What is an AI agency, and what does one in Dublin actually do?

The phrase covers so much ground that it has stopped narrowing anything down, which is a problem when you are the one signing. It helps to ignore what the companies call themselves and look at what gets delivered instead, because that falls into a much smaller number of groups.

Most of it is connecting things. A service business of any age is running a CRM that does not know what the inbox knows, a calendar that does not know what the CRM knows, and a spreadsheet somebody built 3 years ago that quietly outranks all three. Connecting them means deciding which one holds the truth and making the others follow it, so that a change of address gets typed once rather than 4 times, or more realistically, once and then forgotten in 3 places.

Some of it is removing repetition: the onboarding pack, the document chase, the appointment reminder, the monthly report rebuilt by hand from the same 4 sources. None of that work is difficult, and that is exactly why it survives. It is easy enough to keep doing manually, and nobody ever blocks out the afternoon it would take to stop.

A smaller part is answering, which is where voice agents come in. Calls that arrive while the person who would have taken them is on a job, enquiries that land at 9 in the evening, the same 5 questions in slightly different words each time. An AI receptionist handles those, or it does not, depending on whether the volume justifies building one.

And a part that gets skipped almost everywhere is deciding what your own staff should be doing with AI: which tools they may use, on what data, and where a human has to sign before anything leaves the building.

That is the whole category. Everything past it is packaging.

Almost every conversation starts with a tool. Somebody has seen a demo, or a competitor has mentioned a platform, and the question arrives already shaped: should we be using this one or that one.

Answering it in that order is what makes these projects expensive. A process that is broken when a person runs it is still broken when software runs it, only now it breaks faster and at greater volume, and the failure looks like the tool rather than the process. So the tool gets swapped. Six weeks later the same problem returns wearing a different logo, and by then the budget has an opinion about AI.

The order that works is duller and starts with the week itself. Who touches what, how often, where information moves and where it stops, what happens when the person who normally does it is away. Do that first and the tool stops being a decision at all. By the time the process is written down, the requirements have narrowed the field to 2 or 3 options and the choice between them barely matters.

Three kinds of supplier can take this work, and knowing how each one fails is more useful than knowing what each one charges.

A freelancer builds what you specify, quickly and cheaply, which is the right answer when the scope is a single workflow you already understand. The risk sits with the specification. A freelancer is paid to build the brief rather than to argue with it, so if the brief is the problem, you will get an efficient version of the wrong thing.

A software house builds a product. That makes sense when nothing on the market fits and what you need genuinely is new software, and it arrives with the timeline and the budget that implies.

An agency of the third kind sits between them, assembling a system out of tools you already pay for and giving it the logic your business runs on. Less code, more plumbing. It suits the case where the tools exist, nothing talks to anything, and the process needs writing down before it needs building.

There is a fourth option people reach for too early, which is hiring somebody in. That works once the volume of internal systems work is permanent and full time. Until it is, you are paying a salary to wait.

Geography matters less than it used to and more than the remote-first pitch suggests.

Irish service firms tend to run lean. A firm of 12 has no operations manager and no internal IT, and the owner is still the last signature on quotes, complaints and hiring. That shapes what can sensibly be built, because a system nobody has time to maintain is a system that will quietly stop being used, and it shapes how much a handover has to carry.

It also shortens the decision. In most firms this size the person who feels the problem is the person who signs for the fix, so work can be scoped and agreed in one conversation instead of 3 rounds of committee.

The rest is proximity, which is worth more than it sounds. Sitting with somebody through an ordinary Tuesday surfaces things that never come up on a call: the second spreadsheet nobody mentions because it seems obvious, the WhatsApp group doing the scheduling, the paper diary that is still, whatever the CRM says, the real source of truth.

Otomatic works from Dublin with owner-led service businesses across Ireland and the UK, in recruitment, property, marketing agencies, broking, trades and local services, and wellness and coaching.

No credible supplier in this market publishes a price list, and the reason is duller than evasion. The same sentence from 2 different businesses produces genuinely different builds.

Five things move the number. How many systems have to talk to each other, and whether they have proper connections or need custom work. How many points in the process need a human to approve something, because each one is a state that has to be designed, tested and documented. What a wrong output costs, since higher risk buys more testing and a slower rollout. How clean the underlying data is, which is regularly the largest hidden cost in the project. And how long somebody stays with the build after it goes live.

None of that can be known before somebody has looked at your process, which is why a number quoted in the first email is a guess with a margin on it. Scope and price should arrive together, attached to a named deliverable, once the mapping is done.

The first conversation should be about your week rather than their capability. Thirty minutes is enough to map how the work runs now, including the parts that exist only in one person's head.

What comes out of it is a written roadmap, and this is the part worth being firm about. What an audit covers is worth knowing before you sit down for one. It should name the systems worth building in priority order, with the dependencies and the scope boundaries, and it should be specific enough that you could hand it to a different supplier tomorrow. If it is not, it was written for the sale rather than for you. Ours arrives within 24 hours and is yours to keep either way.

After that comes a fixed-scope build, tested against the real workflow rather than a clean example, then documentation, a recorded walkthrough and 30 days of monitoring, because the first fortnight of live use always finds cases the mapping missed. Throughout, the workflows sit in your own accounts. If they sit in the supplier's, you are renting your own operations, and the cost of leaving is a rebuild.

Three situations come up often enough to name, and in all 3 the honest answer is that a system is the wrong purchase.

The first is when there is no repeatable process yet. If the work changes shape every time you run it, there is nothing stable to build around, and whatever gets built will need rewriting before it has paid for itself.

The second is when the goal is to replace somebody rather than to free them up. Those projects fail on adoption rather than on technology, because the people who have to run the system every day have every reason to route around it.

The third is when demand is the constraint. Handling 4 enquiries a week faster still leaves you with 4 enquiries a week. That is a marketing problem wearing a systems costume, and no amount of follow-up automation will fix it.

Any supplier worth hiring will tell you which of these applies in the first conversation rather than the third.

Before you shortlist anybody, write down the 5 things your team did most often this month and roughly how long each one took. It is a dull exercise and it is worth more than any vendor comparison, because it turns a vague interest in AI into a specific brief, and a specific brief is what separates the 3 quotes on your desk.

The Otomatic Systems Audit takes 30 minutes and costs nothing. The written roadmap follows within 24 hours, and you keep it whether you hire us or not.