
A typical mid-sized manufacturer has more than a hundred of these in circulation. Every recipe change triggers a new review. Every legal change can invalidate existing specs. And every spec that is not renewed in time is a potential production stop and, in the worst case, can lead to legal consequences.
What a spec contains
A raw material specification is usually 1 to 4 pages long. But those pages contain a lot, because they include between 15 and 30 required data points that matter to different departments. The spec has to be reviewed by procurement, quality management, product development, packaging, and logistics. One error can, for example, mean that flour suddenly arrives in a big bag that cannot even be stored, or that a product labeled as vegan is unfortunately not vegan. So the review is extremely important.
When a new spec arrives from a supplier, it can be either a new document or an updated one.
If it is an update, the process in practice usually looks roughly like this:
- The spec arrives by email in procurement.
- It is assigned to the correct supplier, item, and previous version.
- Both versions are compared manually, often field by field.
- If there are major deviations, it is sent back to the supplier right away. Response time: days to weeks.
- If the changes are smaller or unclear, the individual departments have to review them.
- If everything is OK: approval, archiving, and linking to master data in the ERP.
Per spec, this quickly adds up to 5 to 45 minutes, and turnaround time to approval is realistically two to three weeks. There are good reasons for this. Several departments must be involved properly, and late follow-up questions to the supplier are real time-killers. This lead time can significantly delay product development and production.
Where AI helps
Take a typical example from daily QM work. In the morning, five new specs are in the inbox. Normally, the responsible colleague opens each one individually and searches for changes. With an AI agent, she instead sees immediately: three are identical to the previous version, one shortens shelf life by 30 days, and one newly declares that the product may contain traces of hazelnut.
That is the key point. Expert review stays with people. The mechanical prep work disappears.
- Critical: allergen trace note now includes barley, oats, and egg.
- Changed: ingredient order and soy lecithin origin.
- New: label details and palletization field added.
Send supplier follow-up for cleaning validation and updated HACCP documentation.
A well-built agent can read the PDF, compare it with the previous version, and present differences in a structured way. It detects missing required fields: if, for example, an organic certificate is missing or there is no percentage for allergen traces, it is marked immediately. If a statement is unclear, it drafts a ready-to-send clarification email to the supplier. That draft can then be quickly reviewed, adjusted if needed, and sent.
The agent also keeps track of deadlines. Specs are usually valid for one to three years. Instead of someone regularly checking an Excel list, the agent can automatically send renewal requests to suppliers by email.
Why people still decide
Food is a regulated domain. A wrongly declared allergen can lead to recalls, fines, or an allergic reaction for the end customer. Food information regulations, organic certifications, contaminant limits: errors are expensive and sometimes dangerous.
That is why such an agent may suggest, but never approve. Every click on "Approve" comes from a human. The full decision trail is documented and therefore clean from a compliance perspective.
What realistically changes
For most mid-sized manufacturers, three things shift.
Initial review becomes much faster, because the agent handles the prep work. Review quality becomes more consistent, because every spec is checked with the same logic, even when the food expert is on vacation. And the process becomes proactive instead of reactive: expiring specs are visible early enough, not only when someone in procurement stumbles across them.
This creates a noticeable gain in time and quality exactly where both are scarce today.
The agent works with existing systems
All of this only works if the agent is integrated where the team already works. Nobody wants another tool where specs first have to be uploaded.
A sensibly built agent therefore plugs directly into the existing IT setup: emails still arrive in Outlook, specs remain in SharePoint, approvals run via Teams, and master data comes from the ERP. The agent sits in between and handles the work that would otherwise have to be coordinated manually.
Who this is relevant for
If you work in procurement or quality management at a food company, regularly review supplier specifications, and feel audit pressure growing faster than your team, this is exactly your process. Domain expertise is not replaced, but specialist staff is used better: only the repetitive work that does not require expert judgment goes away.
If this sounds interesting for you or your company, feel free to get in touch with us. We help you apply AI in the best possible and compliant way.