The first build had no AI in it.

We are an AI company, and the first thing we recommended to a 160-person fit-out manufacturer was a build with no AI in it. That was not modesty. It was the most valuable thing on the list.

The discovery

The company designs, builds and installs retail interiors from its own manufacturing facility, for well-known retail brands, across offices in Ireland and the UK. It had grown for decades. Its systems had not kept up.

We ran a structured discovery across every department. It started with a session with senior leadership. Then came a company-wide workshop, with 17 people over 121 minutes and every department head in the room. Then we held nine department deep-dives over two weeks, covering HR, production, installation, sales, fitout, UK operations and finance, plus two synthesis sessions. In all, it was 22 conversations with the people who do the work.

Tag the work before you touch it

We tagged every process step as one of three types:

• Deterministic: the same every time. Automate it with simple rules. No AI needed.
• Judgement: it needs context and a decision. This is where AI creates the most value.
• Relational: it needs trust and empathy. AI should free up time for it, not replace it.

This one habit removes a lot of bad projects. Much of what businesses ask AI to do is deterministic work that a plain integration does better, more cheaply and more reliably.

What we found

The symptoms were clear early. About 1,500 invoices a month were processed by hand. The same data was keyed into two or three systems: finance in Sage 200, documents in SharePoint, workforce in Softworks, and Excel for everything in between. Production sometimes started without full specifications, because detail was lost between design and estimating. And there was no forward view of capacity.

We mapped seven solutions and ranked them. Two were high priority: a single view of each customer from sale to delivery, and a suite that generates quotes, cutting sheets, contracts and picking lists. Further down came voice notes from site, materials intelligence, estimation from past projects, and finally a command centre for leadership.

Why the first build was the boring one

At the top of the build order sat a link between the finance system and the document system. It needs no AI. It removes the single biggest source of wasted time, which is typing the same data in again.

We chose it first for three reasons:

• It pays immediately. The people doing double entry feel the difference in the first week.
• It builds trust. The operations lead challenged our cost-saving projections hard in the audit presentation, which is exactly the scrutiny that makes recommendations stronger. A quick, visible win answers that better than any slide.
• It lays the foundation. Every AI system on the list needs clean, connected data. Estimation from past projects is only as good as the project records behind it. Build the plumbing first, and every later system is faster and cheaper.

What happened next

The detailed audit made the company’s application for Enterprise Ireland co-funding straightforward. A third-party cybersecurity firm reviewed the integration design. The build phase is confirmed: four systems connected, with 50 validated historical projects loaded, over an estimated 10 to 12 weeks.

When the person most likely to say “that won’t work” is the one pushing for the build to start, the discovery has done its job.

Where to start

Before you ask where AI fits in your business, ask where the same piece of data is typed twice. Fix that first. It is rarely glamorous, and it is almost always the foundation for whatever comes next.