We delivered an AI operations director for a family-run Charleston tour company — rebooking an entire afternoon of tours ahead of a thunderstorm before the first guest thought to complain, balancing capacity across five booking channels, and answering every review within a day through the busiest season they've ever had

A second-generation walking and carriage tour company on the peninsula came to us in April, staring down another summer that would either be their best season ever or the one that finally broke them. The product was never the problem — their guides are the kind who get name-checked in reviews — the operation around the product was. Bookings arrived through five channels that did not talk to each other: their own website, two online travel agencies, the visitor center desk, and a phone line that rang straight to the owner's cell. Every morning she rebuilt the day by hand — which guide takes the 10 a.m. history walk, whether the 2 p.m. is overbooked because an OTA sold four seats the website had already sold, who covers for the guide whose kid got sick. Then there was the Charleston summer itself: a thunderstorm parked over the harbor at 3 p.m. could wreck a whole afternoon, and every washout meant two frantic hours of calling guests one by one to reschedule — guests who were only in town for two more days, half of whom went to voicemail, some of whom just posted a one-star review instead of picking up. Reviews were their lifeblood and their neglected child: hundreds of five-stars thanking specific guides by name sat unanswered for weeks, and the rare bad one — usually a weather refund gone sideways — sat unanswered longest of all, right at the top where every prospective guest would read it. We built them an AI operations director that treats the day as one living schedule instead of five competing ledgers. The booking layer reconciles every channel in near real time, so a seat sold anywhere is a seat gone everywhere — overbookings, which had been a weekly apology, stopped in the first week. It watches the shape of demand and flags the owner when a Thursday ghost tour is pacing to sell out three days early so she can open a second departure while there is still demand to catch, and when a slow Tuesday morning walk has two guests booked against a three-guest minimum, it proposes consolidating them onto the afternoon tour with a personal message and an upgrade to sweeten it. The weather layer is the one the whole family talks about. The AI watches the forecast the way the owner used to — obsessively — but it acts earlier and all at once. When Wednesday's radar showed a storm cell arriving around 2 p.m., it drafted the plan by 10 a.m.: which tours to move, which guests to contact, what alternatives each had based on the length of their stay — tomorrow morning for the family here all week, the indoor-heavy architecture route leaving an hour earlier for the couple flying out Thursday. The owner approved the plan with one tap, every guest got a personal message with their specific options, and by the time the first drops fell, thirty-one of thirty-four guests were rebooked or refunded by their own choice. The old version of that afternoon was two hours of phone calls and three angry reviews; the new version was over before lunch. The guide layer took the morning rebuild off her plate: the AI drafts the day's assignments against each guide's certifications, preferred routes, and hour limits, flags conflicts before they happen, and when someone calls out sick it proposes the swap chain instead of leaving her to solve the puzzle on a sidewalk at 7 a.m. And the review layer finally gave their reputation the attention it earned: every review gets a drafted response in the company's voice within a day — five-stars answered warmly with the guide's name carried through, refund disputes answered with the facts of the weather policy stated kindly and precisely — each one approved or edited by a human before it posts. Ninety days in, through their highest-volume season on record: zero overbookings since week one, weather-day rebooking recovering ninety-one percent of affected guests versus roughly sixty before, review response time down from eighteen days to under one, and their OTA ranking climbed two spots — which the platform rewards with more visibility, which became more bookings. The owner's verdict: the AI didn't replace a single guide, it replaced the version of her that spent every summer drowning — and this is the first July she's actually walked one of her own tours.