AI Transformation · AI Product Engineering
AI Proof of Concept & MVP
Before a full build, we test the idea on your own data. Success criteria are agreed first, the smallest version that can test them is built, and the results lead to a clear decision: go, iterate or stop.
The decision path
- 1HypothesisWhat should change, and for whom
- 2CriteriaAgreed before the build
- 3BuildThe smallest version that can test them
- 4EvaluateOn your own data
The decisionAgainst the criteria
StopReasons recorded
IterateFix, then test again
GoPlan the production build
How to get started
Three steps to a plan for AI Proof of Concept & MVP
- 1Tell us what you needUse the project form or book a call. A few sentences about the goal is enough to start.
- 2Free technical consultationWe go through your goals, users, existing systems and constraints with you.
- 3Your planA detailed plan covering the right tech stack, architecture, timeline and budget. Then you decide.
Which to start with
Proof of concept or MVP
Both test an idea before a full build. They answer different questions, involve different people and are built to different standards.
- The question it answers
- Can it be done, on our data, to the standard we need?
- Who uses it
- The project team and a few of your subject-matter experts.
- Data
- A representative sample of your real data, including the difficult cases.
- Built to
- Test feasibility. It is not hardened for production and may be set aside once it has answered the question.
- Success is measured by
- Technical criteria agreed in advance, such as accuracy, grounding, response time and cost per request, measured on an evaluation set.
- What happens next
- A go, iterate or stop decision. On go, an MVP or the production build.
Success criteria
What “success” means, agreed in advance
An example sheet for an assistant that answers customer service agents' policy questions. The targets are illustrative; yours are agreed with your team before the build.
Criteria sheet
Example: a policy assistant for service agents
| Criterion | Measured as | Example target |
|---|---|---|
| Answer quality | Correct answers on an evaluation set of real questions from agents, checked by your policy experts | At least 90% correct |
| Grounding | Answers that cite a passage which supports them | Every answer cites its source |
| Knowing its limits | Questions outside the policy content declined rather than guessed | At least 95% declined |
| Speed | Time until the answer starts to appear, for 95 in 100 questions | Under 3 seconds |
| Cost | Model and hosting cost per question at the expected volume | Within the business case |
| Access | Agents see only content their role permits, in permission tests | No failures |
The decision rules
Go
Every criterion met. The production plan and estimate follow.
Iterate
Close to the targets, with a clear cause and fix. Another round, against the same criteria.
Stop
A criterion can't be met with the data, technology or budget available. The reasons are recorded for later.
What we need from you
A fair test needs your people and your data
A business owner
Who agrees the criteria and makes the go, iterate or stop decision.
Subject-matter experts
Who supply real examples and judge the outputs, for the evaluation set.
Representative data
A sample of the real data, including its difficult cases, with access agreed with your security team.
The pilot group
For an MVP: the users who will use it in their daily work and report what they find.
Deliverables
What you receive
- A hypothesis and success criteria, agreed in writing
- An evaluation set built with your experts
- A working proof of concept or MVP
- An evaluation report against every criterion
- A go, iterate or stop recommendation
- A production plan, architecture and estimate
After a go decision
Where the work goes next
- AI Product EngineeringGenerative AI DevelopmentBuild the assistant, search or agent the proof of concept tested, to production standard.
- AI Product EngineeringMachine Learning DevelopmentTrain the full model on all the data, validated and explained.
- Enterprise AI IntegrationData Foundations for AIFix the data gaps the evaluation exposed, before the full build depends on them.
Questions
Questions about
AI Proof of Concept & MVP
Should we start with a proof of concept or an MVP?
Start with a proof of concept when the question is whether the AI can reach the standard you need on your data. Start with an MVP when that is already known, for example with an established technique, and the question is whether people will use it and what it is worth.
Why agree the success criteria before building?
So the decision depends on results, not on how impressive a demonstration looks. Criteria set afterwards tend to fit whatever was built.
Is stopping a failure?
No. A stop decision, with its reasons, saves the cost of a full build that would not have met its targets, and often shows what would need to change for the idea to work later.
Is the proof of concept code reused?
Sometimes. A proof of concept is built for speed and may be set aside. An MVP is built to production standard for its scope, so it is extended rather than rebuilt.
How is a project priced?
Well-defined scopes are delivered as fixed-price engagements; when requirements are still evolving, we provide a dedicated team instead. Either way, the free technical consultation ends with a plan covering tech stack, architecture, timeline and budget, so you know the cost before work starts.
AI Product Engineering
Other services in this line
- Generative AI DevelopmentAssistants, search and agents built on large language models, grounded in your own content and tested before release.
- Machine Learning DevelopmentPrediction, classification and recommendation models trained on your data, validated on data held back and explained.
- Intelligent Process AutomationProcesses redesigned and automated with rules, AI and human approval at the steps that need judgement.
- MLOps & AI OperationsDeployment, monitoring, retraining and support for AI in production, with its cost and results tracked.
Ready to talk about AI Proof of Concept & MVP?
Start with a free technical consultation: a plan covering the right approach, architecture, timeline and budget for your AI work.
Start a project