Roadbook Quality Check: What We Verify Before It Ships

A roadbook quality check verifies, before a trip goes out to the traveler, that every night and every leg of the itinerary line up, that mandatory fees are disclosed, and that no single day pushes past the pace the client's profile can actually handle. It's the last line of defense before the traveler finds the mistake themselves, on the ground.

Why a travel itinerary can't afford to be wrong

A client paying for a tailor-made trip isn't just buying an itinerary. They're buying the assurance that someone already thought through everything they didn't want to think about. The moment they find out at the Chichen Itza ticket booth that the entrance fee wasn't in the price, or that breakfast opens after their scheduled departure, that assurance is gone. It doesn't matter how flawless the rest of the trip was.

That's what makes this job hard in a specific way: you get judged on the one detail you missed, never on the ninety-nine you got right. I've lived this from the inside with Mexikoo for over ten years, and the thing that actually changed it wasn't hiring people who pay closer attention. It was building a process that doesn't depend on how sharp the person reviewing it is at midnight, the night before it goes out. At Mexikoo, the scan has become systematic: it runs once bookings are finalized, and always before clients arrive.

This picks up where an earlier piece left off, on the AI tools I actually recommend for a receptive agency, where I explained why I chose Claude over ChatGPT and Gemini to run Mexikoo. This quality check is the first of the five skills I named there. The other four will follow in later articles.

What a quality check actually verifies

The method I use runs four passes, in a specific order, because the order changes what you find.

Mechanical consistency first. Does every night have accommodation? Does every change of city have a way to get there? Do the dates fall on the right days of the week? These are the fastest, highest-yield checks: an orphan night or a mismatched date is an automatic red flag, and it's exactly the kind of thing a human eye stops seeing after the third read-through.

Pace and feasibility next. Each day gets chained hour by hour: implied wake-up, breakfast service, travel to the meeting point, activity length. Then you look at accumulation across days: two pre-6am starts in a row, a long drive after a late evening, a long-haul arrival followed by a packed day. The same day works fine for a couple in their thirties and it's brutal with a four-year-old.

Accuracy next. Every stated time gets cross-checked against something verifiable: check-in and breakfast hours, closing days, the season of a promised natural phenomenon. A turtle hatching promised out of season isn't a small detail. It's a promise you can't keep.

Completeness last, filtered by one question: would the traveler have acted differently knowing it? A hotel's postal code never changed anyone's behavior. Breakfast opening hours on an early-departure day would. Run completeness without that filter and you end up with forty flagged fields nobody reads. Run it with the filter and you're left with the three or four gaps that actually matter.

There's one rule I'd call the single highest-leverage move in the whole method: the exclusions line. Most travel booklets list what's included and go silent on what isn't. That silence is exactly where undisclosed entrance fees, uncovered meals, and optional supplements hide. Forcing an explicit "not included" line next to every paid service is, across most documents I've reviewed, the one structural fix that closes the most gaps at once.

Why you never correct a figure by feel

There's one non-negotiable rule in this method: never invent a correction. A wrong number in a quality audit is worse than no number at all, because it makes a correct document wrong, and it destroys trust in the other fifty findings in the report.

In practice, every checked item carries one of three states, never a fourth: [VERIFIED] with a source and a date, [CONSISTENT] cross-checked by calculation or against another part of the document, or [TO CONFIRM] not verifiable online, flagged for the supplier. That third state isn't a failure. It's a work instruction. You never swap it for a value that sounds plausible.

And the reverse matters just as much: you never flag an absence as if it were automatically an oversight. A dietary requirement can be handled perfectly well behind the scenes with the supplier, without ever appearing on the page meant for the traveler, and there's no reason to remind someone on their own itinerary that they're vegetarian. What triggers an alert is never absence. It's contradiction: the document announcing a service that's incompatible with something already known.

A real example, on a test document

To write this piece, I ran the method on a Yucatan test roadbook. The travelers, agency, hotels and prices are entirely fictional; external facts and rates were checked online in September 2026. The scan starts with what works: all eleven nights connect, both Quintana Roo/Yucatan time changes are clear, and the first day handles jet lag well. Then it surfaces four concrete findings.

Blocker — undisclosed entrance fees. The document listed only the Chichen Itza parking fee ("around 120 MXN"). The mandatory entrance fee appeared neither as an inclusion nor on an exclusions line. In 2026 it runs roughly 680 to 700 MXN per foreign adult[1]. The same gap appeared at Uxmal, at around 566 MXN[2].

Blocker — six hours without a proper meal. A ten-hour excursion started at 7am with a young child, while breakfast opened at 8am and lunch was not scheduled until around 1:30pm. Neither fact was wrong on its own; their overlap revealed the problem.

Risk — one activity in the wrong language. A four-hour cycling activity was offered in English and Spanish while every other service for this family was planned in French. The scan does not invent a correction: it asks the team to confirm a French-speaking guide or tell the travelers.

Hospitality — extend a meaningful encounter. After an activity with a Maya family, the report suggests packing a few handmade tortillas to share at dinner. A simple, low-cost gesture anchored to a real itinerary moment.

Get the Roadbook Quality Scan skill and the complete Yucatan report by email.

Can AI replace a DMC's human review

No, and it's worth saying that plainly instead of selling the opposite. A 2026 study across 356 AI-generated itineraries and 2,735 planned days found that roughly 43% of days carried at least one verifiable fault: a venue closed at the scheduled hour, a visit running past closing time, a day that backtracks across the map[3]. A March 2026 survey of more than 7,000 travelers points the same way: nearly half of those who used AI to help plan a trip ran into incorrect or outdated information, most often around hours, pricing, or availability[4].

My take isn't that AI should stay out of tailor-made travel. It's that it needs to stay in its lane: the first pass, not the final word. AI can absorb the mechanical part of a review, the part that's pure calculation and cross-checking, and that takes hours from an exhausted human on a Friday night. What it can't do: be there when a flight gets cancelled at 11pm in Cusco, call the right doctor in Oaxaca, sense that a traveler isn't doing well and improvise a detour, know the local guide who doesn't have a website.

The real question for a DMC today isn't human versus machine. It's what you do with the time you get back once the machine takes the mechanical part off your plate.

This skill isn't meant to sit outside Odys as a standalone tool. It's one of the architectures we're building for Odys Copilot, the agentic layer we're developing for DMCs on the Odys network. It's already running at four agencies, Mexikoo included, but it's still exploratory, with pricing not yet set. Downloading it on its own is a way to try it before it's built in natively. The beta program for Odys Copilot is open if you'd rather test it wired straight into your own production.

And what about the unreasonable hospitality part

There's a second half to this method, kept entirely separate from error-hunting and never mixed with it: spotting where to place an attention nobody asked for. The philosophy, borrowed from the idea of unreasonable hospitality popularized by Will Guidara (former restaurateur at Eleven Madison Park), is easy to state and hard to do well: service is delivering what was sold, correctly. Hospitality is what you add on top of that.

Three rules keep a suggestion actually useful instead of just nice. Personalized, or nothing: a generic idea is noise, every proposal has to hang on a specific element of the profile or the itinerary. Anchored to a real moment of the trip: name the day, the time, the place, never "add a nice touch somewhere." Doable by the team: an idea nobody will execute has no value, favor what costs little and can be prepped in advance.

This is where I think the real value of a DMC sits for the years ahead. If a traveler tells us during the brief that they love magic, we can arrange a private moment with a local magician we know. If someone does classical dance, we can get them into a local company's private rehearsal, backstage. None of that gets improvised the day before. It gets prepared because someone actually read the brief and has a contact book Google doesn't.

That's exactly where I see AI as leverage, not a threat. The time it hands back by absorbing the mechanical review doesn't go toward something else. It goes back into the one thing only a human can do: knowing the magician, calling the dance company, being there when it counts. AI in the back office, the human out front, more present than before, not less.

FAQ

How long before departure should this check happen? As late as possible before the document ships, but with enough margin to act on what turns up: a missing entrance fee gets fixed in five minutes, confirming a French-speaking guide's availability can take days.

Who should review the roadbook, sales or operations? Both, because they're looking at different things. Sales knows what was promised to the client; operations knows what's actually feasible on the ground. A good check crosses the two instead of putting the whole job on one person.

What are the most common mistakes in a travel itinerary? Undisclosed entrance fees top the list, followed by breakfast hours that don't line up with an early departure, and leftover fragments in the wrong language from a copy-paste in production.

Can AI replace a DMC? It can absorb the mechanical review of an itinerary. It can't be there on the ground when something goes wrong, and it doesn't know the local supplier with no website.

What is unreasonable hospitality? An attention the traveler didn't ask for, anchored to a specific moment of the trip and a real element of their profile, prepared in advance by the team on the ground.

Does this skill replace Odys? No. It scans and prepares corrections; Odys stays the system that owns the data and runs production. It's one of the building blocks integrated natively into Odys Copilot, our agentic layer for DMCs, currently in beta.

A travel itinerary never needed to be perfect to be good. It needed someone to read it as if they were about to board that flight themselves.