The pattern matters more here than the list of names. In all eight the client came not for “AI” but for one specific bottleneck, and the bottleneck had the same shape every time.
First: the bottleneck is the exception, not the norm. The routine 80% of the process already works somehow in the existing system. What burns people is the rest — a shipment with a questionable HS code, a parcel stuck in a locker, an inbound delivery short of stock, an application without a certificate. Those cases eat the hours, and they are precisely the ones a template cannot close.
Second: the answer lives in documents, not in a database. A service bulletin, an SOP, a destination-specific customs requirement, a warranty procedure, a retail chain’s rules. So the core of such a system is retrieval over the client’s own documents with a citation back to the source, not text generated from nowhere.
Third: the decision stays with a person. A risky shipment escalates to an operator, a warranty case goes to a service engineer, a contested application goes to a consultant. The system prepares the case, shows the grounds and proposes the next action — it does not decide on behalf of whoever is accountable.
That third property is not our good manners but a requirement wherever a regulator exists. In recruitment it is written into the EU AI Act, which is why an architecture proven in logistics transfers to an agency without redesign.
| Company | Sector | What was built |
|---|---|---|
| InPost | Logistics, parcel lockers | AI dispatcher for delivery exceptions |
| DHL | International logistics | Copilot for customs and transport documents |
| FIEGE | Warehousing, contract logistics | Warehouse exception manager |
| Bridgestone | Automotive components, dealer network | Assistant for dealers and service teams |
| JDE Peet’s | FMCG, retail chains | Order fulfilment and activation control |
| CEVA Logistics | Freight forwarding | Multi-agent control tower |
| Škoda | Automotive, dealer network | Dealer knowledge hub |
| TRANSITUM | International employment | Lead generation and routing |
There are no figures for the seven enterprise deployments, and that is not an oversight. The volumes, timelines and results of those projects are confidential — that is how agreements at this level work, and we do not bend them for a better-looking page.
So the only case with figures on this site is our own agency, where the data is ours: $1.63 per lead on the winning combination, 115 leads backfilled in one pass, zero minutes of manual work from Meta to CRM. If you want proof of effect in numbers, look at that one — it is broken down to the mechanics.
Under NDA on a call we walk through the architecture of any of these projects in more detail than this page allows — including what did not work the first time.
The volumes, timelines and results of those deployments are confidential — that is what the agreements say. Where the data is our own, the numbers are on the site and broken down to the mechanics.
We build exception-handling systems grounded in documents; the sector decides what the documents contain. Logistics gave us volume and discipline, the international employment corridors gave us the regulatory knowledge. We speak to agencies on this site because that is where our depth is greater.
A chatbot answers questions. These systems act inside the process: they classify an incident, raise a task in the WMS, escalate a risky shipment, build a prioritised card. Answering is only one of the functions.
A person. The system prepares the case, shows the grounds with a citation to the document and proposes the next action. Where a regulator exists there is no alternative: in recruitment it is an explicit EU AI Act requirement.
A reference is the client’s call, not ours. On a call we walk through the architecture under NDA, which usually tells you more than a testimonial.
Start with a short brief — we reply with an estimate.