I've been spending time with Account Executives who are new to out-of-home, because I wanted to understand something specific.
This is the first cohort entering the industry for whom AI is not a tool they adopted but a way of working they already had. They used it at school, they used it at the job before this one, and they arrived fluent. I wanted to know what someone like that does when you hand them a billboard network and a quota.
What I found is that they build things, quietly, that the operators employing them have never seen. And it reframed a question I thought I understood. The interesting one is not whether this industry should adopt AI. The people it just hired settled that on their own. The question is what happens to the work they're already producing with it.
What they had already built
They had all been in out-of-home under a year, and none had come from the industry. Each had built their own sales process out of consumer AI, alone, without being asked to.
Start with one of them, and the onboarding that came before. A manager taught the right things: know your traffic counts, know your seasons, know who is passing the board. That is the craft, correctly stated. The manager had a number to carry too, so a single day was all there was to give. After that came about ninety minutes on the operator's system of record and some version of go figure it out. That is not negligence. It is what happens when the only copy of the knowledge sits inside a person who is also carrying a quota.
Handed a job and no instructions, they did what anyone their age does. They opened ChatGPT and started asking it how to sell billboards.
Within a few weeks they had built something more sophisticated than that. They loaded a file of their inventory into a project so that, as they put it, the tool knew what they were selling. When they needed to plan a run through a part of their territory they didn't know, they'd ask it who was worth pitching there and start the day from the list. When an agency sent an RFP, they'd paste the brief in, ask which of their boards fit and why, and work from the answer. They closed business in their first month off that process. They'd have told you they were just muddling through.
That phrase, the tool knew what they were selling, is where the whole thing turns, and it is worth slowing down on. A board is not its dimensions and its coordinates. It is the corridor it sits on, who moves past it and when, the trade area around it, the seasons that swing its audience, the businesses it faces. A file with a board list in it gives a general model the labels and almost none of that. So the tool did not know what they were selling. It knew the names of the things they were selling, and produced confident prose about the rest. They had no way to see the difference, because seeing it would require exactly the location context the tool was missing. The gap was invisible from where they sat, and it stays invisible until someone checks the output against what is actually true on the ground.
Their company had not left them with nothing. They had good tools for the parts of the job those tools cover, the operational system, the customer records, the platforms that profile the businesses they call. None of it was built to answer the question they took to ChatGPT: why one of their boards is right for a given advertiser. That question sits between the customer and the inventory, and they were answering it in a separate window, on a personal account, with a model that could not see the boards.
Another was running the same play from the other direction. They sell in markets they have never physically been to, so they open the map, read the geography cold, and try to assemble enough local knowledge to sound credible on a call. They described it as building the market knowledge alongside the research, which is a precise way of putting it. They are doing two jobs at once and only one of them is selling.
Why this is a 2026 problem
This is a 2026 problem, not a 2019 one, and the reasons all happened outside your company. Consumer AI got good enough to produce work that survives contact with a customer, plausible if not always correct, which is the more dangerous threshold. It got cheap enough to adopt on a personal card. And the people entering the industry arrived already fluent, so it needed no push from the top and left no trace.
What that changed is not whether a new Account Executive struggles. They always did. It changed whether the struggle is visible. No support used to mean thin pitches and a slow pipeline a manager could read in the numbers by month three. Now it means fluent, confident pitches in week two, with no way to tell from the outside whether the substance is real. The failure mode moved from underperformance to invisibility. Underperformance you can manage. Invisibility you cannot.
Where it comes apart
The same failures came up independently, from different people, which is what made me pay attention.
Nobody can check it. I asked one of them whether the tool had ever let them down on quality. Probably, they said, but they'd never know. And then, more starkly: it could have made all of it up, and they'd have sent it anyway. Sit with what that means. A general-purpose model will produce a confident paragraph about a trade area it has no real information about, and the person best positioned to catch that error is someone with twenty years in the market, who can feel that a claim is wrong before they can explain why. The person least positioned to catch it is a new Account Executive with no ground truth to check against, which is exactly the person for whom the tool is most useful. The value and the risk land on the same desk.
Feeding it costs as much as the work. One of them put it as spending as much time giving it context as the work would have taken anyway, maybe more, because it's such a broad tool. A general model doesn't know your boards, your corridors, your seasons, or your rates, so every task begins with an act of manual translation. One of them solved it by maintaining an inventory file. Another solved it by describing the same market from scratch, repeatedly. Both were doing unpaid data entry to make a consumer product behave like an industry one.
None of it belongs to the company. That process lived in a personal account. It got better every week they used it, and every improvement was theirs alone. When they left, the whole thing went with them. A colleague sitting at the next desk wasn't using AI much and got no benefit from any of it. This industry talks constantly about the institutional knowledge that walks out the door when a thirty-year veteran retires. The same thing is now happening at nine months, and faster, because nobody thought there was anything there to protect.
It runs across four tools. Chat in one window, the operational system in another, the inbox in a third, and, as one of them counted it, about five separate searches to work out what sits near a location and what is happening there this month. Every handoff is a place for the work to degrade.
Coverage depends on who the Account Executive happens to be. Every Account Executive sees the opportunities their own background primes them to see, and misses the rest. The sports fan jumps on a game-day signal the moment the schedule drops. The one who came up in finance notices tax season is a quarter out and starts working the accountants. Someone who follows local politics catches the new development before it breaks ground. None of them are wrong, and none of them see what the others see. Which opportunities a territory surfaces comes down to the accident of who is working it, not to anything anyone decided. A whole category of business can go unpitched simply because it never landed in the right person's field of interest.
The wrong conclusion to draw
An owner reading this could reasonably land on a policy. Write it down, circulate it, prohibit unapproved AI in customer-facing work. I'd argue against that, and not on principle.
Start with why the tool got reached for at all. Nobody opened ChatGPT looking for a shortcut. They opened it because they had been handed a market, a quota, and almost no support, and it was the only thing on the desk that helped. A prohibition takes the help away and leaves the gap exactly where it was. The pitch still has to be written, the unfamiliar territory still has to be learned, and now it happens without even the imperfect tool that was getting it done.
Then look at what you would actually be prohibiting. The work in question produced revenue inside four weeks, at a company that had been able to offer one day of training. The tool was not the problem. The problem is that the work was unsupervised, unverifiable, and evaporated the moment its author walked out the door. A ban addresses none of those three. It moves the same activity somewhere you cannot see it at all, which is a worse version of where it already is.
Range is not help
There is a single mismatch under all of these failures. A general AI platform is an open canvas: ask it anything, build any workflow, point it at any problem. For someone who knows exactly what they want built, that range is the value. For a new Account Executive it is a tax. An open canvas hands the person with the least command of the job the work of designing their own support, knowing what to ask for, what good output looks like, and where the tool is quietly wrong.
Build any tool you want is a generous offer. It is not a better offer than a screwdriver when what you need is a screwdriver. The newest person on a sales team does not need a workshop and an instruction to invent their own instruments. They need the specific tool for the specific job, already shaped and already correct, so the judgment lives in the tool instead of resting on the person least equipped to supply it.
Shadow process, or infrastructure
None of this is an argument that any of them did anything wrong. They did something resourceful with nothing, and one of them turned it into revenue inside a month.
The argument is that an operator now has a choice it didn't have three years ago and mostly doesn't know it's making. Your Account Executives are running an AI-assisted sales process either way. It is either a shadow process, invisible, unverifiable, rebuilt from scratch by each new hire and owned by whoever leaves next, or it is infrastructure the company can see, trust, and keep.
Building the second version is what we are doing at doohthis, and it meets those failures point for point. It runs on proprietary tooling built for out-of-home, not a general model pointed at billboards, and it holds your inventory natively, the corridor, the audience, the trade area, the season of each board, so nobody pays the context tax and the scattered windows become one workflow.
The rest follows from that. Every claim it makes is drawn from real, observable data with the source in view, so nothing ungrounded reaches an Account Executive and nothing invented gets passed along. It surfaces the opportunities across every board and category, so coverage stops riding on what one person happens to know or follow. The work accumulates in the company instead of a personal account, still there after whoever built it has gone. And the credits are priced under a personal subscription, so the version that is supervised, verifiable, and owned is also the cheaper one.
The frame worth holding
There is a version of this that matters more than any software decision.
The people in this industry who can look at a board and know which advertiser belongs on it built that over decades, and almost none of it exists anywhere outside them. One of the newer Account Executives got the real thing from a manager in a single day and still remembers it. The constraint was never the veteran's willingness to teach. A day was simply all there was time for.
The people arriving are filling the rest of the gap with tools nobody gave them, trained them on, or checks. Discussed as two problems, they are one: the knowledge the veterans never wrote down, and the private, unverifiable replacement the newcomers are building for it. doohthis is where both can live, checkable and shared, so what the veterans know and what the new Account Executives work out stop walking out the door and start compounding inside the company.
The newest person on your sales team has already written the first draft of your sales process. doohthis is what makes the finished version the company's, and keeps it there long after the author has gone.
doohthis builds proprietary AI tooling for out-of-home operators — the kind that holds your inventory natively, verifies every claim, and keeps the work inside the company. If you're thinking about how to turn your team's AI workflows into company infrastructure, we'd like to talk.