Questions about working with us

Every project is fixed-price, agreed in writing before work begins. There are no hourly rates and no billing surprises.

Simpler automations — a single regulatory source, a straightforward clause-extraction pipeline — typically start from £2,000–£5,000. More complex work, such as multi-jurisdiction regulatory extraction, full M&A pack automation, or bespoke AI agents, typically ranges from £5,000 to £25,000 depending on scope, data complexity, and the number of sources involved.

We provide a written scope document with a fixed price after a free scoping conversation. You approve the scope before work starts.

Most projects are delivered within 2–4 weeks from scope sign-off. Simpler automations — a single-source extraction pipeline, a straightforward AI agent — can be built and deployed in under two weeks. Larger multi-source pipelines or complex agent workflows may take 4–6 weeks.

The delivery timeline is agreed in the scope document before work starts and is treated as a commitment, not an estimate.

Yes — once the system is in production, we operate it on your behalf. There is a fixed monthly retainer covering hosting, monitoring, maintenance, and minor changes. This is sized during scoping so you know the running cost before the project starts.

Infrastructure costs — cloud hosting, API calls to third-party services such as Azure OpenAI — are typically rolled into the retainer for a single predictable invoice, or paid directly by you to those providers if you prefer. No per-seat fees, no per-document licensing, no surprise increases.

We treat this as a hard requirement, not a bolt-on. In practice:

• We sign NDAs as standard before any documents are shared.
• Client documents and data are never used to train AI models.
• All data handling is under UK GDPR. Full data processing documentation is provided for your compliance and information security records.
• We can work within air-gapped environments, on-premises deployment, or your firm's existing Azure / AWS tenancy if required.
• We work within your firm's information security policies and can complete standard supplier questionnaires.

Yes. UK AI Automation (DataNorfolk Limited) is registered with the Information Commissioner's Office (ICO) as a data controller. Compliance with UK GDPR and the Data Protection Act 2018 is built into every project from the start.

We provide full data processing documentation, suitable for submission to your firm's information security and risk team. ICO registration number available on request.

DataNorfolk Limited holds £5 million professional indemnity insurance and public liability insurance. A certificate of insurance is available on request for procurement and supplier approval processes.

We use the appropriate tool for the task. The most common stack for legal document work:

• Document extraction: Azure Document Intelligence, OCR pipelines, GPT-4 class models via Azure OpenAI (data stays within Azure, no training on your data).
• Clause extraction and classification: Fine-tuned LLM pipelines or RAG (retrieval-augmented generation) over your document corpus.
• Regulatory filing parsing: Bespoke parsers for SEC EDGAR, Companies House, FCA, Legifrance, AMF, and others — built to handle each source's specific structure and encoding.
• Backend infrastructure: .NET/C#, Azure Functions, Azure Service Bus, SQL Server.

The tool choice is driven by accuracy requirements, data sensitivity constraints, and running cost — not by a preference for any particular vendor.

In most cases, yes. We build integrations against REST APIs, SQL databases, SharePoint, document management systems (iManage, NetDocuments, HighQ), Azure Blob Storage, and most enterprise data stores.

If your system has an API or accessible data store, we can almost certainly connect to it. If it does not, we can discuss what data export options exist.

Accuracy depends on the task. For structured regulatory filings with consistent formats, extraction accuracy can be very high — effectively 100% on well-formed documents. For more variable documents such as contracts or unstructured correspondence, we calibrate during the scoping phase using sample documents, and we are direct about what level of accuracy is realistic.

We design every pipeline to flag uncertain extractions for human review rather than silently returning low-confidence results. The goal is to reduce analyst time, not to create a new accuracy problem.

We retain ownership of the source code and operate the system on your behalf. You own all extracted data, all outputs, and all integrations into your systems — this is written into our standard engagement terms.

This is how managed AI and document-automation engagements typically work. It keeps your costs predictable (no per-seat or per-document licensing) and lets us continue maintaining, improving, and re-tuning the system over time without re-quoting every change.

Outputs are delivered in standard formats — JSON, CSV, REST APIs into your existing systems — so your data is never trapped. If you ever wanted to move providers, your downstream workflows would not need to be rebuilt.

Foundational pipeline patterns, utility libraries, and infrastructure components are reused across engagements — that is how we keep delivery times short and prices fixed.

Anything specific to your firm — your prompts, your schemas, your business logic, your data, your output formats — is exclusive to you and never shared with another client.

It starts with a free, no-obligation conversation — typically 30–60 minutes — in which we discuss the workflow, the data sources, the volume, and the output format you need. We may ask to see sample documents (under NDA) to calibrate what is technically realistic.

Following that, we provide a written scope document setting out exactly what will be built, at a fixed price, with a delivery timeline. You review and approve the scope before any work begins. There is no pressure to proceed.

We'll tell you. There is no benefit to selling a project that won't deliver real value. If a workflow has characteristics that make automation genuinely difficult — highly variable documents with no consistent structure, volumes too low to justify the build cost, or accuracy requirements that current AI cannot reliably meet — we will say so clearly during scoping.

A short honest conversation is worth more to both sides than a project that under-delivers.