Artificial Intelligence FAQ

A proof-of-concept starts at £4,500. That covers the discovery call, data audit, model training and a working prototype you can test with real scenarios. Full production projects typically range from £12,000 to £45,000. The price depends on how many data sources need connecting, whether you need a custom front-end dashboard and whether you opt for ongoing monitoring.

We quote fixed prices for the PoC and time-and-materials for production work, capped at an agreed maximum. There are no hidden licence fees for models we build from scratch. If we use a third-party API (for example, a large language model provider), we pass through the API cost at our negotiated rate with no markup.

It depends on the task. For tabular prediction problems like sales forecasting or churn scoring, 2,000 rows of clean historical data is usually enough to build a useful first model. More data improves accuracy, but we can start small and iterate.

For NLP tasks such as document classification or entity extraction, 500 labelled examples often produce a model accurate enough for a proof of concept. We can help you label the first batch during the discovery phase if your data is not annotated yet.

If you genuinely have very little data, we sometimes use transfer learning: we fine-tune a pre-trained model on your small dataset rather than training from zero. This works especially well for text and image tasks.

Most projects move from the first discovery call to a live production system in 8 to 14 weeks. The proof-of-concept phase takes three to four weeks. After that, production deployment adds another four to eight weeks depending on how many integrations are needed and how quickly your IT team can provide API access.

The single biggest delay we see is waiting for data access. If you can share a sample dataset before the PoC officially starts, we can shave a week or two off the timeline.

Your data stays in the UK. We use cloud infrastructure hosted in London-region data centres, and all data is encrypted at rest (AES-256) and in transit (TLS 1.3). We never use client data to train models for other clients.

Before any project begins, we sign a data-processing agreement that specifies retention periods, access controls and deletion procedures. At the end of an engagement, we delete all copies of your data within 30 days unless you ask us to keep them for ongoing support.

If your compliance requirements demand on-premise deployment, we can package models as Docker containers that run entirely within your own network. The model never phones home.

That is exactly why we start with a paid proof of concept rather than jumping straight to production. During the PoC, we agree on specific accuracy thresholds with you. If the model cannot meet those thresholds after reasonable tuning, you are free to walk away. You pay only the PoC fee and keep the evaluation report.

In practice, about one in eight PoCs does not clear the bar. The usual reason is data quality: missing fields, inconsistent labelling or too little historical depth. When that happens, we provide a written recommendation on what data to collect and how to structure it so a future attempt has a better chance.

Yes. We have built integrations with Xero, Sage, SAP Business One, Salesforce, HubSpot, Shopify and several bespoke ERP systems written in .NET and Java. If your system exposes a REST or SOAP API, we can connect to it. If it does not, we can often pull data via scheduled database exports or SFTP file drops.

We deliver our models as containerised microservices with a documented REST API, so your developers can call them from any language or platform. We also provide a simple web dashboard for non-technical users who just need to see results.

We do. Our standard support plan includes automated accuracy monitoring, retraining when performance drifts by more than 3 percentage points, priority bug fixes and a monthly review call. The plan costs a flat monthly fee agreed at contract signing, typically between £800 and £2,400 depending on the number of models in production.

Quarterly, we run a half-day workshop with your team to review model behaviour, discuss edge cases and plan any feature additions. These workshops are included in the support plan at no extra charge.

You do. All custom code, trained model weights and documentation become your intellectual property upon final payment. We retain no licence to use your models or data for any other purpose.

We sometimes use open-source libraries and frameworks (scikit-learn, PyTorch, FastAPI) as building blocks. Those remain under their original open-source licences, which allow commercial use. We will list every dependency in the project handover documentation so your team knows exactly what is under the hood.