Frequently asked questions about Artificial Intelligence

Most projects run between six and fourteen weeks from the initial scoping call to a deployed, monitored model. The biggest variable is data preparation. If your data is already in a clean, structured format, we can move straight to feature engineering and training, which cuts the timeline by two to three weeks. An AI readiness audit is faster: two days on-site plus roughly a week for us to write and deliver the report.

We work with structured data (CSV, database tables, spreadsheets), semi-structured data (JSON, XML, log files) and unstructured data (images, PDFs, free-text fields). The minimum volume depends on the problem. For a classification task we usually want at least a few thousand labelled examples per category. For time-series forecasting, two years of daily records is a good starting point. During the scoping call we will tell you exactly what we need and whether your current data is sufficient.

An AI readiness audit starts at £3,500. Custom model training projects typically range from £12,000 to £60,000. The price depends on complexity: a single demand-forecasting model for one product line sits at the lower end, while a multi-model computer-vision pipeline with edge deployment sits at the higher end. We quote fixed prices per project after the scoping call so there are no surprises. Cloud compute costs during training are included in the quote.

Yes. We offer quarterly monitoring and retraining packages. These cover model-drift detection, data-pipeline maintenance and up to four retraining cycles per year. If your model's accuracy drops below an agreed threshold, we retrain on fresh data and redeploy within five working days. The quarterly package starts at £2,200. Clients who do not want a package can request ad-hoc retraining at our standard day rate.

In our experience, no. The models we build handle repetitive, high-volume tasks: scanning thousands of invoices, ranking leads by conversion probability, flagging anomalous sensor readings. Your staff then spend their time on the decisions that require context, judgement and customer relationships. One logistics client freed up 14 hours per week of analyst time by automating route-optimisation scoring. Those analysts now focus on supplier negotiations, which the model cannot do.

We sign a data-processing agreement before any data transfer. All data is encrypted in transit (TLS 1.3) and at rest (AES-256). We use UK-based cloud infrastructure unless you specify otherwise. Access is restricted to the project team, and we delete all client data within 30 days of project completion unless you ask us to retain it for future retraining. We carry professional indemnity insurance and are registered with the ICO.

Almost certainly. We have deployed models into environments running on AWS, Azure, GCP, on-premise Linux servers and even Raspberry Pi clusters at the edge. Our standard delivery format is a Docker container with a REST API, which slots into most architectures. If you use a specific orchestration tool like Kubernetes, Airflow or Prefect, we configure the deployment to match. During the scoping call we map out your current infrastructure so there are no integration headaches later.

We agree on success criteria before training starts: a target accuracy, precision, recall or business metric such as cost saved per month. If the model does not meet those criteria after two rounds of iteration, we discuss options honestly. Sometimes the issue is data quality, and collecting more labelled examples solves it. Sometimes the problem is not well suited to the approach we chose, and we pivot to a different algorithm. We have walked away from two projects in seven years because the data simply could not support the desired outcome, and in both cases we refunded the training fee.