Common questions
Discovery and prototyping are fixed-price. For most projects that means between £4,000 and £12,000, depending on data complexity and the type of model. A straightforward churn-prediction model on clean tabular data sits at the lower end. A multi-class computer vision system with custom annotation sits at the higher end.
After deployment, ongoing monitoring and quarterly retraining start at £800 per month. We quote exact figures after the discovery call so there are no surprises.
It depends entirely on the task. Tabular predictive models generally need a few thousand rows at minimum. If you have fewer, we can sometimes supplement with synthetic data generation or Bayesian approaches that handle small samples better than deep learning.
Computer vision projects can begin with as few as 200 labelled images when we use transfer learning from a pre-trained backbone like ResNet or EfficientNet. For natural language tasks, the threshold is similar: a couple of hundred annotated examples can be enough if we fine-tune a large language model.
We sign a mutual NDA before you share anything. All data transfers happen over TLS 1.3 encrypted channels. When we work on our own infrastructure, data is stored in AES-256-encrypted volumes on UK-based servers.
Whenever possible we work directly on your infrastructure via a VPN or SSH tunnel so the data never leaves your environment. After project completion we delete all local copies within 14 days and send you written confirmation.
Our completed projects span logistics, healthcare, retail, manufacturing and financial services. The common thread is structured or semi-structured data and a decision that currently relies on human judgement or manual rules. If your industry is not on that list, it does not mean we cannot help. The discovery call exists precisely to figure that out.
Yes. Every line of code we write during your project belongs to you from the moment we write it. That includes training scripts, inference pipelines, deployment configurations and documentation. We do not retain any proprietary rights. If you decide to part ways after deployment, you can hand the codebase to another team and they will be able to maintain it.
During the prototype phase we agree on a performance threshold together. If the model does not reach it after the agreed iteration cycles, we present the results honestly and discuss options: collecting more data, simplifying the problem scope, or stopping the project. We have never charged a client for work that did not deliver measurable value. Two projects out of 34 ended at the prototype stage because the available data could not support the desired accuracy, and we refunded the iteration fees in both cases.
In most cases, yes. We package models as REST APIs that any system can call over HTTP. We have integrated with Salesforce, SAP, custom Django applications, .NET back-ends and Excel-based workflows (via a lightweight Python bridge). If your software can make an HTTP request or read from a database, we can connect the model to it.
45 minutes, and yes, genuinely free. We do not ask for payment details before the call. About 20 per cent of discovery calls end with us recommending a non-AI solution. We would rather point you in the right direction than sell you something you do not need.
Topics people also ask about
Data privacy and the UK GDPR
If your data includes personal information, we follow UK GDPR requirements throughout the project. That means data minimisation (we only use the fields the model actually needs), pseudonymisation where feasible, and a clear data-processing agreement. We can also help you prepare a Data Protection Impact Assessment if the ICO requires one for your use case.
Explainability and bias
Black-box predictions are hard to trust, especially in regulated industries. We default to interpretable model families (gradient-boosted trees, linear models with interaction terms) and add SHAP-based explanations when we use neural networks. Before deployment we run bias audits across protected characteristics and document the results in a model card you can share with your compliance team.
Cloud versus on-premise
We deploy wherever your data lives. Cloud deployments (AWS, Azure, GCP) are fastest to set up and easiest to scale. On-premise deployments take longer but keep data entirely within your network. We have also deployed models on edge hardware for factory-floor use cases where internet connectivity is unreliable.
Team training
After deployment we offer optional half-day training sessions for your engineers. The goal is to make your team self-sufficient for day-to-day monitoring and minor adjustments. We cover the model architecture, the retraining pipeline and common failure modes. The session costs £600 and includes printed reference materials.
Still have questions?
Drop us a message and we will get back to you within one working day. No sales pitch, just a direct answer.
Contact us