AI integration
AI integration means putting a model to work inside a process that already exists, with the guardrails and fallbacks that make it safe to rely on. The hard part is not the model — it is deciding where a model genuinely beats ordinary code, and building the checks that catch it when it is wrong.
Companies that have run AI trials without shipping anything, or that have a specific task involving messy documents, text or conversations.
Typical projects
Document understanding
Contracts, invoices and forms read and turned into structured data, with a human check on anything low-confidence.
Support triage
Incoming messages classified, routed and drafted against your own history — with a person approving before anything sends.
Search over your own material
Retrieval over your documents so staff get answers with citations from your material instead of guesses.
Private and on-premise deployment
Where data cannot leave your perimeter, running open models on your own hardware instead of a public API.
Deliverables
- A written recommendation on what should and should not use a model
- The integration running in your environment with usage limits in place
- Evaluation cases so you can tell when output quality moves
- Cost projections at your real volume, before you commit
$6,000 – $18,000 CAD
2–4 weeks per sprint, including an evaluation pass before launch.
Fixed scope, fixed price. If it doesn't ship, you don't pay.
Priced in CAD. USD contracting available through our US entity.
Common questions
Should we use AI for this, or is regular software enough?
Regular software is enough more often than the market admits. Use ordinary code when the rules are knowable and the input is structured — it is cheaper, faster and predictable. Use a model when the input is messy language, images or documents and the task needs judgment. We give you that answer during the free scoping call, including when the answer is that you do not need us.
Why not just ask ChatGPT to build it ourselves?
You can, and for small internal experiments you probably should. The gap is that a model will produce something that looks finished and is subtly wrong, and you only find out in production. What you are paying for is someone who can tell when the output is wrong in this domain, and who is accountable when it is.
Can we keep our data private?
Yes. Where data cannot leave your perimeter we deploy open models on hardware you control, so nothing is sent to a third-party API. Where a hosted API is acceptable we configure retention and access settings explicitly rather than relying on defaults.
What does it cost to run once it is live?
We project running costs at your actual volume before you commit, and design to keep them low — caching, smaller models for easy cases, and hard usage limits so a bug cannot produce a surprise bill. Typical integrations at small-business volume run in the tens of dollars per month, not the thousands.
Want a fixed price for this?
Thirty minutes on a call and you will have a written scope and a number.