The most useful number published about AI customer service last week was not a benchmark score. It was a 6.26.
That is the average rating buyers gave AI customer support agents for quality of support, in a July 31 analysis by G2 of more than 7,900 verified reviews of the category. Ease of use came in at 6.36 out of 10. Meeting requirements, 6.34. These are not disaster scores, but nobody would call them love. The same analysis found the average time to ROI in the category is 6.1 months — faster than both chatbots and help desk software.
Read those two facts together and you get the actual state of the market in 2026: the technology pays back quickly, and buyers are still only moderately happy with the experience of owning it. For anyone running a salon, a spa, a barbershop or a small clinic and wondering whether this is the year to put an AI agent on the phone and the website, that gap is the whole story. It tells you where the risk is, and it is not where most demos point.
The failure mode is implementation, not intelligence
G2’s read on where these products fall short is specific: vendor support, implementation experience, and demonstrating long-term business value. Not accuracy. Not conversation quality. The report’s summary of what reviewers reward is blunt — users value products that perform well after implementation, not products with impressive feature lists.
That maps almost exactly onto how a wellness business actually breaks a software rollout. The agent that answers “do you do balayage” correctly is table stakes. What decides whether it earns its keep is whether it can see that a colour correction needs 180 minutes and a men’s cut needs 30, whether it knows Marta is off on Tuesdays and her regulars will wait for her rather than take another stylist, whether it writes the booking into the calendar your team already works from instead of a second calendar nobody checks, and whether it flags the client with a patch-test note in their history before it confirms anything. Every one of those is an integration and configuration question. None of them shows up in a scripted demo.
The practical version: when you evaluate a vendor, spend most of the call on the boring half. Who does the setup. How your service menu, durations and staff rules get in. What happens to a booking when two clients ask for the same 2pm. Who picks up the phone when something is wrong on a Saturday morning, which is when it will be wrong. The way those details get handled in appointment-led businesses is what we break down on our wellness and personal care page.
Buying got harder, and small operators feel it differently
The other half of the picture comes from G2’s 2026 Buyer Behavior Report, published July 22 from a survey of more than 1,000 B2B software buyers. Evaluation has overtaken research as the longest stage of the buying journey. Finance involvement in software decisions jumped from 31% to 46% in a year, and nearly half of buyers said their CFO vetoed an already-approved deal in the last twelve months. Seventy percent say the pace of AI change pushes them toward shorter contracts.
A four-chair salon has no CFO. The owner is the CFO, and the veto happens at 11pm in the back office with a calculator. That is an advantage, not a disadvantage. Enterprise buyers are stuck in evaluation because of security reviews and budget committees; you can run a real test in a month and decide with your own numbers. The shorter-contract instinct is worth borrowing too — in a category moving this fast, a twelve-month lock-in on a tool you have not lived with is the expensive part, not the monthly fee.
Voice is still the thin part of the market
One more finding is worth flagging, because it runs against the marketing. G2’s earlier AI in Customer Support report from January looked at how support software vendors run AI in their own operations. All five were using AI in chat, email or ticket workflows. Only one applied it to voice support as of late 2025. None described a fully autonomous support model; hybrid, with AI assisting people, was the norm across the board.
For most businesses in this sector that is exactly backwards from where the pain is. The missed call during a treatment is the lost booking. The 9pm “can I move Thursday” text is the no-show you could have saved. Text is the mature side of the market and voice is the underbuilt side — which means voice is where a working agent still creates a gap between you and the salon down the road, and also where you should test hardest before signing.
Measure it against one number
The honest lesson from 7,900 reviews is that this category rewards operators who treat it as an operations project rather than a purchase. Pick a single metric before you start — calls that go unanswered in a week, or bookings taken outside opening hours, or the no-show rate. Take the baseline. Run thirty days. Compare.
If the number moves, 6.1 months to payback is a reasonable bet on a business where one recovered appointment a week covers the subscription. If it does not move, you will know in a month rather than a year, which is the only part of enterprise software buying nobody should imitate.
The TalkRev blog
One short email per new post
No ads. Unsubscribe anytime.
