Last week in Mexico City, I could not remember if I had tipped our guide. The van was pulling up. The guests were standing on the sidewalk with their bags. I had a tipping guideline sheet I had printed before the trip, but it was back at the hotel in the pile of paper I had not brought with me that morning. My co-leader was with a different part of the group. I could not text her fast enough to check. So I tipped him, which meant I tipped him twice, because she had already done it.
That was not a small amount of money. It came out of petty cash that the office would eventually reconcile, which means the operator ate the cost of my information gap. On a trip with twenty guests and dozens of small transactions like this, those gaps stack up fast.
A decade ago I was guiding cycling trips. Pre-GPS. We ran on paper cue sheets and printed rider rosters. The cue sheet told you the route. The roster told you who was on it, what bikes they had, and who was a strong climber versus who would need the sag wagon by mile thirty. You folded both and put them in your back jersey pocket. That was the system.
I left the field. Ran travel companies from the executive side for years. Came back to guiding last year partly to see what had changed.
Nothing had.
The answer, up front
Tour operators, travel advisors, DMCs, and trip designers still run their operations on paper, scattered spreadsheets, and email chains in 2026 because the software built for the industry was built for the office. Booking engines, CRM, financial reporting. The whole last decade of travel tech solved the desk. The field, the research hour, and the proposal deadline were left alone. AI changes what is possible, but only if an operator is willing to own the operational layer instead of letting ChatGPT own it for them.
Ten years ago, this was part of the expected
Same trip, different failure. Later that week in Mexico City, we could not reach one of the restaurants I had planned. A 10K was happening that Sunday and the park was closed to traffic. The van dropped us almost a mile away. We walked. We were thirty minutes late to the reservation.
Ten years ago, this was part of the expected. You planned, you confirmed, you printed, you showed up, and if the world rearranged itself overnight, you improvised. That was the job.
In 2026 there is no reason a guide should be caught by a street closure. I could have had an auto-scheduled daily itinerary audit running every morning, pulling local news, traffic, event calendars, and weather, cross-referenced against my specific pickup times, drop-off points, and reservations. It would have flagged the 10K before I poured my coffee. I would have rerouted the morning, pushed the reservation back an hour, and the guests would have never known anything had almost gone wrong.
Nobody has built that for tour operators. Not because it is technically hard. It is not. It is because the software companies that took the last decade of venture funding in travel built booking engines, and the people trying to fix operations right now have not been on the ground.
The five layers that still run on paper
Field coordination is the visible one, because it happens in front of the guest. The paper in the pocket. The tipping question answered wrong. But it is only the first of five operational layers that still run exactly the way they did before AI existed. Each one is leaking time and money at every operator-side business in travel, whether that business is a global tour operator, a luxury trip designer, a DMC, or a solo travel advisor.
Layer 1: Field coordination
What happens now. Guides carry printed documents because the office software will not fit in their hand while a guest is waiting. Co-leaders text each other and miss the thread. A pre-trip binder gets handed over at the briefing and lives in the glovebox because it has everything a guide needs and therefore nothing a guide can find. The only reliable system is the one the guide builds herself, usually the morning of the trip, usually on paper.
Before this Mexico City trip, I was a contractor on the ground. The office was designing logistics. I was helping with logistics. I was also looped out of roughly fifteen different email threads with vendors. I spent two hours the weekend before the trip piecing together which reservations were confirmed, which needed follow-up, what the current budget was, and what had already been discussed with each vendor. That is two hours the operator paid for a contractor to do archaeology instead of preparation.
What could happen. A shared operational hub where every decision, update, and vendor exchange is instantly visible to every person on the trip, in one view, organized by day and activity. The contractor on the ground opens the app on Saturday night and sees: here are today's seven bookings, here is each vendor's last message, here is the current budget for each activity, here are the dietary flags for each guest. No assembly required. The two hours go back into the trip.
The in-the-moment version is the same principle. If my co-leader had tipped the guide and logged it in three seconds on her phone, I would have known before I pulled out my wallet. Not because I was looking for it. Because the tool would have surfaced it when I opened the guide's name.
What it costs today. Two to four hours of guide time per trip on pre-trip assembly. Uncounted small losses like double-tipped cash. The slow, corrosive thing where every guide gets good at building her own workaround and none of those workarounds ever become company knowledge.
Layer 2: Expenses
What happens now. A decade ago, when I was guiding cycling trips, it took me about five hours after a trip to categorize my expenses. Paper receipts in envelopes, cash tips logged in a notebook, a currency conversion or two, a shared expense to untangle with my co-leader, and a spreadsheet to reconcile against the petty cash reimbursement.
In 2026, using the manual systems most tour operators still run on, it takes five hours. Same envelopes, different decade.
It is worse than that, actually. Guides forget. Every guide I know has tipped someone fifty dollars in cash out of their own wallet, meant to log it, and then forgotten by dinner. That money comes out of the guide's own pocket. On a multi-trip season, a guide can easily eat a few hundred dollars of untracked cash before anyone notices.
What could happen. I built this one myself on the Mexico City trip, because nobody wants to sit down to five hours of expenses after a trip. On the fly, I took a photo of every receipt and recorded a short voice note describing what it was for. In parallel, I built a small tool that internalized the operator's expense-cataloging PDF, the one that normally tells a human guide how to file each line. The tool learned the operator's categories, pair rules, and vendor list.
At the end of the trip, I opened the tool and hit one button. Every receipt was categorized, matched with its voice note, and paired against the right budget line. I spent one hour reviewing and approving. Five hours of work became one. On the same trip, on the same guide, on the same pile of receipts.
Cash tips logged live, too. As soon as I handed over the money, I said it into my phone: tipped the driver fifty dollars, part of the morning transfer allowance. By the end of the trip the number was already in the log. No envelope of mystery receipts. No tipped-and-forgot.
What it costs today. Four hours per trip per guide on expense reconciliation. Hundreds of dollars per season in forgotten cash tips that come out of guides' pockets. Plus the thing that does not show up in any budget report: guides burning out on admin work that has nothing to do with the craft of guiding.
Layer 3: Research and vendor work
What happens now. Operators pay researchers and trip designers to negotiate with vendors, design itineraries, and build proposals. A meaningful share of the hours the operator pays for go to assembly, not to the work. The researcher is handed a dietary list, a budget (which keeps changing), a vendor list, a guest profile, preferences from the sales call, and a brand tone guide. All of these live in different places. Before the researcher can start researching, she has to piece them together.
I designed a Danube river cruise for a large active travel company a while back. The design phase was a long chain of back-and-forth emails about budgeting requirements, vendor availability, revised guest lists, and updated pricing from the ground operator. I am good at this kind of work, and it still took longer than it should have, because the assembly layer was manual. Every time a guest swapped out, I re-opened the thread and reread context. Every time the budget shifted, I re-ran the math. Every time a vendor quoted new pricing, I copy-pasted it into the master.
What could happen. A system where the researcher opens a trip and the context is already assembled. Current guest list with dietary flags and past-trip notes. Current budget with the pending changes flagged. Every vendor's last quote and last conversation already summarized. The research phase starts with synthesis instead of with archaeology.
The deeper version is that the system does not just assemble, it evaluates. The researcher drafts the itinerary and the tool cross-checks it against guest notes. Would this itinerary appeal to the Martins, who on their last trip said they wanted more downtime? Would it clash with the Kim family's observed preference for early dinners? The tool flags the mismatch before the proposal ever ships. The itinerary reaches the guest already pre-audited against what is actually known about them.
What it costs today. Anywhere from a third to half of a researcher's paid hours, depending on the complexity of the trip and the messiness of the operator's inputs. That is money the operator is paying its own people to do administrative glue-work instead of the negotiating and designing they were hired for.
Layer 4: Guest knowledge
What happens now. Every trip begins with a guide staring at a guest roster and trying to internalize twenty people. Who are they, where are they from, what do they like, what do they need, what do they not eat, and what have previous guides noted about them. That last part is the weakest. Most of what is known about a returning guest lives in short notes at the bottom of past trip reports, if they exist at all. "Great attitude, loves a good pisco sour" is a typical entry. Useful, but thin, and often the previous guide wrote it from memory a week after the trip ended.
Meanwhile, during the trip, the current guide is learning new things constantly. Cindy loves orange juice. Marcus has knees that cannot handle the morning hike. The Kim family prefers eating early. Most of this never gets captured, because capturing it means sitting down at the end of the day and typing notes into a report, and by the end of the day the guide has been up for sixteen hours.
What could happen. On the ground, every guide has a tool for capturing live notes in two seconds. Voice or tap. Cindy loves orange juice. Marcus, knees, no more morning hikes. The Kim family, early dinners, no exceptions. At the end of the trip, the guide opens the report and every note is already there, organized by guest, ready to be polished and submitted.
The compounding effect is where this gets powerful. The next guide who has Cindy does not start from zero. The system surfaces her notes before the trip starts. Her preferences, her quirks, what landed last time, what did not. The guest is known before she arrives, and known better than she would have been even if she had written in to the office with preferences, because the notes come from a guide who watched her for seven days.
What it costs today. The guest experience ceiling. Every operator's brand promise is personalized service. The industry is running that promise on sparse, retrospective, lossy notes. The guests feel the gap, even if they cannot name it.
Layer 5: Trip design and proposals
What happens now. Roughly every researcher I know is using ChatGPT. Not because the company adopted it, usually. Because it is faster than starting from a blank page. They draft the itinerary in ChatGPT, paste it into the company's Word template, and reformat. The AI is doing the design work, and the company is paying for the copy-paste.
This is the most urgent of the five, because it is the one where the train has already left the station. Operators are reporting that stitching AI into the existing proposal process does not save as much time as they expected. The AI output drifts from brand. It does not know past trips. It does not know the specific guests. The researcher spends the time she saved on drafting on the new task of reining the output back in.
What could happen. The operator owns the layer. The AI is trained on the company's brand, past trips, vendor relationships, and the specific guests the proposal is for. The researcher opens a new proposal and the first draft is already in the brand voice, already structured according to the company's itinerary format, and already audited against what is known about the guests.
Beyond the proposal document, the whole trip design phase gets rethought. Itineraries get tested against real guest data before they ship. Proposals go out hours after the discovery call instead of days. Sales cycles compress. Close rates go up because the proposal arrives while the lead is still warm.
What it costs today. A generation of trips being designed by an AI the company does not own, in a tone the company did not set, against guest context the AI does not have. Every operator is effectively outsourcing its most strategic work to a chat window.
Why the apps didn't fix this
The travel-operator software market is not small. Tourplan, TourWriter, Lemax, Rezdy, Peek Pro, FareHarbor, Xola, Bookeo on the booking and reservation side. Vamoos, mTrip, and Simplified.Travel on the guest-companion side. Plus the homegrown Google Sheets systems most operators actually run on.
None of them show up in a guide's back pocket. None of them tell a researcher which vendor thread is current. None of them catch the Sunday 10K.
The reasons are structural. Most of these tools were built to solve booking, payment, and back-office reporting, because that is where the money in travel B2B software has been. The operational layer, what happens between the office and the guest, was addressed by exporting a PDF or generating a guest-facing app that the guide could not edit. Nobody built the hub where the guide, the co-leader, the office, the researcher, and the vendor could all operate in the same shared reality.
That gap is the thing. It requires deep knowledge of how trips actually run. Most of the people building software for this industry have never been on the ground. For the broader picture of which workflows to tackle first across the operation, see the 12 tour operator workflows worth automating with AI, ranked by ROI.
What changes for the guest
When these five layers work, the trip feels different to the person on it.
The guide is present. She is not pulling out a printout twenty times a day, because the information she needs is on her phone and she already checked it at breakfast. She is not distracted by whether the next restaurant is still open, because her morning itinerary audit already answered that. She is not mentally tracking her own expenses during dinner, because she logged them between courses in three seconds. She has attention left for the guests.
The guest is known. The guide greets Cindy on day one already knowing she loves orange juice, because the last guide wrote it down in two seconds six months ago and the system surfaced it this morning. Cindy does not have to explain herself again. She feels like a regular at a restaurant she has never been to.
The trip bends. A road closes, a storm comes in, a guest gets sick, a vendor cancels. None of these break the trip because the system catches the disruption the morning it happens, not the afternoon the van arrives at the closed park. The guest might never know anything had to be rerouted.
That last one is the whole business. A tour operator's product is not the itinerary. It is the experience of the itinerary unfolding without friction. Every one of these five layers is either adding friction or removing it. Right now, almost all of them are adding it.