Artificial Intelligence FAQ

Honest answers to the questions prospective clients ask us most often.

General questions

Most engagements run between eight and fourteen weeks from the start of discovery to a production-ready deployment. Discovery itself takes roughly one week and results in a scope document, a data audit and a fixed-price quote. Model development and testing usually occupy four to six weeks. Integration, load testing and handover fill the remaining time. Larger projects with multiple models or complex data pipelines can extend to five or six months, but we break those into phases so you see working software after each one.

It depends on the task. For text classification we can often achieve useful accuracy with a few hundred labelled examples, especially when we fine-tune a pre-trained language model. Image-based tasks like defect detection usually need a few thousand annotated images. If you have raw data but no labels, we can set up an annotation workflow using Prodigy or Label Studio and train annotators from your team. We have also built synthetic-data generators for clients whose real data was too scarce or too sensitive to use directly.

Discovery is a flat £1,200. During that week we review your existing data, interview stakeholders, identify the highest-impact use case and produce a written scope document with a binding project quote. If you decide not to proceed, you keep the scope document and owe nothing further. About 85 % of discovery clients move forward to a full project.

Yes. Once final payment clears, all intellectual property transfers to you. That includes trained model weights, training scripts, data-processing pipelines and deployment configurations. We retain no licence to use your data or your models. If we built on top of an open-source framework (PyTorch, scikit-learn, Hugging Face Transformers, etc.) the framework itself remains under its original open-source licence, but everything we wrote for you is yours.

Absolutely. We deploy to UK-based cloud regions (AWS eu-west-2, Azure UK South, GCP europe-west2) by default. For clients with strict on-premises requirements, we can train and serve models on your own servers. We have done this for two NHS-adjacent organisations and a financial-services firm that needed to keep patient and customer data behind their own firewalls at all times.

Technical questions

Python is our primary language. For deep learning we use PyTorch; for tabular data, scikit-learn and XGBoost. NLP projects lean on Hugging Face Transformers. Deployment usually involves FastAPI or Flask behind an Nginx reverse proxy, containerised with Docker and orchestrated by Kubernetes or AWS ECS. We write infrastructure-as-code in Terraform. If your existing stack is different and you need the model to integrate as a library rather than a service, we can package it as a pip-installable wheel or a .NET-compatible ONNX runtime.

Every deployment ships with a monitoring dashboard (Grafana or a custom panel) that tracks prediction distributions, latency and key accuracy metrics. We set alert thresholds during the scope phase. When metrics breach those thresholds, an automated pipeline re-trains the model on fresh data and runs a validation suite. If the new model passes, it is promoted to production via a blue-green deployment. During the 90-day support window, we manage this pipeline. After that, your team can run it themselves or we can provide ongoing maintenance under a separate agreement.

The smallest engagement we take on is the £1,200 discovery phase. Full projects typically start around £8,000 for a straightforward classification or regression model with a single data source. Complex multi-model systems, real-time pipelines or projects requiring significant data engineering tend to fall in the £25,000 to £60,000 range. We do not charge by the hour, so you always know the total cost before signing.

We design every model to be consumed through a REST or gRPC API, which means any system that can make an HTTP call can use it. We have integrated with Salesforce, SAP, custom Django applications, legacy .NET services and several in-house ERPs. During discovery we map out the integration points and include the necessary adapter code in the project scope.

Still have questions?

If your question is not covered here, send us a message. We respond within one working day, and there is no charge for an initial conversation about your project.

You can reach us by email at [email protected] or by phone on 01557 488835.

Go to contact form

Team discussion about an AI project at Ai Accord