Operational analysis & applied AI · Dublin
We take one important part of how your business runs, work out how it works in practice, and find where time, information and responsibility are being lost. Then we fix one or two things, using AI where it genuinely helps, and measure whether it worked.
As a business grows, the work spreads out faster than anyone's understanding of it does.
Nobody planned it. Responsibilities get split. Systems get added one at a time. Knowledge ends up in a few people's heads and in a lot of inboxes. Owners and managers become the bottleneck because they are the only ones who can see across the whole thing.
Then AI arrives, and it gets pointed at a process nobody has properly looked at.
AI runs on whatever coherence a business already has. Where things are clear, it helps. Where they are muddled, it spreads the muddle faster.
One workflow that matters and costs you real time. Whole-business reviews take months and produce documents nobody reads.
Not how the procedure says it runs. We talk to the people doing the work, follow the information, and pay attention to the exceptions and the workarounds. That is usually where the cost is hiding.
Handling time, cycle time, volume, backlog, where things get stuck. If the data isn't there, we build a baseline by sampling and observation. Without a baseline, an improvement is just an opinion.
Better intake, automated follow-up, drafting, classification, retrieval, a dashboard: whatever the analysis points at. AI where it earns its place, ordinary software where that is the right answer.
Against the same baseline, and again a month after handover, because plenty of improvements slip back. You own what we build.
If your product includes AI features at a fixed price per seat or licence, every AI call is a cost inside that price. Most software companies cannot say what that cost is per customer, so they cannot tell whether their heaviest users are profitable.
Unfold is an AI platform for teacher development, co-founded by Conal. Before launch we measured what it cost to run, against what we had budgeted. We were wrong three times out of three.
Modelled from those measurements: a heavy user could cost about 42% of the licence price.
Measured on Unfold over 30 days of cloud monitoring and three months of billing, during development and before launch. Replies averaged 256 tokens. The hosting forecast was about $425 a month; the whole account ran at about $70 to $76 a month at pre-launch usage. This was development and demo traffic, not production use. These are our figures, not a projection of yours. The point is the size of the gap between an informed assumption and a measurement.
Cost sprint · 2–3 weeksFixed scope and fixed price. Every figure is labelled as measured, modelled or estimated. We need read-only access to billing and usage data, never your customers' content.
Automating a broken process makes it break faster and more consistently. The analysis comes first, every time.
Measured, modelled and guessed are three different things. We label which one we are giving you, including on this page.
Every automated workflow gets defined limits: what it may do, what it may not, where a person has to approve, what gets logged, and how to switch it off. Anything consequential keeps a human in it.
Data minimisation, access control, audit trails and GDPR obligations are part of the build, not a checklist in the final week.
Accounts, data, systems and documentation are yours and stay in your name. We are not trying to become part of your monthly costs.
You know what you are getting and what it costs before we start. No open-ended retainers.
If some part of how your business runs is costing more time than it should, or you don't know what your AI features cost to serve, a short conversation costs nothing. It will usually tell us both whether there is something here worth doing.
conal@optech.ie