Skip to content
AI Private LimitedAI Advisory & Delivery

AI Consulting

Find the two or three things worth building. Then build them properly.

Most businesses do not have an AI problem. They have a process nobody wrote down, in a system nobody trusts, run by someone who has been quietly compensating for it for years. AI is sometimes the right fix for that, and often it is not.

Book a free 30-minute call

Two levels

Some of you need a map. Some of you need the machinery.

The first engagement tells you what to build and what to leave alone. The second is for teams who already know, and need retrieval, tools and agents that hold up under real traffic and real documents. Plenty of clients start with the first and move to the second once the roadmap is signed off.

01

Typically 1–3 weeks. Fixed price, quoted after a free scoping call.

AI Consulting

A short, sharp engagement that turns 'we should be doing something with AI' into a costed roadmap with two or three things worth building first, and a clear list of what not to bother with.

  • What you receive
  • Opportunity audit across your workflows, ranked by value and by how hard each one is
  • A data readiness check: what you can use today, what you cannot, what it would cost to fix
  • A working proof of concept for the top priority, on your data, in your stack
  • A 90-day roadmap with owners, costs and the numbers each step is accountable for
  • A governance note: what is allowed, what needs approval, and where the risk actually sits
Start a conversation
02

Architecture review £2,500–£4,500 over two or three days. Full builds usually £12,000–£45,000 over 6–12 weeks.

Advanced AI Consulting

For teams who are past the pilot and need the machinery built properly: retrieval that survives real documents, tools an agent can call safely, orchestration you can trust in production, and the control plane that makes all of it auditable.

  • What you receive
  • Retrieval architecture with an evaluation harness, scored against your hardest questions
  • Custom MCP servers, so an agent can use your systems under explicit permissions and an audit log
  • Agentic and multi-agent platforms with orchestration, acceptance gates, human approval and rollback
  • An AI control plane: model routing, version management, spend ceilings and tracing
  • Threat and isolation review as we build, then runbooks and pairing until your team can run it
Start a conversation

Also on offer

Five more engagements, if the need is narrower.

Each is fixed price and quoted after a call. The security review is a standalone deep dive into something already live; the platform work above reviews what we build as it is built.

Engagement 01

Knowledge Assistants

A retrieval-grounded assistant over your documents, policies and past work, so staff stop spending the afternoon looking for the one paragraph they know exists.

Usually 2–5 weeks. Fixed price, then an optional support retainer.

  • Ingest and structure your real sources, including the PDFs nobody has opened since 2019
  • Answers that cite the exact passage they came from, so people can verify instead of trust
  • Refusal behaviour: it says it does not know rather than inventing a policy
  • Evals built from your own hard questions, so regressions get caught before a client does
  • Access control that mirrors who is allowed to read what
Engagement 02

Workflow Automation

The unglamorous work that pays: documents read, triaged, drafted and routed without a human touching each one, with a person approving anything that leaves the building.

Usually 3–6 weeks. Fixed price against an agreed accuracy target.

  • One painful process, mapped end to end with the real exceptions in it
  • Classification and extraction you can trust, measured on your own labelled examples
  • Drafts into the tool your team already uses, with a human approval step that is hard to skip
  • An audit log of every decision, so you can defend one three years later
  • A fallback for when the model is wrong or the API is down
Engagement 03

AI Training Sessions

Half-day and full-day sessions for school staff, teams and boards: what the tools genuinely do, where they fail quietly, and the guardrails that keep you on the right side of safeguarding and data policy.

Half day from £850, full day from £1,500. INSET days and multi-academy trusts quoted per cohort.

  • A session built on your real documents and your real workload, not vendor slides
  • For schools: planning, marking workload, SEND and EAL support, and where a pupil's data must never go
  • For teams: the three workflows worth automating first, practised live on your own files
  • A written policy you can adopt the same week: acceptable use, procurement questions, disclosure rules
  • A follow-up clinic four weeks later, because that is when the real questions arrive
Engagement 04

AI Security Reviews

An outside read on an AI feature before an incident writes the report for you: what can reach it from the internet, who is allowed to ask it what, what data leaves the building to a model vendor, and what it can be talked into giving up.

From £2,500 for one deployed system. Platforms and multi-tenant setups quoted after scoping, usually 1–2 weeks.

  • Exposure map of the system as actually deployed, covering every way in from the public internet, with the ones that should not be open evidenced rather than assumed
  • Access control review: authentication, authorisation, API keys and service accounts, and what the assistant can reach once it is in
  • Data exposure audit: what it can read, what is sent to a model provider, what they retain, and whether any of it trains their model
  • Live prompt-injection and data-exfiltration testing against your own system, with the transcripts that worked
  • Audit trail check: who asked what, and whether you can reconstruct it afterwards when the question comes from a regulator
  • A prioritised fix list with owners, a one-page summary your board or DPO can read in five minutes, and a retest once the fixes are in
Engagement 05

Advisory Hours

For organisations that have their own builders and just need the awkward questions answered: senior time booked in blocks, with no retainer and no markup on anything you buy.

£150 an hour, in blocks of five or ten. No minimum term; unused hours roll for six months.

  • Architecture and model-choice review before you commit to a stack
  • Second opinion on a vendor proposal, with the questions to put back to them
  • Pairing with your developers on retrieval, evaluations, or a build that has stalled
  • A same-week read on something that has already gone wrong in production
  • Written notes after every session, so the reasoning outlives the person who asked

Your data

The question everybody asks second, after price.

Where your information goes is a design decision, so it gets made at the start and written down. These are the three rules we work to.

Least data that works

If anonymising or narrowing the corpus gets the same result, we do that first. Most assistants need your policies, not your customer records.

Where it is processed

A UK or EU-hosted model, a self-hosted open model on your infrastructure, or a commercial API under contractual data protections. Chosen and written down at the start, not discovered later.

Provable, not promised

An evaluation set built from your hardest questions, a score you can look at, and an audit log of what the system actually did.

How we behave

Four commitments

  • We will tell you when AI is the wrong answer, including when it costs us the work.
  • We do not build anything whose output cannot be checked by a person in ten seconds.
  • We do not train a model on your data, or a client's data, without it agreed in writing.
  • We do not take a percentage of your infrastructure or model spend.

Equally important

What we will not build

  • A dashboards-and-slides programme that produces a strategy nobody implements.
  • An assistant over documents nobody has reviewed, because it will be confidently wrong about the wrong things.
  • Automation that removes the last human from a decision with legal or financial consequence.
  • A six-month discovery phase before anything usable exists.

Next step

Bring the thing that is slowing you down.

Thirty minutes is usually enough to say whether it is worth building, and roughly what it would cost. There is no deck and no follow-up sequence.

Book a free call[email protected]