How we work
Production is where AI succeeds or fails.
Most AI projects die between the demo and daily use. Our method is built around that gap: fixed scope, a metric that matters, weekly demos of a real system and a handover your team can run with.
The 90-day method.
Five steps, one metric, no surprises. Click through the steps or let them play.
Step 01
Discovery
A short call and a look at your data. We tell you honestly whether AI is the right tool for the problem and what it would take.
Week 0 · 30-minute call
Step 02
Blueprint
A scoped proposal with a success metric in your units, a timeline and a fixed price. No surprises later.
Weeks 1–2 · Fixed price
Step 03
Build
The baseline is beaten in week four or we stop. From then on you see a working system every Friday, on that week’s data.
Weeks 3–8 · Weekly demos
Step 04
Deploy
We integrate with your systems, run alongside your current process, set up monitoring and hand over documentation your team can run with.
Weeks 9–12 · Shadow run, then live
Step 05
Run & improve
Models drift. We watch them, retrain them and keep improving the numbers that matter to you.
Ongoing · Monitoring and retraining
From our hands to yours.
Every engagement is designed to end with your team in charge. The share of the work we do shrinks by design.
Weeks 0 to 12
First use case to production
We lead. Discovery, blueprint, baseline, build and deployment of the first system, with your team in every weekly demo.
- Discovery call and data audit
- Fixed-price blueprint with a success metric
- Baseline beaten by week four or we stop
- Shadow run, then live
- Runbook and handover
Months 3 to 6
Your team runs it with us
Shared. Your engineers make changes with us reviewing. Retraining, monitoring and the next use case are planned together.
- Pairing on changes and retraining
- Monitoring reviews on a schedule
- Second use case scoped from what you learned
- Playbooks for the questions that came up
Month 6 onward
You own the roadmap
Yours. You decide what comes next. We stay available for reviews, hard problems and new problems, on a retainer or per project.
- Quarterly architecture and model review
- On-call for the hard cases
- New problems scoped when you are ready
Who is on your project.
A small team, senior from the first call. No hand-off to juniors after the sale.
Lead data scientist
Owns the model, the baseline and the numbers. Runs the weekly demo and answers the hard questions.
Engineer
Builds the pipeline, the API and the integration into your systems, and makes sure it survives without us.
Project manager
Keeps scope, timeline and language plain. Your single point of contact from the first call to handover.
Designer
Joins when the system has a face: dashboards, assistants and product interfaces people will actually open.
Forward-deployed AI engineers
Our engineers, inside your team.
Not every problem needs a 90-day project. Sometimes you need a senior AI engineer in your team now, without the hiring. That is the second way we work, at a fraction of what building the team yourself costs.
What they actually do
- Week one: read your code and your data, sit in your standups, pick the first thing worth shipping.
- Build production AI features in your stack: retrieval, extraction, forecasting, evaluation.
- Put in the evals, monitoring and retraining your team will need after they leave.
- Pair with your engineers so the knowledge stays when they go.
- Say no to AI when a rule or a query does the job.
How it works
A senior DataMinds engineer joins your team for a fixed number of days a week and reports to your product or engineering lead.
- Works in your repositories, your tools and your meetings.
- Backed by DataMinds: design, project management and a second senior pair of eyes on every model.
- Starts with a short scoping call, not a 90-day blueprint.
- Ends when you say so, with a handover.
Where the fraction comes from
You pay for senior engineering days, not for building and keeping a team. No recruiting, no ramp-up, no headcount you cannot scale back.
| Hiring in-house | Forward-deployed | |
|---|---|---|
| Time to start | Months of recruiting | Weeks |
| Seniority | Whoever the market offers | Senior from day one |
| What you pay for | Salary, employer costs, tooling, recruiters, ramp-up | A monthly rate for the days you use |
| Behind the engineer | Nobody, until you hire more | The whole DataMinds team |
| Commitment | Permanent | Month to month, scale up or down |
| Ownership | Yours | Yours: your repos, your models |
Runs where your data lives.
Hosting is a choice you make in the blueprint, not a constraint we impose.
Your cloud
Deployed in your own Azure, Google Cloud or AWS tenant. Your accounts, your billing, your access controls.
EU-hosted by us
We run it in EU regions and hand over credentials and documentation. Good for a first project without infrastructure work.
On-premise
Inside your data centre, including open-weight models when data cannot leave the building.
Hybrid
Sensitive data stays with you; compute and models run where it makes sense. Designed case by case.
One number, agreed up front.
The blueprint fixes scope, timeline and price for the first 90 days.
Included
- Discovery and blueprint
- Design, build and evaluation
- Deployment and integration
- Monitoring and runbook
- 30 days of support after launch
Changes the number
- A second use case in the same window
- New data sources found mid-project
- Custom interfaces beyond a dashboard
- Hosting we set up and run for you
Agreed in writing before the work, never after.
Questions people ask before they sign.
What if the data cannot support the goal?
We say so in the blueprint and stop there. You get a written plan for what to collect and why. It has happened, and those clients came back later with the data in hand.
How is the price set?
The blueprint fixes the scope and the price for the first 90 days. It covers design, build, deployment, the runbook and 30 days of support after launch. Changes in scope change the price, and we agree them before we do the work.
Who owns the code and the models?
You do. The code lives in your repositories from the first week, the models are trained on your data and stay with you, and nothing we build is tied to a licence from us.
Do you work with our existing vendors?
Yes. We integrate with your ERP, CRM, data platform and cloud rather than replacing them. If a vendor is the wrong choice for the job, we say so and explain why.
What happens after the 90 days?
Either your team runs the system with us reviewing, or we stay on a retainer for monitoring, retraining and the next problem. Both are common; neither is required.
Can we bring in one engineer instead of running a project?
Yes. A forward-deployed engineer works inside your team on a monthly rate, part-time or full-time, and reports to your lead. Most teams start with a short scoping call, ship something in the first month and decide from there.
Start with a 30-minute call.
Tell us the problem. We will tell you if AI is the right tool, what it would take, and what it would cost.