AI that plugs into the systems you already run
Can AI work with the ERP and systems we already run, without breaking the business?
It can, if the architecture accounts for how those systems actually work. AI UVD helps UK operations businesses decide where AI fits, which route into each system is safe, and who checks the output, before anything is built. The work rests on more than 20 years of connecting enterprise systems. The first conversation is free.
The first step
The first step is an Opportunity Map: £4,500, fixed, ten working days from access, and a written answer on what to do first.
AI UVD is me, Sam Naseem, and I do the work. Where this site says we, it means the firm around that: an accounts lead for proposals, contracts and invoices, and trusted engineers I bring in for specific work and name in the proposal before they go near your systems.
Every engagement is architected by me, and I'm the person you deal with.
Sam Naseem
We take on a small number of clients at a time.
Why the integration is the hard part
A model can read an email well. Then someone has to put the answer into the ERP, against the right account, with a record of who approved it. That step is the work.
The work rests on more than 20 years of connecting enterprise systems: ERP, WMS, EDI, PMR, CRM and EPOS. AI adds a new component to the same discipline. The hard part is connecting an agent to a twenty-year-old ERP without breaking the business.
The architecture every engagement works to:
Which route into each system. If there is an API, the AI uses it, with the fewest permissions that do the job. If there is no API, the route is what the system already produces: the nightly export, an emailed report, an EDI feed, a read-only copy of the database.
What the AI may change. Whether it may write to the ERP at all. Anything written back goes through a route the system already trusts, such as an import file a person approves.
Who checks the output. A person approves the AI's work until measured results justify checking less.
What gets recorded. What the AI read, what it proposed and who approved it, so a mistake can be traced.
Whether to build or buy. Whether a product already on the market does the job, or something has to be built, and what each would cost to run.
Simple things are simple. If an off-the-shelf tool already does what you need, I will say so.
Expectations set on what the technology can actually do today, measured as we go — so what we agree at the start is what gets delivered.
The services
Start with the Map. Anything after it is scoped and priced in writing before it starts.
We use AI agents in our own delivery, and that is why the prices above are fixed and the timelines short. Agents I have trained draft, research and test under my review, so fewer engineering hours go into each deliverable. Judgement, architecture and anything that touches your live systems stay with me.
Opportunity Map
A written assessment of where AI fits your operation, what each option costs and needs, and what to do first.
Agent Pilot
One AI agent in one real workflow, measured against how the work runs today and handed over with a runbook.
Private AI
AI on your own infrastructure, or in the cloud region you require, inside a written data boundary, measured against hosted models.
Fractional AI Architect
An external AI and systems architect for three to five days a month. It is for businesses running older systems that need someone senior to steer AI and integration decisions, and cannot hire that person full time. It is the retainer previously offered as Stewardship, and work I have done for years.
£6,000 a month · three to five days a month · three-month initial term, then rolling. Positions are written, not verbal. Building anything is a separate, priced piece of work.
How an engagement works
You write. A few lines on what the business does, which systems it runs, and the decision or problem in front of you. The contact page says what helps.
Sam Naseem reads every message and answers it in writing. If a call would help, the reply suggests one. The first conversation is free.
If there is paid work, you get a written proposal: fixed scope, a price in pounds, exclusions, and what AI UVD needs from you. Each service page says who does which part and who checks it.
It also says what AI agents are never allowed to touch.
Opportunity Map. Interviews with four to six of your people, a walk through your systems and real data, and a written Map: the top opportunities in order, each with its cost, risk and first step.
Agent Pilot, measured against a baseline. How the work runs today is measured first. The agent then runs in shadow for at least two weeks, with a person reviewing every output. The result is compared with the baseline, and the Pilot is handed over with a runbook your people can use to operate, pause and roll back the agent.
Nothing is left in anyone's head. Decisions are written down and handed over, so your people can carry on without AI UVD. If you want someone senior steering, that is the Fractional AI Architect retainer.
Being small keeps this quick. Decisions are made in the room with the person who designs the system, the person who scopes the work is the person who builds it, and if the work shows a better route mid-project, changing course is a written note, not a change-request process.
| Candidate | Value | Feasibility | Data | Risk | Decision |
|---|---|---|---|---|---|
Column headings only. Every candidate considered gets a row, including the ones removed, and each score has a one-line reason.
Frequently asked questions
Why not just use Copilot or ChatGPT?
For drafting, summarising and searching documents, they may be all you need, and I will say so. Reading a purchase order, checking it against the ERP and raising an exception someone approves needs integration, permissions, a test set and a measured result. Deciding how that is done safely is architecture work, and on every engagement it is mine.
Our ERP is ancient and has no API.
That is common, and workable. An AI tool can read from a nightly export, an emailed report, an EDI feed or a read-only copy of the database. Anything written back goes through a route the ERP already trusts, such as an import file a person approves. The route is agreed in writing before anything is built.
What happens when it gets something wrong?
It will, sometimes, so the design assumes it. A person checks the output before anything reaches your systems, at least until measured results justify less checking. Items the AI is unsure about go to a person, the design lists what it must never do, and every action is logged so a mistake can be traced and put right. In an Agent Pilot, a person reviews every output during a shadow run of at least two weeks.
Who else have you done this for?
No client is named or quoted on this site. The work rests on more than 20 years of connecting enterprise systems, and the first conversation is the place to test that against your kind of system. Every service page says what you get, what you don't, and what can go wrong, so the method can be judged before you buy it.
Can you just take a quick look at our system first?
The first conversation is free. Anything after it is scoped and priced in writing before it starts. That way you know what a closer look would cost, and what you would get, before you agree to it. That closer look is the Opportunity Map: £4,500, fixed, ten working days from access.
Ask about an Opportunity Map
Say what the business does, which systems it runs, and the decision or problem in front of you.
Sam Naseem reads every message and answers it in writing.
Every engagement is architected by me, and I'm the person you deal with.
Sam Naseem
We take on a small number of clients at a time.