The Workflow Pilot is a two-week, fixed-scope engagement: we build a working extraction or decision pipeline for one workflow, run it on your real documents, and measure its accuracy on about 200 samples. Nobody should sign a build contract to find out whether an AI system can handle their paperwork.
Not a mockup. A running pipeline pointed at your own documents, producing output you can inspect field by field.
A labeled set built from your documents with the people who do the work, and the system scored against it, per field and per document type.
Where it broke, which formats and sources it struggled with, and what fixing each would take. The uncomfortable half is usually the most valuable part.
If the workflow is a good fit, a scoped build proposal. If it is not, we say so with the evidence, and you keep the code and the labeled set either way.
Around 200 real documents from the workflow and an hour or two from whoever runs it today. Everything else is on us.
One workflow and one document family, built as a standalone pipeline with its labeled set and accuracy report. Connecting it to your live systems, more document families, or a second workflow are build work, and the pilot is credited in full toward that build.
Then we tell you, with the evidence. That is a legitimate outcome of the pilot. A vendor who has never given a client that answer is not measuring anything.
Yes, including the labeled set. If you never speak to us again you keep the code and the measurement.
NDA before anything is shared. We can work on redacted samples, and where your policy requires it we work entirely inside your environment.
Core · 4–10 weeks
The Production Agent Build ships the agent, its integrations and a human exception queue in 4–10 weeks, with evals and monitoring running from the first day rather than bolted on at the end. Most agent projects stall because nobody can prove the thing is right. We ship the proof alongside the system.
Ongoing · Monthly
The Reliability Retainer keeps a production agent accurate after launch: regression runs on every change and every model release, judge recalibration as your traffic shifts, prompt and tool fixes, and a monthly report on where quality stands. Document formats change, user behaviour drifts, models get deprecated. Something has to notice.
Next step
Audit the agent you already run, or pilot one workflow on your own documents. Either way, you leave with a number you can check.