AI proof of concept,
on your data.
Validate an AI idea on your own data in 4–6 weeks, from £10,000, before you commit to a full build. Magora builds the AI for one job on a sample of your real data, measures how accurate it is and gives you a costed plan for production. A free working prototype comes first, in 5–10 working days.
AI proof of concept,
in one paragraph.
Most AI ideas fail on one question: does it work well enough on our data? An AI proof of concept answers that before you pay for a full build. Magora takes one job, such as classifying documents, answering from your knowledge base or running a step of a workflow, builds it on a sample of your real data and measures how often it gets the answer right. In 4–6 weeks, from £10,000, you get a working pilot, the measured results and a costed plan for production, or a clear reason not to build it.
Does it work on your data?
Accuracy measured on your own examples, not on a vendor demo.
From £10,000, 4–6 weeks
Fixed scope agreed up front. A free working prototype comes first.
Build, change or stop
A costed plan for production, or a clear reason not to spend more.
Prototype, proof of concept
or MVP?
Three different questions, three different steps. For AI projects the proof of concept matters most, because accuracy on your data is the biggest unknown.
Do people get it?
A working prototype on a sample of your data in 5–10 working days, to see the idea before any paid work.
Can AI do the job?
An AI agent or model on your real data in a controlled pilot, with its accuracy measured, in 4–6 weeks.
Will people use it?
The first version of the product in real users' hands, typically 6–8 weeks after Discovery. See MVP development.
AI ideas we validate
most often.
One job per proof of concept, chosen because it is frequent, costly by hand and easy to check.
Document processing
Reading, classifying and extracting fields from contracts, forms, invoices or reports.
Answers from your documents
An assistant that answers from your own files with a cited source. See RAG development.
One workflow automated
An agent that runs a step of a process in your systems, with a person approving. See AI automation.
Email and support triage
Sorting, routing and drafting replies to requests, measured against what your team actually sent.
AI inside your product
A new AI feature tested in your application before it goes to all users. See AI & ML development.
An AI agent for one role
An agent that uses your tools and data to do a defined task. See AI agent development.
From an AI idea
to a decision.
Each step ends with something you can see, and you can stop after any of them.
Free prototype
5–10 working daysA working prototype on a sample of your data, before any paid work.
Discovery
2 weeks, from £3,000The one job to test, the data to use, what counts as a correct answer and the scope of the proof of concept.
Proof of concept
4–6 weeks, from £10,000The AI working on your real data in a controlled pilot, with accuracy measured against your own examples.
Decision
End of the pilotResults, risks and a costed plan for production, or a clear reason not to build.
Production
Typically 6–8 weeksIf you go ahead, a first production release connected to your systems, with monitoring and support.
What an AI proof of concept
costs.
Our 2026 prices in GBP, the same as on our AI agent development page. Adding AI to an existing application typically costs £20,000–£80,000.
A working prototype on a sample of your data in 5–10 working days.
Two weeks: the job to test, the data, success criteria and scope.
AI on your real data in a controlled pilot, accuracy measured, 4–6 weeks.
An AI agent that automates one internal workflow in production.
Answers from your own documents with a cited source.
A full application built around a language model.
AI projects
we've built.
Three AI products we built for a university, a healthcare company and a call-centre technology company.
University of Essex
- Problem
- Sensitive files had to be classified by confidentiality level, and cloud AI services were not allowed.
- Built
- A self-contained AI module with two local models: one proposes tags and confidentiality rules for human review, the other applies the approved rules.
- Outcome
- Classification is automated without sending data to third-party cloud APIs.
BodySync
- Problem
- Clinicians spent long hours drafting personalised treatment plans by hand.
- Built
- A web app that drafts a treatment plan from structured assessment data, restricted to pre-defined peer-reviewed clinical sources.
- Outcome
- The MVP is in use in UK clinics, and further development is planned.
Empath AI
- Problem
- Call-centre agents could not pick up undeclared signs of caller vulnerability during a live call.
- Built
- A SaaS platform that analyses vocal biomarkers in live calls and shows guidance prompts to the agent inside Microsoft Teams.
- Outcome
- Agents get live guidance in the tool they already use, and managers see trends without exposing individuals.
More AI work: AI & ML development. All projects →
AI proof of concept
questions.
A short, fixed-scope project that tests whether AI can do one specific job well enough on your own data: classify your documents, answer from your knowledge base, draft replies, extract fields or run a step of a workflow. At Magora it is an AI agent or model working on your real data in a controlled pilot, with its accuracy measured, in 4–6 weeks.
At Magora an AI proof of concept starts from £10,000 and takes 4–6 weeks. A free working prototype on a sample of your data comes first, and Discovery from £3,000 fixes the scope. If the proof of concept works, an agent for one workflow starts from £20,000, a knowledge assistant over your documents (RAG) from £30,000 and a production LLM application from £40,000.
A prototype tests whether people understand and want the idea. A proof of concept tests whether a piece of technology can do the job. An MVP is the first version of the product that real users use. For AI projects the proof of concept matters most, because accuracy on your data is the biggest unknown.
A representative sample of the real thing: the documents, emails, tickets or records the AI will work on, plus examples of the correct result so accuracy can be measured. The time depends mostly on how many sources there are, how clean they are and how strict the access rules need to be. We sign an NDA on day one.
Yes, when it has to. For the University of Essex we built an AI module with two local models, because cloud AI services were not allowed for their files. Whether a cloud model or a local one fits is one of the questions the proof of concept answers.
Then you have learned it for the price of a proof of concept instead of a full build, with the results measured on your own data. Often the answer is partial: the AI handles most cases and the rest go to a person for review, and the proof of concept shows where that line sits.
Have an AI idea
you want to test?
30-minute call with a senior Solution Architect. Bring the idea and a description of your data, and we will tell you whether a proof of concept is worth doing.