Data readiness audit
£4,500 fixed
- Two-week turnaround
- Up to 5 data sources
- Written feasibility report
- One follow-up call
What each service includes
Every engagement begins with a paid two-day discovery workshop. We charge this separately so you get an honest feasibility report before committing to a full build.
Predictive analytics
We train supervised-learning models on your historical data to forecast outcomes you care about: weekly sales volume, equipment failure windows, customer churn probability, inventory shortfalls. The work starts with a data audit where we assess completeness, label quality and potential bias.
After the prototype passes evaluation against agreed accuracy thresholds, we package the model as a containerised REST API, write integration tests and deploy it to your cloud environment. A Grafana monitoring dashboard tracks prediction drift, latency and input-data health in real time.
- Data audit and feature-engineering report
- Trained model with versioned artefacts
- REST API with OpenAPI documentation
- Grafana dashboard and alerting rules
- Three-month post-launch support
Natural-language processing
Text is everywhere in a business: support tickets, contracts, survey responses, internal wikis. We fine-tune transformer-based language models on your domain vocabulary so they understand the jargon your staff and customers actually use.
Common deliverables include multi-label ticket classifiers, extractive and abstractive summarisers, sentiment analysers calibrated to your product categories, and named-entity recognisers trained on your specific entity types (part numbers, policy codes, medication names). We handle the full annotation pipeline, either with your subject-matter experts or through a managed labelling team we have worked with since 2021.
- Annotation guidelines and labelled dataset
- Fine-tuned language model (hosted or on-premise)
- Batch and real-time inference endpoints
- Evaluation report with precision, recall and F1 per class
- Retraining script and schedule recommendations
Computer vision
We build image and video analysis systems for manufacturing inspection, retail analytics, agricultural monitoring and security applications. Our engineers work with convolutional neural networks and vision transformers depending on the accuracy and speed trade-offs your use case demands.
For clients who cannot send image data to public cloud providers, we deploy inference on our on-premise GPU cluster or on edge devices (NVIDIA Jetson, Intel NUC) installed at your facility. Each project includes a retraining toolkit so your team can update the model when product lines change or new defect types appear.
- Image collection and augmentation pipeline
- Trained detection or classification model
- Edge or cloud deployment with TensorRT optimisation
- Retraining scripts and data-versioning setup
- On-site installation support where applicable
Data readiness audit
Not sure if your data is good enough for machine learning? This standalone two-week engagement answers that question before you spend anything on model development. We connect to your databases, warehouses or flat-file exports and produce a written report covering volume, completeness, label distribution, class imbalance and potential leakage risks.
The report ends with a clear recommendation: proceed to modelling, invest in data collection first, or solve the problem with simpler methods. About 30 per cent of audits conclude that the client does not yet have enough labelled examples, and we provide a concrete plan for closing that gap.
- Secure data-access setup and NDA
- Exploratory data analysis notebook
- Written feasibility report (typically 15–25 pages)
- Go/no-go recommendation with next steps
Model monitoring and retainer
Production models degrade. Input distributions shift, upstream schemas change, and business definitions evolve. Our monitoring retainer catches these problems before they affect your bottom line.
We set up automated drift-detection pipelines that compare live prediction distributions against your baseline. When drift exceeds agreed thresholds, our engineers investigate, retrain and redeploy. The retainer includes a fixed number of retraining cycles per quarter, a monthly performance summary and a dedicated Slack or Teams channel for ad-hoc questions.
- Drift-detection pipeline (Evidently or custom)
- Quarterly retraining cycles (number agreed at sign-up)
- Monthly performance report
- Dedicated communication channel
- Priority response for production incidents
Indicative pricing
Final quotes depend on data volume, model complexity and deployment environment. These ranges cover most mid-size engagements.
Full model build
£12k–£45k project
- Discovery workshop included
- Prototype in 4 weeks
- Production deployment
- 3-month support window
- Knowledge-transfer workshop
Monitoring retainer
£1,800 / month
- Drift detection pipeline
- Up to 2 retraining cycles / quarter
- Monthly performance report
- Dedicated Slack channel
Frequently asked questions
Most projects run between six and fourteen weeks from data handover to production deployment. The four-week prototype phase is the core of the work; the remaining time covers integration testing, stakeholder review rounds and the knowledge-transfer workshop. Simpler classification tasks with clean labelled data can ship faster. Complex multi-model pipelines or projects requiring new data collection take longer.
Yes, model training requires access to representative data. We sign an NDA and a data-processing agreement before any transfer. For clients with strict residency requirements, we can work inside your VPN or on air-gapped infrastructure. We also support federated-learning setups where the data never leaves your premises and only model gradients are exchanged.
That is exactly what the data readiness audit is for. We assess what you have, identify gaps and give you a concrete plan to fill them. Sometimes that means collecting more labelled examples, sometimes it means fixing upstream ETL pipelines, and occasionally it means the project should wait six months. We would rather tell you that upfront than build a model on shaky foundations.
We prefer it. Our best outcomes happen when one or two of your engineers pair with our ML team throughout the build. They learn the model internals, review pull requests and take ownership of the system after handover. The knowledge-transfer workshop at the end formalises that, but the real learning happens during the project itself.
We deploy on AWS, Google Cloud and Azure. Most of our tooling is cloud-agnostic: Docker containers, Kubernetes manifests and Terraform modules that work across providers. If you run on-premise infrastructure, we can deploy there too, provided the hardware meets the minimum GPU and memory specs for the model in question.