Artificial Intelligence FAQ

It depends on the scope, but most engagements fall between four and twelve weeks. A straightforward classification model trained on a clean, labelled dataset can reach production in about a month. Projects that require data collection, cleaning, or integration with legacy systems tend to sit closer to the twelve-week mark. We give you a week-by-week schedule before any contract is signed, so there are no surprises.

Not necessarily. For some tasks, a few hundred well-labelled examples are enough to train a useful model, especially when we fine-tune a pre-trained foundation model rather than training from scratch. If your data is sparse, we can help you set up a collection pipeline or use techniques like data augmentation and synthetic data generation to fill the gaps. We will tell you honestly during the audit phase whether the data you have is sufficient or whether you should invest in gathering more before spending money on model development.

Our past projects span logistics, retail, healthcare administration, and financial services. That said, the techniques we use transfer across sectors. A demand-forecasting model for a retailer and a patient-flow prediction model for an NHS trust share the same underlying time-series methods. What matters more than industry is whether you have a clear business question and data that relates to it.

Yes. We deploy on AWS, Azure, or Google Cloud, and we have experience integrating with on-premises servers for clients with strict data-residency requirements. If your application layer is in Java, .NET, or Node.js, we expose our models as REST APIs with clear documentation so your developers can call them without learning Python. We also support containerised deployments via Docker and Kubernetes if that is what your ops team prefers.

We offer two pricing models. The first is a fixed-fee engagement: we scope the work, agree on deliverables, and quote a total price. You pay in milestones tied to specific outputs, not hours logged. The second option is a time-and-materials retainer, which suits ongoing advisory work or projects where the scope genuinely cannot be pinned down in advance. In either case, we share a detailed cost breakdown before you commit.

We do not enforce a hard minimum, but in practice our smallest engagements start at around two weeks of work. Anything shorter usually means the problem is simple enough that an off-the-shelf tool will serve you better, and we would rather point you to that tool than charge consulting fees for something you do not need.

Every project includes a 30-day support window after deployment. During that window we fix bugs, tune thresholds, and answer questions from your team at no extra cost. Beyond that, you can opt into a monthly maintenance plan that covers model monitoring, retraining on new data, and performance reporting. Many clients choose this because model accuracy drifts over time as real-world patterns change.

All client data stays within the UK or EU by default. We use encrypted storage on AWS (London region) or Azure (UK South), depending on which cloud provider you already use. If your compliance requirements mandate on-premises storage, we can work with that too. We sign a data-processing agreement before any data is transferred, and we never use client data to train models for other clients.

We have worked with datasets containing personal health information, financial records, and customer contact details. In each case we follow UK GDPR requirements, apply pseudonymisation or anonymisation where possible, and restrict access to the specific team members assigned to your project. If your dataset includes special-category data under the GDPR, we conduct a data-protection impact assessment before processing begins.

You do. Our standard contract assigns all intellectual property created during the engagement to you upon final payment. That includes trained model weights, training scripts, data pipelines, and documentation. We retain no licence to use your model or data after the project ends. The only exception is generic utility code that we use across projects (logging wrappers, deployment templates), which remains ours but is freely licensed to you.