AI goes where the work already is
Most businesses that ask about AI do not want a new product. They want the software they already use to do a little more. Summarize the notes in a record. Draft the reply a staff member always writes. Read an incoming form and fill in the fields. Search years of documents by meaning rather than keyword.
This is much easier to do well in a system someone already understands. That is why this service is an expansion for software we already support, not a first engagement. If you are not yet a client, the route is a Takeover Assessment, which includes a modernization and AI opportunity map.
The packages
| Package | What it is | Price |
|---|---|---|
| AI Quick Win | One feature, about two weeks, with a 25-case evaluation set | $9,000 |
| AI Feature Pack | Three features, guardrails, an evaluation harness and a runbook | $26,000 |
| AI Workflow or Agent | A workflow that spans systems, quoted after scoping | From $60,000 |
| AI Ops | Ongoing evaluations and regression checks | $1,500 per month |
Prices are in USD with a fixed scope.
AI Quick Win
One feature, built into your existing application, in about two weeks. We start by agreeing what a good answer looks like and writing 25 test cases. We build the feature, run it against the cases, and show you the results, including where it gets things wrong. You decide whether to keep it.
AI Feature Pack
Three features that share one design: the guardrails that limit what the AI may do, an evaluation harness so you can re-test as models change, and a runbook describing how it works and how to switch it off.
AI Workflow or Agent
A workflow that crosses systems, for example reading incoming documents, updating two applications and drafting a reply. Because the scope depends heavily on your systems, we quote it after a scoping conversation.
AI Ops
Models change. A feature that worked in March can behave differently in September. AI Ops runs your evaluation set on a schedule, flags regressions and keeps the feature honest. It is $1,500 per month.
Your account, your data
Shipped features run on your own AI provider account, not ours. You pay the provider directly, you can see usage, and you can change providers later. We do not resell AI access, and we do not keep copies of your data to improve anything of ours.
What we do before we write anything
- Pick a feature with a clear test. If we cannot describe a right answer, we will say so and suggest something else.
- Check the data. Old systems often have messy data. We look at it before we promise anything.
- Decide what the AI may and may not do. For example: it can draft but not send, suggest but not change a record.
- Keep a person in the loop wherever a mistake would matter.
Honest expectations
AI features are useful and they are imperfect. We will show you failures during the quick win, not after launch. Some ideas will not be worth building, and we will tell you. We do not promise a percentage saving, and we will not quote statistics about what AI does for businesses in general. We show you what it does on your data, using your test cases.
How it fits with the rest
If you are on a retainer, AI work gets scheduled alongside your normal support, and we already know the codebase, the hosting and the deployment steps. Changes arrive as reviewed pull requests, and nothing reaches production without your written approval. Our own delivery work is AI-assisted and human-reviewed, and we apply the same standard to what we build for you.
If an upgrade has to happen first, for instance because the application runs on an unsupported runtime, we will say so. That is usually a fix and upgrade project before the AI work starts.
Next step
Tell us about your software and the task you would like to improve. We will reply with whether an AI feature is a good fit, what it would cost and what we would need to see first. Pricing for every option is on the pricing page.
