Artificial Intelligence FAQ

It depends on scope, but most projects follow a pattern. A focused proof-of-concept, such as a churn-prediction model trained on your CRM data, usually takes three to five weeks from kick-off to a working demo. A production-grade system with API integration, monitoring and retraining pipelines typically runs eight to fourteen weeks. We break every project into two-week sprints with a demo at the end of each, so you can see progress in real time and adjust priorities along the way.

We price by milestone, not by the hour. A discovery workshop and feasibility report starts at £2,400. Proof-of-concept builds range from £8,000 to £18,000 depending on the complexity of the data and the model architecture required. Full production deployments are scoped individually, but the majority land between £25,000 and £75,000. We share a detailed cost breakdown before any commitment, and the price doesn't change unless the scope does.

Often, yes, because the model needs to learn from real examples. We sign a mutual NDA before any data transfer and can work within your own infrastructure if that's preferable. For clients in regulated sectors like finance or healthcare, we routinely use anonymised or synthetic datasets during early development and only introduce real data in a controlled environment with strict access logging. We are registered with the ICO and comply with UK GDPR requirements.

We set a measurable accuracy or performance threshold at the start of every project. If the proof-of-concept doesn't meet that threshold, we investigate why: sometimes the issue is data quality, sometimes it's a modelling choice, sometimes the problem itself isn't well suited to machine learning. In that last case, we'll tell you directly and recommend an alternative approach. You won't be billed for work that doesn't deliver value, because our milestone payments are tied to agreed deliverables, not time spent.

Our client base spans retail, logistics, financial services, manufacturing and professional services. The common thread isn't the industry but the problem type: if you have structured or semi-structured data and a decision that currently relies on manual judgement or simple rules, there's a good chance we can improve it. We've built demand-forecasting models for a Welsh food distributor, fraud-detection systems for a London fintech, and defect-classification pipelines for a Midlands automotive parts supplier.

Yes. We deploy on AWS, Azure and Google Cloud, and we've integrated with on-premise servers running everything from PostgreSQL to legacy COBOL systems. Our models are typically containerised with Docker, which means they can run wherever your infrastructure allows. If you have a preferred CI/CD pipeline or monitoring tool, we'll adapt to it rather than forcing our own preferences.

We do. Machine learning models degrade over time as the real world shifts, a phenomenon called data drift. Our maintenance plans include monthly performance checks, automated drift alerts and scheduled retraining runs. Plans start at £800 per month for a single model and scale based on the number of models and the frequency of retraining required. Clients on a maintenance plan also get priority access to our engineering team for ad-hoc questions.

Size, mostly. We're a team of nine, so the people who scope the project are the same people who write the code and present the results. There's no handoff to a junior bench. That also means we're selective about the projects we take on: we typically run two to four engagements at a time, which keeps quality high and communication fast. You'll have a named lead engineer who responds to emails within a working day.