AI for Pool Service Businesses: What Works and What Doesn't

Every piece of software sold to pool companies has acquired an AI feature in the last two years, and most of the lists ranking those features are published by the vendors who appear in them. That makes it unusually hard to work out which of these things would change anything on a route of six to sixty pools, and which are a checkbox on a comparison sheet.

We sell two of the things on this list, so read the rest with that in mind. What follows is still the honest version: six categories, what each one does mechanically, what it cannot do, and whether an owner should pay for it. One of them is worth real money to almost every pool company. Two are useful at the margin. Three are mostly a word applied to software that already existed.

The missed call is the real problem

Start with the gap rather than the technology. A pool company's phone rings hardest between roughly nine and four on weekdays, which is exactly the window in which every person capable of answering it is standing on somebody's deck with a pole in both hands. The call goes to voicemail. Some fraction of those callers leave a message. A smaller fraction wait for the call back. The rest ring the next company on the list, and you never learn that they existed.

You can price this yourself, which is why it is the one item on the list that does not require taking anybody's word for anything. Take your own average first-year value of a residential account. Multiply by the number of unreturned calls you can find in a month of phone records. That is the number any phone-answering product has to beat, and for most route businesses it is not a close contest — which is also why the category is where every vendor has put its effort.

AI voice agent

Software that answers an inbound call, holds a spoken conversation with the caller, and takes an action in your system as a result — looking the caller up, booking a visit, drafting a work order or transferring the call to a person. It is distinct from an answering service, which is a human in a call centre, and from an IVR phone tree, which only routes the call.

The six things you will be sold, and what each one is

The word covers at least six unrelated products. Separating them is most of the work, because the weakest items in the category borrow their credibility from the strongest.

Phone answering, and what it costs

Two things decide whether a voice agent earns its keep: whether it can see your customer records, and how it is priced. On the first point, an agent that cannot identify the caller is a receptionist with no filing cabinet — it can take a name and a complaint, which is roughly what voicemail does. An agent connected to your records can tell a caller when their next visit is, what they owe, and that the pump they are asking about was replaced in June. Ours does that: it identifies the caller by number, pulls up their properties, upcoming services and balance, can schedule a visit or draft a work order, optionally quotes, and hands the call to a person when it should. It also logs a transcript, which is the part owners end up valuing most, because you can see what callers are actually asking for.

On price, the category has settled into two shapes: a flat monthly add-on, or a monthly fee plus usage. Both are defensible; the usage model is cheaper for a quiet phone and more expensive for a busy one. The figures below are each vendor's own published prices on the day we checked, and they are the only two in this market we could verify from a published page — several competitors publish no add-on price at all, so they are left out rather than guessed at.

It is worth being clear about the ceiling, too, because the category is oversold on exactly this point. A voice agent is good at the calls that are mostly lookup and mostly routine: when am I next due, what do I owe, can you add a filter clean, my gate code changed, I am moving. Those are the majority of a pool company's inbound calls and they are the ones that eat an afternoon. It is not good at the calls that need judgement — pricing an unusual repair, reading that a customer is on the point of cancelling, deciding whether to go out tonight for a green pool before a party. Those calls want a person, and the only thing you should ask of the software is that it recognises them quickly and hands them over cleanly. Test the transfer path on the demo before you test anything else; it is the failure mode that costs you customers, and it is the one vendors demonstrate least.

The chatbot question is really a traffic question

Website chatbots get bundled with voice agents and sold as the same purchase, and they are not. The economics are completely different. A voice agent works on demand that already exists and is already reaching you — the phone is ringing whether you buy anything or not. A chatbot only converts visitors who are already on your website, which means its value is a direct function of how many of those there are. If your site gets a few dozen visits a month, a chatbot will handle a handful of conversations a year and no amount of configuration will change that. If you are running paid search, ranking locally, or getting steady referral traffic to a quote page, the same feature starts catching enquiries that would otherwise have bounced at eleven at night.

So the honest sequencing for most pool companies is: fix the phone first, because the demand is already there; then get found, which is local search work rather than a software purchase; and add the chatbot when there is traffic for it to answer. Buying it in the other order is the most common way to end up paying monthly for a feature that has spoken to nobody. The one exception worth noting is after-hours quoting: if a meaningful share of your enquiries arrive in the evening and your competitors are all answering the next morning, a chatbot that captures a name, an address and a rough scope overnight is doing work that no amount of staffing would cover economically.

The three where the word is doing the work

Route optimisation is the clearest case. Sequencing stops to cut drive time is a well-understood computational problem that logistics companies have been solving since long before anyone said AI, and the good implementations — including ours — are solvers, not language models. This is not a criticism: it works, and it is one of the few features that gives back time every single day. But it means the AI label tells you nothing about quality, and the only sensible test is to run your own route through it and look at what comes out. It also will not rescue a route whose real problem is geographic, which is a reorganisation job, not a software one.

Chemical dosing is the same story with a better ending. A dose recommendation is pool chemistry applied to a reading and a volume. Because it is formulas rather than a model, you can check it, and a technician who understands why the number is what it is will catch the day the volume on file is wrong. Be wary of any version of this that cannot show you its working, and remember that it inherits every error in the test — a confidently wrong dose is worse than no recommendation at all.

Marketing copy is the weakest of the three. A general-purpose writing tool bolted into service software is the same tool you can use directly, usually at a lower price, and it does not know your market, your prices or your reputation. It is a reasonable convenience and a poor reason to choose a platform.

How to evaluate an AI feature before you pay for it

The same five questions separate the useful from the decorative, whatever the feature is called.

Where we actually stand

To be explicit about our own position, including where it is narrower than this article might imply. Pool Runs sells a voice agent, and its minutes sit in the Communications add-on priced above. It has route optimisation that is a solver rather than a model. It does not write your reviews, and it does not bolt a general-purpose writing tool into the product and call that a reason to switch platforms. On chatbots we are the weaker option as things stand: the thing we call an AI chatbot is an internal assistant that answers your own questions about your customers, invoices, routes and work orders in plain English, read-only and inside your permissions. It is genuinely useful if you are the person who keeps asking the office which accounts are overdue — and it is not a website chatbot that talks to visitors and captures enquiries. If that is what you are shopping for, we do not currently have it, and you should weigh that against whoever does. If you are weighing us against the market leader more broadly, the side-by-side comparison is a better starting point than this article, because the decision turns on far more than which features carry the word AI.

The honest summary is that one of these six categories is close to a straightforward buy for a route that misses calls, two are worth having when the conditions are right, and three should not move a purchase decision at all. If you want to hear what a voice agent sounds like answering as your company before you pay anybody for one, that is the quickest thing to arrange.