A monthly quality run
Each capability re-tested on your own cases: the set agreed at the start, plus real questions from that month. You get a short report of what improved, what slipped and what we fixed.
AI Care · $1,500 a month · per capability
AI Care. Your live AI, kept accurate, safe and affordable.
AI that worked at launch drifts: real questions change, providers ship new models, costs creep up. Every month we re-test it on your own cases, watch quality and cost, test model updates before they reach your customers and make the fixes.
Every month
Re-tested on your own cases
Cost per customer
Tracked, with savings proposed
Model updates
Tested before any switch
Live products we keep shipping for
AI Care report · October
Meridian · Support assistant
Cases answered correctly47 of 50
Answers that cite a source50 of 50
Cost per customer
$0.82 a month
Fix hours used
6 of 10
- Fixed: refund questions now read the new policy page
- New model version tested and switched: 2 more cases correct
Illustrative example on a fictional product. Not client work.
Brownfield AIProduction AI inside software people already rely on. It is all we do.
- In AI and data since 2018
- 150+ projects delivered
- 60+ AI projects
- AWS Select Tier Partner
- Claude Partner Network
- AG-UI contributor
For EdTech
AI that keeps working through the school year.
For learning, training and student-records platforms, AI Care re-tests each live AI feature every month on your own cases, such as a new term's courses or this month's support questions, tests model updates before your learners see them and keeps student data in your own cloud. $1,500 a month per capability, month to month.
What it covers
Six things done every month, for each live capability.
The same tests that cleared your AI for release, run again on real use. If we did not build it, the AI Scorecard sets the baseline first.
Monitoring and alerts
Answer quality, failed tasks, response time and cost per customer watched every day. When one moves past the line we agreed, we look at it before your customers notice.
Model updates without surprises
When a provider ships a new model or retires an old one, we test it on your cases first and switch only when the results hold.
Fixes included
Up to 10 hours a month of prompt, tool, retrieval and data fixes. Anything larger gets a fixed quote before work starts.
Security tests every quarter
The data separation and prompt injection tests run again every three months, so your answers to customers' security questions stay current.
A named engineer
One engineer who knows your product and your cases, with a reply within one business day.
After launch
Most AI is left alone once it ships. Yours is not.
What usually happens to AI after its launch week, and what changes under AI Care.
AI left alone
It worked at launch.
- After launchThe team moves on to the next feature
- When quality slipsA customer finds it first
- New model versionsSwitched on, or forced by a retirement date
- What it costs to runFound on the monthly bill
- Security questionsAnswered from tests run once, at launch
- CommitmentAn open-ended support retainer
Your customers find the problems first.
AI Care
It keeps working.
- After launchEach capability re-tested every month on your own cases
- When quality slipsMonitoring flags it, and we fix it within the month
- New model versionsTested on your cases before any switch
- What it costs to runCost per customer tracked, with savings proposed
- Security questionsAnswered from tests re-run every quarter
- CommitmentMonth to month per capability, stop at any time
- Monthly tests
- Cost tracked
- Quarterly security
How it runs
A baseline, then the same checks every month.
Four steps, from a free call to every month after. Each one ends with something you keep.
30 minutes
No chargeAI Care call
You show us the AI you have live. We agree which capabilities AI Care covers and the cases each one is re-tested on.
You keep
A clear scope and price
Before the first month
Included after a Datics buildBaseline
For AI we built, the release tests are the baseline. For AI built elsewhere, the AI Scorecard ($5,000, two weeks) tests it on 50 of your cases first.
You keep
The test set and today's results
Every month
$1,500 per capabilityQuality run and fixes
The cases run again, the dashboards are read, model changes are tested and up to 10 hours of fixes are made.
You keep
A monthly report
Every quarter
IncludedSecurity re-tests
Data separation and prompt injection tests run again, with the results written up for your security reviews.
You keep
Current test results
Before a model switch
A new model reaches your customers only after it passes your cases.
- Run on the same cases as the current model
- Compared on answers, response time and cost
- Switched only when the results hold
Sample model check
Before a switch
The new version run on the same 50 cases as the current one.
Current model: cases correct47 of 50
New model: cases correct49 of 50
Switch approved
Two more cases correct, response time the same, cost per answer 18% lower.
Illustrative example on a fictional product. Not client work.
The offer · AI Care
Your live AI, re-tested and looked after every month.
One price per capability, month to month. The first 30 days after a Datics build are included.
$1,500
AI Care · a month per capability
What you keep
- A monthly quality report: what improved, what slipped and what we fixed
- Up to 10 hours of prompt, tool and data fixes each month
- New model versions tested on your cases before any switch
- Data separation and prompt injection tests re-run every quarter
- The test cases, the dashboards and the fixes, all yours
Month to month, with no minimum term. The first 30 days after a Datics build are included. Stop at any time and keep everything.
It fits if you have
- AI already live with your customers, built by us or by your own team
- A team that wants an outside engineer watching quality, cost and model changes
- Customers who ask how you keep the AI accurate over time
It is not for
- AI not live yet: start with the Build Sprint
- AI we did not build and have not tested: start with the AI Scorecard, which sets the baseline
- New features: those get their own fixed quote
Is the AI not live yet? The Build Sprint puts it in your staging in three weeks.
What clients say
Live products, kept shipping.
Clients whose products we keep working on after launch, in their own words.
CleanHaus · Every agent run traced and logged
“I want to recognize and let you know that your work is having a truly positive outcome. Our recent tests are going so much better!”
Coast Technology · Shipping every sprint
“You've always achieved everything that we've asked for, and it's amazing. The quality of your work is second to none, and we couldn't be more pleased.”
5.0 average client rating
Questions
AI Care, answered.
Want an engineer inside your team instead? See forward deployed engineers.
What does AI Care cost?
$1,500 a month for each live AI capability, month to month, with no minimum term. That includes the monthly quality run, monitoring, model updates, up to 10 hours of fixes and the quarterly security tests.
Is AI Care included after you build something for us?
Yes, for the first 30 days after go-live. After that it continues only if you want it, at $1,500 a month per capability.
Can you look after AI we built ourselves?
Yes. We start with the AI Scorecard, two weeks and $5,000, which tests it on 50 of your own cases. Those cases become the set AI Care re-tests every month.
What counts as a capability?
One AI feature your customers use, such as an assistant, an agent, a document extraction or a search. A capability that serves every customer of your product still counts as one.
What if a fix takes more than 10 hours?
We tell you before the hours run out, and larger work gets a fixed quote before it starts.
Who owns the code, the tests and the dashboards?
You do. Everything stays in your repositories and your cloud account, so you can stop AI Care at any time and keep running it yourselves.
Have AI live with your customers?
Show us what it does today. In 30 minutes you will know what AI Care would watch, which cases it would re-test and what it costs.