Adding AI to How an Established SME Already Works
Last updated:
An established business has an advantage
Owners of established businesses often feel behind on AI, as if the startups have something they lack. In practice it is usually the other way round. A 25-year-old engineering supplier with 80 staff has what an AI startup would love: years of customer history, a known process, real volume and people who understand the work deeply.
What it lacks is spare time to experiment, and patience for anything that disrupts a business that already works. That is fair, and it shapes how we approach AI for SMEs. The goal is to make the existing business run with less tedium, not to turn it into a technology company.
Start with the process that eats the most hours
The best first AI project in an established SME is almost never the most exciting one. It is the job where somebody reads something and types what it says into a system, or sorts things into piles, or writes the same kind of reply repeatedly. Our post on what AI can actually do for a business today lists the categories that reliably work.
- Purchase orders and delivery notes arriving by email that someone keys into the order system
- Supplier invoices that need matching to orders before payment
- Customer enquiries that need routing to the right person and a first reply
- Quotes built by copying and adjusting old quotes
- Job sheets or site reports that need turning into structured records
If a task like that takes a person half their week, and the business has several such people, the first project nearly chooses itself.
Your existing systems are usually fine
A common worry is that AI requires replacing the ERP, the accounts package or the old database someone built years ago. It rarely does. Most AI features can sit alongside existing systems, reading from them and writing results back through an API, an import file or, for older software, a carefully built connector to the database.
| What you run | How AI usually connects | Typical difficulty |
|---|---|---|
| Modern cloud tools with APIs | Directly through the API | Low |
| Accounts packages such as Xero | Through the published API | Low to moderate |
| An older on-premise ERP | Database connector or scheduled import and export | Moderate |
| Spreadsheets and shared drives | Structured templates plus a small service | Low, but fragile if formats drift |
| A system with no access at all | Screen-level automation as a last resort | High and worth questioning |
The last row is where we would look hard at whether the system itself is the real problem. Our automation work often starts by making older systems reachable before any AI is involved.
Bringing your staff with you
Long-serving staff in an established business have seen technology projects come and go, and they will reasonably ask whether AI is a threat to their job. How that question is handled decides whether the project succeeds.
What works is involving the person who does the task from the start. They know the exceptions, the awkward customers and the documents that always come in wrong. We ask them to review what the AI produces during testing, and their judgement becomes the quality bar. In our projects the effect has usually been a change in what people spend time on, moving from typing to handling exceptions, rather than a change in headcount. We will not promise a business case built on redundancies we cannot see.
The person who has done the job for twelve years is the best tester you will ever have. Treat them as one.
What a first project costs and how long it takes
A single focused AI feature on real data typically reaches a working first version in two to three weeks, and a production version with error handling and monitoring follows once it has proved itself. Larger projects that span several systems take longer. Every SpiderHunts project is scoped and priced before work starts, so there is no hourly meter running while you find out.
The measurement is simple and worth agreeing in advance: how long the task takes now, how many errors slip through, and the same figures after launch.
Compliance you should think about early
Most SME AI projects involve some personal data, even if only names on invoices. Under UK GDPR and EU GDPR that means knowing which provider processes the data, where, and on what terms, and making sure your privacy notices cover it. Businesses selling into the EU should also be aware that EU AI Act obligations are phasing in, with the heavier duties aimed at high-risk uses such as recruitment and credit decisions rather than invoice processing.
We use API arrangements that exclude training on your data, redact personal details where the task allows, and document the data flow so your data protection lead can sign it off. None of this is exotic; it just needs doing before launch rather than after.
A sensible first year
- Pick one high-hours task and measure it properly
- Build a focused AI feature inside the existing system and test it with the people who do the job
- Run it with human review until the evidence says review can be reduced
- Choose the second project using what the first one taught you
- Decide whether someone internal should own AI tools day to day
Two or three well-chosen projects in a year will change an established SME more than a dozen experiments. If a task is better served by an off-the-shelf tool, build versus buy for AI features will help you decide, and we will tell you the same on a call.
Frequently asked questions
Do we need to replace our old software to use AI?
What is the best first AI project for a small or medium business?
Will adding AI mean making staff redundant?
Does the EU AI Act affect a UK SME using AI?
Running a well-established business and wondering where AI fits?
Tell us which part of the week eats the most staff time. We will suggest where AI could help, what it might cost and what we would leave exactly as it is.