One manual process, replaced. In production in 2 to 4 weeks.
- Sprint
- Feature Sprint
- Price
- $10,000 to $25,000
- Time
- 2 to 4 weeks
- Fix window
- 30 days
Example builds.
- Invoice and contract intake into your ERP
- Shared-inbox triage with drafted replies and routing
- Form and PDF extraction into a database with a review queue
- Several workflows with an admin console and audit log
What is included.
- Included: One primary workflow, end to end
- Included: Up to two integrations (ERP, CRM, inbox, storage)
- Included: A review queue for low-confidence cases
- Included: An evaluation set built from your real data
- Included: Authentication and roles
- Included: Deployment to your cloud account, with monitoring
- Included: A runbook and a 30-day fix window
How the AI part works.
- What the model does
- Reads, classifies and extracts. It fills fields, routes items and drafts text. It never takes an irreversible action on its own.
- What it must not do
- Approve payments, send external email or delete records without a person confirming. These stay behind a review step.
- How we test it
- Before launch we build an evaluation set from 100 to 300 of your real cases and publish the accuracy. After launch, every correction in the review queue goes back into that set.
- What it costs to run
- Model usage is on your account. For a typical document workflow it is tens of dollars a month, not thousands. We show the number before launch.
What this looks like
Most operations teams have one process that eats a person’s week: reading documents and typing what is in them into a system, sorting a shared inbox, or copying data between two tools. The software we ship reads the input with a model, fills the system of record, routes what it is sure about, and sends what it is not sure about to a short review queue.
The person who used to do the work now checks the exceptions. The volume the team can handle goes up without hiring.
How the sprint goes
Week one: we map the process with you, collect real cases and set the accuracy target. You see the first working slice at the end of the week, and you can stop there if it is not what you expected.
Weeks two and three: the workflow is built out, integrated with your systems and measured against the evaluation set every day.
Week four, where needed: production deploy on your cloud account, monitoring, the runbook, and a handover session with your team.
What you get at the end
Software in production under your own accounts, an accuracy number you can quote internally, a review queue that keeps improving the system, and 30 days of fixes.
Questions we get on the first call.
How much manual work has to exist for this to pay off?
As a rule of thumb, from about 20 hours a week of repetitive reading, typing or routing. Below that, the fixed price takes longer to earn back and we say so on the call.
What data do you need from us?
A sample of real cases, 100 or more, with the correct outcome where you have it. Access to the systems the workflow reads from and writes to. We sign an NDA first if you want one.
What happens when the model is wrong?
Low-confidence cases go to a review queue, a person decides, and the decision becomes training data for the evaluation set. The system gets more precise with use, and nobody loses a case to a silent error.
Can it run inside our existing tools?
Usually. We integrate with the inbox, ERP, CRM or storage you already use. If a tool has no API, we tell you on the first call and quote the workaround.
Book a call, or send the brief. A written price in three working days.
No NDA needed for the call. We sign one within 24 hours if you want it.