The problem
Theo Pouw Groep processes contaminated soil into reusable building materials. Its Environment department receives hundreds of requests a day for the intake of soil batches, each with soil surveys, analysis reports and borehole logs that together can run to hundreds or thousands of pages. Far from every request is worth pursuing: some are not feasible, and often several organisations ask for the same work. That only became clear after someone had worked through the documents. Account managers laid reports from different agencies side by side and checked from memory whether anything was missing, such as a PFAS analysis in an older report. A lot of attention went to requests that led nowhere or were assessed twice.
The solution
We built an AI assistant that sorts and prepares the flow of requests before an account manager looks at them. A connector on the central mailbox recognises requests and links the attachments to a file automatically. The assistant then does three things:
- Insight into which requests are worth pursuing: the AI reads the reports, extracts the relevant data (soil type, contaminants and concentrations, tonnage, location) and tests cleanability against the applicable guidelines. It is clear straight away which requests are promising and which are not feasible.
- Spotting duplicates: when several organisations ask for the same work, the assistant puts those requests side by side. The same work is assessed once, and the team sees at a glance how much demand there is for it.
- Preparing documents for review: from hundreds of pages the assistant pulls out only the parts that matter and lays them out clearly, with missing or outdated information marked. Every value traces back to report, page and borehole with one click.
The account manager reviews and decides. The judgement stays with people.
The result
The assistant has been in use since September 2026. Account managers see at a glance which requests are worth taking on, instead of reading every request from front to back. Requests from different organisations for the same work are recognised and handled once. Whoever picks up a request gets the relevant passages rather than a stack of reports, and gaps are flagged consistently. The team can handle a much larger flow of requests without the care of the assessment suffering.
