Where should businesses actually use AI?
Richard Hardy · CTO
Where does AI genuinely add value in a business, and where is it hype?
AI earns its place in a business where the work is high-volume, pattern-based and currently absorbing skilled people's time: document classification and processing, knowledge search across your own material, first-draft production work, intelligent routing of requests, and decision support with a human still deciding. It is mostly hype where it is bolted on for the demo — chatbots nobody asked for, AI strategy decks with no workflow behind them, and anything promising to remove human judgement from work that clearly needs it.
We added AI to RJDM's toolkit in 2023 — not as a product line, but in the workflow, where it either pays back or gets removed. Two years of building AI into real client systems has left us with a fairly unglamorous view of where it works. That view is the useful part.
The pattern that works
AI adds value where three things line up: the task is high-volume, the inputs follow patterns (even messy ones), and skilled people are currently doing it by hand. Document classification is the cleanest example — invoices, referrals, applications and reports that someone reads just to decide where they go. Knowledge systems are a close second: letting staff ask questions of your own manuals, policies and project history instead of asking the one person who knows. Add workflow automation with AI handling the ambiguous middle steps, AI-assisted reporting that drafts what a human then owns, and intelligent routing that gets each request to the right queue.
The test we apply
Before any AI feature goes into a system we build, it has to pass one question: if this worked exactly as promised, whose week gets lighter, and by how many hours? If the answer is vague — "it will make us more innovative" — it fails. If the answer is "the finance team stops hand-keying 400 invoices a month", it is worth scoping. AI used where it pays back is a measurable claim, and you should hold your suppliers to it.
Where the hype lives
Be suspicious of AI that exists to be seen. Customer-facing chatbots deployed because competitors have one, not because customers wanted one. Strategy decks recommending "an AI transformation" with no named workflow attached. Anything that promises to remove human judgement from decisions with real consequences — clinical, financial, legal — rather than support it. And AI features added to a product pitch to justify the price. The tell is always the same: no specific task, no specific hours saved, no owner.
Keep the human where the judgement is
The systems that stick are copilots, not autopilots. AI drafts, a person approves. AI sorts, a person handles the exceptions. AI surfaces the relevant history, a person makes the call. Decision support beats decision replacement in almost every business context we have deployed into — not for sentimental reasons, but because the exception cases are where the business risk lives, and pattern-matching is worst exactly there.
How to start without a moonshot
Pick one workflow that is annoying, measurable and safe to get slightly wrong at first. Build the smallest version that touches real work. Measure the hours before and after. Then decide whether to widen it. That is the whole method. It is less exciting than an AI strategy, and it is the difference between AI in the workflow and AI on the slide.
