Transform — 6 weeks
AI Pilot
A working AI prototype for one concrete operational problem, deployed in 6 weeks. Cost and efficiency impact measured; scale decision made on evidence.Working AI prototype in 6 weeks. Connects to one customer segment, channel or order flow — metric impact measured, scale decision is yours.
No LLM hype — business problem, data, model and integration. A prototype that goes to field test in six weeks, used by real operators. If the pilot succeeds, the production roadmap is ready. If not, what was learned and what should be done differently is documented.One specific e-commerce problem, working in 6 weeks. Product recommendation engine, cart abandonment prediction, customer segmentation, order forecasting — we pick the one with the highest ROAS or LTV impact using data. Real users test it for 2 weeks; metric impact measured.
Scope
What's included?
- Use case selection and value validation — start with no more than two candidates, one selected against data and business impact criteria
- Data inventory and quality check — existing data sources, gaps and cleaning requirements identified
- Model selection (large language model / classical machine learning / hybrid) — with rationale and estimated performance expectations
- Prototype and interface development — at a level real operators or end users can use directly
- 2-week field test with real users — usage metrics, feedback and pilot success criteria measured
- Use case selection — we use data to identify the problem with the highest conversion or LTV impact, maximum two candidates
- Data inventory and quality check — existing customer, product and order data; gaps and a cleaning plan
- Model selection (recommendation engine / classification / large language model) — with rationale and expected metric impact
- Prototype and interface development — usable directly by the marketing team or end customer
- 2-week field test with real users — conversion rate, basket value or retention impact measured
Deliverables
What you get.
- 01Working prototype (source code included, full ownership transferred) — core architecture ready to move to production
- 02Pilot report: usage metrics, cost analysis, efficiency impact and scale recommendation for moving to production
- 03Production roadmap: technical steps, estimated budget and timeline
- 01Working prototype (source code included, full ownership) — architecture ready to move to production
- 02Pilot report: metric impact (conversion, LTV, ROAS), cost analysis and scale recommendation
- 03Production roadmap: technical steps, estimated budget and timeline
Who it's for
Who it fits.
- Industrial firm with a concrete AI use case — production quality control, demand forecasting or maintenance scheduling
- COO or CDO who sees AI potential but is unsure which problem to start with
- Data-rich, process-intensive organisation that wants to see what that data is worth
- E-commerce or D2C brand with a specific AI idea — product recommendations, personalisation or basket optimisation
- Brand or growth leader asking 'we need AI but where to start' — wanting to begin with a concrete use case
- Customer and order data is accumulating but not being processed — wants to turn that data into growth metrics
FAQ
Frequently asked.
Is the software delivered at the end of the pilot production-ready?
The pilot is not a production-ready product — but how to get to production, at what cost and on what timeline, is documented clearly. The six weeks produce a working prototype, a pilot report measuring the metric impact, and a production roadmap. Source code is handed over with full ownership, and the core architecture is built so it can move to production. The decision to proceed belongs to the client, and the scale recommendation sits in the report with its estimated budget.
Do we have enough data for an AI project, and how do we find out?
A data inventory and quality check run in the first weeks, and the answer comes from there. Customer, product, order and process sources are examined, and the gaps and a cleaning plan are documented in writing. Where the data is insufficient, that is said while the use case is still being chosen rather than halfway through the pilot — which is why at most two candidates are taken up, and whichever has ready data gets picked. If it falls short, either the use case changes or measurement is set up first.
Who chooses the use case?
The choice is made together, starting with at most two candidates, and it rests on data suitability and business impact rather than preference. A concrete problem is required — production quality control, demand forecasting, maintenance scheduling, product recommendation, cart abandonment prediction or segmentation. "Trying AI" is not a use case on its own. Any of them can be the candidate; not all at once. The success criteria for the chosen scenario are written in week one, not at the end.
What happens if the AI pilot project fails?
The outcome gets documented and the learning is written down. The pilot report covers what was tried, what the metric did, what should be done differently and the cost analysis; the source code still transfers in full ownership. Eliminating a use case in six weeks costs less than committing an annual budget to the same mistake, or than a project running unmeasured for a year. Because the success criteria were written up front, the result is not open to argument.
How does the AI cost run after the pilot, and what is excluded from the price?
Model usage fees, cloud infrastructure and tool licences sit outside the package price. The price covers the five scope items: use case selection, data inventory, model selection, prototype development and the two-week field test. Post-pilot running cost is written into the production roadmap as an estimated budget line. Consumption-based lines cannot go into a fixed price; that figure rises as usage rises, and presenting it as fixed would mislead.
We have no AI roadmap. Should we build one first?
A roadmap is not required, but a problem definition is. The pilot locks onto a single use case, and a concrete operational bottleneck is enough in place of a corporate roadmap. The pilot itself produces the production roadmap as an output — technical steps, estimated budget and timeline. If even the candidate use cases are unclear, the Digital Transformation Audit comes first; three weeks ranks the candidates by ROI and the pilot starts from there.
Who from our team needs to be involved?
A technical counterpart with data access is needed, along with the side that will genuinely use the prototype in the field for two weeks — operators, the marketing team or end customers. The field test is not simulated; real usage is what gets measured. The process owner signing off the pilot success criteria is a first-week step. Access to customer, product, order and source systems has to be opened during the data inventory.
What happens if the scope changes within the 6 weeks?
The use case is fixed; switching it after the field test starts resets the pilot. Minor corrections inside the prototype are within scope, but a new problem definition is not. A second use case is priced as a separate pilot and needs its own six weeks — six weeks is for one problem, and split in two both come out half-finished. Holding that boundary is what buys a measured metric impact by the end of week six.
Who is this package not for?
Not for organisations with no accumulated data — a model learns from history, and without customer, order or process history the pilot turns into guesswork. Also not for firms expecting a finished, production-ready product to go live; that work belongs to MVP Build. Where no concrete operational problem can be named, the Digital Transformation Audit comes first. Six weeks is planned to answer one defined question.
Is the price fixed, or do extra line items appear along the way?
The price is fixed and tied to the scope list. No extra line item appears inside the six weeks; if scope changes, repricing is put in writing and no surprise line emerges. Model usage fees and cloud infrastructure vary with consumption, so they are excluded from the outset and appear in the roadmap as estimates. The post-pilot running budget is visible with the report rather than after the fact.
AI Pilot or MVP Build — which one fits us?
Evidence points to the AI Pilot; a product points to MVP Build. The pilot measures an idea in six weeks, closes with a two-week field test and ties the scale decision to data. MVP Build reaches production in eight weeks, opens to real users, wires up the metrics and hands over to the internal team after a 30-day stabilisation. The pilot is the cheaper of the two, and the one that makes the decision cheaper; the two can also run back to back.
Related case
10× more quote requests, driven by an AI technical advisor.
Meccanotecnica Umbra is the Türkiye arm of one of the world's leading mechanical seal manufacturers, yet its technical visibility in the local market lagged behind its global standing. We connected the product catalogue to an AI advisor that lays out the right equipment for an engineer describing their plant, and to a quote portal. Quote requests rose tenfold and response time dropped by ninety percent.
ViewWhere do we start?
Three entry doors at three speeds. Pick the one that fits.