Where to Start With AI in 2026: How to Pick the First Task Worth Setting Up
Three AI tasks pay back almost every time they get set up properly in a small business in the property economy. Most owners, once they accept that, immediately try to set up all three. Six months later, the most common outcome is that none of them are properly in production.
The question worth answering right now isn’t which tasks does AI do well? — that map has steadied, and we covered it in What AI Actually Does. The question is which of the three to start with, when to start the next, and which of the showier capabilities to leave alone for now.
What follows is a picker’s guide.
The three tasks, in one place
For context, the short list. The one-sentence version of each, plus what it actually saves.
- Drafting routine replies in the firm’s voice. Booking confirmations, tenant queries, supplier chase emails, renewal reminders, quote follow-ups. A four-person firm typically spends 8–12 hours a week on routine outbound; AI drafting with the right setup cuts that to about 2 hours of review.
- Pulling structured information out of messy messages. A booking with the dietary note in paragraph three. A landlord enquiry mentioning the EPC in passing. A quote request with the scope buried in informal language. AI turns the message into a clean structured record so nothing gets lost between systems — which removes a whole category of complaints and re-work, not just a stack of hours.
- Generating client-facing documents from rough notes. Six bullet points from a site visit. Ten minutes of voice notes from a discovery call. A few lines from a meeting with a landlord. AI produces the first draft of the proposal, the brief for the design team, the letter to the tenant — in the firm’s voice, structured the way the firm structures its documents. The value is asymmetric: this saves the most senior people the most time, because they’re the ones who’d otherwise have written it at 7pm from scratch.
Each one is dull. Each one currently sits with people more senior than it needs to. Each one is reviewable in seconds. Those conditions are the filter we unpacked in the 4-question test — and they’re why these three pass while most of what gets sold as “AI for business” doesn’t.
Why setting up all three at once doesn’t work
The setup work is small per task — usually a couple of hours of structured conversation, a handful of example documents, an agreement on the standing rules, a review of the first week’s output. None of it is hard. All of it requires attention from the people who already have the least of it.
Stack three of those efforts on top of each other and the attention runs out. The first task gets done. The second gets half-done. The third never makes it past the kickoff. Two months in, the team is using the first one inconsistently because nobody had time to embed the routine, and the conclusion in the room is that “AI didn’t really land for us.”
The pattern that works is the opposite. Pick one. Get it embedded in the team’s week for a month, until the people using it stop thinking about it. Then start the next.
How to pick the one to start with
The most reliable picker is one question: which of the three tasks is the most senior person currently spending the most hours on?
That’s the one. Not because it’s the easiest to set up, but because the recovered hours land back with the person whose time is most expensive and whose attention does the most when freed up. In most firms in the property economy, that’s also the person whose Admin Tax is the biggest hidden cost on the P&L.
Two other pickers come up. Which one is the team most frustrated by? — useful, but frustration doesn’t always track to hours saved, and the team will sometimes nominate the noisy task over the expensive one. Which one is the easiest to set up? — tempting, but easy-to-set-up usually means low-volume, which means low payback, which means the team never quite cares enough to embed it. Senior-hours-on-it wins because it’s the picker most likely to produce visible value in the first month, which is what carries the rest of the year.
Once the task is picked, the last check before starting is to run it through the 4-question test — a quick filter for whether the task itself is genuinely AI-ready, or whether there’s a process gap to fix first. The picker tells you which task; the test tells you whether that task is ready to be handed over. Both, briefly. Then start.
What to stop worrying about
A useful inverse of the list. The things that get a lot of airtime but aren’t reliably ready in a small business right now — and shouldn’t distract the picker conversation:
Fully autonomous customer service
Voice agents have crossed into reliable for narrow tasks like taking a booking or triaging an enquiry. They aren’t yet ready to handle a complaint, manage a relationship with a regular, or improvise around an unusual request without supervision. (More on where agents do fit, in AI Agents for Small Business.)
End-to-end document generation with no review
The “AI writes and sends the proposal” version of task three. The review step is what keeps the rest of it safe.
Custom AI models trained on the firm’s own data, end to end
Possible. Expensive. Almost never the right answer when an off-the-shelf tool with good setup gets ninety percent of the result for five percent of the cost.
Anything that promises to replace a role wholesale
The honest model in 2026 isn’t to remove a person from a function. It’s to take a slice of that function — the dull, repetitive, reviewable slice — and give the person their time back for the work that needs them. The same pattern holds in cleaning and facilities firms as in property or construction.
When to start the second one
The honest signal is invisibility. The first task is ready to be followed by the second when the team has stopped talking about it — when nobody mentions the drafting tool in the Monday meeting because it’s just how Mondays work now. For most firms that’s four to eight weeks after go-live, depending on how routine the underlying task is.
If the team is still discussing whether to trust the output, still rewriting most of it, or still forgetting to use it, the second task waits. Layering a second setup on top of a half-embedded first one is the same mistake as starting all three at once, just slower.
The recovered hours from the first task are the budget for the second. Both in literal time — the senior person now has the bandwidth to be properly involved in the next setup — and in credibility, because the team has seen one work and is more open to the next.
The shape of a sensible year
For a typical four-to-eight-person firm in the property economy, a sensible run looks roughly like this. Pick task one this month. Get it embedded over the next four to six weeks. Run it cleanly for another month before starting task two. Repeat. Inside a year, two of the three are in production, the team trusts the approach, and the third is queued for the following quarter.
That’s a much smaller ambition than most owners arrive with. It’s also the version that delivers. The version that tries to do everything in one quarter is the one that quietly stops being mentioned by the next.
If you’d like to see the numbers before any conversation, the free AI Value Calculator gives you an instant estimate of hours reclaimed and annual savings based on team size and salary data. If you want help running the picker on your specific week — which of the three a senior person is currently spending the most hours on, and what setting it up properly would actually take — an AI Strategy & Operations Audit is built around that conversation. For the one you pick, a Bespoke AI & Automation Build is what gets it sitting properly inside the workflow your team already uses.
Want help picking the right first task for your business?
Book a free consultation. We’ll look at last week’s calendar, identify which of the three tasks would give the most senior person their hours back fastest, and tell you plainly what setting it up properly would actually take — no jargon, no commitment.
Book Free Consultation