Artificial Intelligence FAQ

Honest answers to the questions clients ask most often. If yours is not listed, ring us on +44 845 390 6877 or email [email protected].

Common questions

Discovery calls are free. Data audits typically range from £800 to £2,000 depending on how many data sources we need to review. Prototype builds run between £5,000 and £25,000: a straightforward demand-forecast model sits at the lower end, while a multi-field document extraction pipeline with OCR pre-processing sits at the upper end.

Deployment adds £3,000 to £10,000 and includes 90 days of monitoring and bug-fix support. Every fee is fixed and quoted before work begins, so there are no surprise invoices.

Most projects deliver a testable prototype within four to six weeks of the signed proposal. Simpler models can be ready in two weeks. Complex pipelines with many field types or multiple languages may take up to eight weeks.

The biggest variable is usually data preparation. If your data arrives clean and well-labelled, the timeline shrinks. If we need to help you build a labelling process first, add one to two weeks.

It depends on the task. A time-series forecast usually needs at least 12 months of daily observations to capture seasonal patterns. Document extraction can work with as few as 200 labelled examples if the layout is consistent across documents.

For classification tasks, like routing support tickets into categories, we generally want at least 500 labelled examples per category. We assess your data during the audit phase and tell you honestly whether it is enough.

Yes. We hold Cyber Essentials Plus certification. All data is transferred over TLS 1.3 and stored encrypted at rest on UK-based cloud servers. We sign a data processing agreement before receiving any personal or commercially sensitive data.

Access is restricted to the engineers assigned to your project. When the project ends, we delete your data within 30 days unless you ask us to retain it for future retraining.

We agree a specific performance metric and target value in the scoping document before any engineering starts. If the prototype does not hit that target after the iteration round, you pay nothing for the prototype phase. You keep the audit report and scope document regardless.

This has happened twice in 47 projects. In both cases the root cause was insufficient training data, which we flagged as a risk during the audit. We offered to pause and resume once the client had collected more examples.

Almost certainly. We deploy models as containerised REST APIs, which means any system that can make an HTTP request can consume the output. We have integrated with .NET backends, Node.js services, Python Flask apps, Salesforce via middleware, and even legacy VB6 applications through a thin proxy layer.

If you use Azure, AWS or Google Cloud, we deploy directly into your account. If you prefer not to manage infrastructure, we host the model on our own UK-based Kubernetes cluster and give you an API key.

We do. Our maintenance plan costs a flat monthly fee, typically between £400 and £1,200 depending on model complexity and retraining frequency. The plan covers monitoring, drift detection alerts, quarterly retraining and priority bug fixes.

About 60% of our clients choose this option. The rest prefer to bring maintenance in-house. We provide full documentation and a handover session to make that transition straightforward.

Our client base spans recruitment, logistics, healthcare administration, retail and financial services. The common factor is not the industry but the problem shape: you have structured or semi-structured data, a decision that currently requires human judgment, and enough volume to justify automating that decision.

We have turned down projects in sectors where we lacked domain knowledge and could not partner with a subject-matter expert. Honesty about our limits saves everyone time.

What clients say

Short excerpts from recent project reviews.

"We gave them 14 months of delivery data and got back a forecast model that cut our over-ordering by 18% in the first quarter. The weekly progress emails meant we were never guessing about status."

Rachel Dunne, operations lead at a Belfast food distributor

"Our HR team was spending two full days a week sorting CVs. The classifier they built routes candidates to the right recruiter automatically. Accuracy after six months is still above 91%."

Tom Hargreaves, director at a Manchester recruitment agency

"I appreciated that they told us up front our data set was too small for one of the three features we wanted. They suggested we collect another three months of examples before attempting it, which saved us from paying for a model that would not have worked."

Sienna Kapoor, CTO at a London fintech startup
Data visualisation dashboard on a laptop with coffee and notebook on a wooden desk