Our Artificial Intelligence services

Below is a detailed look at what we deliver, how each engagement works, and what it costs. Every project follows the same five-step process, but the scope and timeline vary depending on your data and goals.

Service details

We organise our work into four service lines. Most clients start with one and add others as they see results.

Data scientist reviewing time-series charts on a curved monitor

Predictive modelling

We train supervised-learning models, primarily gradient-boosted trees and neural networks, on your historical data. The output is a production-ready prediction endpoint your systems can query via a REST API or a batch job.

Typical use cases include demand forecasting for retail and wholesale, customer churn prediction for subscription businesses, and equipment failure prediction for manufacturing lines. A logistics company we worked with in Edinburgh reduced late deliveries by 22% after deploying a route-delay model we built on 14 months of GPS and weather data.

Turnaround from data handover to a validated prototype: four to six weeks. Full production deployment, including monitoring dashboards, adds another two to three weeks.

RegressionClassificationTime series
Text analysis interface displayed on a tablet

Natural language processing

Text data piles up fast: support tickets, survey responses, contract clauses, social media mentions. We build NLP pipelines that classify, extract entities from, and summarise these documents automatically.

Our standard approach uses fine-tuned transformer models running on your own infrastructure, so sensitive text never leaves your network. For a Glasgow-based insurance firm, we built a claims-description classifier that routes incoming claims to the correct department with 92% accuracy, cutting average handling time from 11 minutes to under three.

We also build chatbot back-ends. Not the scripted decision-tree kind: retrieval-augmented generation systems that answer questions using your internal knowledge base. Setup takes roughly six weeks, including a two-week testing phase with real users.

Text classificationEntity extractionSummarisationRAG chatbots
Industrial camera inspecting products on a conveyor belt

Computer vision

If your problem involves images or video, we train convolutional and vision-transformer models for object detection, defect inspection and image classification. A food-processing plant near Aberdeen uses our system to flag packaging defects on a line running at 120 units per minute. The false-positive rate sits below 1.5%.

We handle the full pipeline: image collection guidance, annotation tooling, model training, edge deployment on devices like NVIDIA Jetson, and a web dashboard for reviewing flagged items. For projects with fewer than 2,000 labelled images, we use transfer learning from pre-trained models to keep accuracy high without months of data collection.

Object detectionDefect inspectionImage classification
Server room with fibre optic cables and indicator lights

Data pipeline automation

Before any model can work, the data feeding it needs to be clean, timely and structured. We build ETL pipelines using Apache Airflow, dbt and Python that pull from your existing databases, APIs and file shares, then transform and load the results into a warehouse or a model-ready format.

One healthcare client had clinical trial data arriving in 14 different CSV formats from partner sites across Europe. Our pipeline normalises, validates and merges those files overnight, flagging anomalies for a human reviewer by 8 a.m. each morning. What used to take a data analyst two full days now runs in 40 minutes.

We also set up monitoring alerts: if a source stops sending data, if row counts deviate by more than 10% from the rolling average, or if schema changes break a transformation, your team gets notified immediately.

ETLAirflowData qualityMonitoring

How every project runs

Five steps, the same every time, so you always know where things stand.

1. Scoping call

A 30-minute video call where we learn about your data, your goal and your constraints. No charge, no obligation. By the end we can tell you whether AI is the right approach and give a rough timeline.

2. Discovery phase

A fixed-price engagement, usually one to two weeks, where we audit your data, identify gaps and produce a written project plan with deliverables, milestones and costs. You own the plan whether or not you proceed.

3. Build and train

Our engineers build the pipeline and train models iteratively. You get a demo at the end of each week so you can course-correct early. Typical duration: four to eight weeks.

4. Testing and deployment

We validate the system against held-out data and deploy it to your infrastructure, whether that is a cloud instance, on-premise servers or edge devices. We write runbooks and train your team.

5. Support and iteration

After launch we monitor performance for 90 days at no extra cost. If model accuracy drifts or data patterns shift, we retrain. Beyond the initial period, support plans start at £800 per month.

Indicative pricing

Every project is quoted individually after the discovery phase, but these ranges give you a starting point.

Discovery

£2,400 fixed
  • Data audit and gap analysis
  • Written project plan
  • Go/no-go recommendation
  • 1–2 week delivery
Get started

Ongoing support

£800+ /month
  • Model monitoring and retraining
  • Pipeline maintenance
  • Priority email and phone support
  • Quarterly performance reports
Learn more

Frequently asked questions

Answers to the things clients ask most often during the scoping call.

It depends on the task. For tabular prediction, a few thousand rows is often enough. Image classification typically needs at least 500 labelled images per category, though transfer learning can reduce that. During the scoping call we review what you have and tell you honestly whether it is sufficient.

Yes. About 40% of our clients are based elsewhere in the UK, and we have completed projects for companies in Ireland and the Netherlands. All communication and demos happen over video call, and we travel for on-site workshops when the project warrants it.

You do. From day one, all code lives in a repository you control. On project completion we transfer any remaining assets and documentation. There are no licence fees or lock-in clauses.

We agree on performance targets during the discovery phase. If the prototype does not meet those targets after a reasonable iteration cycle, we discuss options: collecting more data, adjusting the target, or stopping the project. You only pay for work completed up to that point.

Absolutely. We regularly work inside client VPNs and on-premise environments. For healthcare and financial-services clients this is the default arrangement. We bring our tools to your data, not the other way around.