Every conversation about deploying AI into a company hits the same wall, usually in the first ten minutes: "our business is different." Payments has regulation. Gaming has compliance and churn. Health has claims you cannot make. Logistics has physical reality. Each industry is convinced its complexity is the moat that AI, and outsiders in general, cannot cross.
Having crossed a few of those moats myself, I can report what is on the other side: the same five bottlenecks, wearing different uniforms.
The view from the deployment seat
Strip any business to its loops and it stops looking like an industry and starts looking like a machine. Leads come in and someone qualifies them. Content gets produced and someone approves it. Questions arrive and someone answers them from knowledge that lives in three heads and one outdated document. Reports get assembled by hand every Monday. Decisions that follow a pattern get made one at a time, slowly, by the most expensive people in the building.
That list is not from any particular industry. It is every industry. What changes across domains is the vocabulary and the constraints. What does not change is the shape of the work, and the shape is the thing you automate.
The bottleneck is never "payments" or "gaming." It is judgment applied at volume. Context scattered across heads. Handoffs where information dies. Work that repeats but resists a template. Those are the five or so shapes I have found under every hood I have opened, and each one has a known deployment pattern.
I have crossed industries my whole career
I started in advertising as a designer, working on cars, soft drinks, airlines, banks, and oil in the same building. Nobody asked whether I had automotive experience before handing me an automotive brief. The craft was the constant; the industry was an input.
Then I crossed into performance marketing and consumer health, a domain I had zero background in. It took weeks, not years, to become dangerous in it, because the method carried over: understand the customer, find what the market rewards, iterate against the number. Now I build and run AI systems at Uponly, my own company, doing work that did not exist as a job when I started my career.
Every crossing taught me the same lesson. Domain knowledge is real, but it is the cheapest ingredient in the room, because your company already owns it. It is sitting in your operators, your compliance leads, your best support agent. What you do not own, and what does not come with the industry, is the ability to turn that knowledge into running systems.
The method does not care about your vertical
Here is what I actually do when I walk into a business, and notice how little of it depends on the industry.
Find the loop that hurts. Volume, repetition, a clear definition of done, expensive people doing it. Sit with the person who owns that loop and extract what is in their head: the facts, the standards, the examples of good and bad, the reasons behind the exceptions. That extraction is the real work, and it is precisely where "our business is different" turns out to be true and useful, because the difference is content, not structure. Then wire a small agent to that context, put its output where the work already flows, and give the owner a kill switch. Measure work cleared, not applause. Let the first loop teach you the second.
The domain expertise enters the system through your people. The machine that extracts it, encodes it, and runs it at volume is what I bring. I do not need to know your industry on day one. I need your best operator for two weeks, and the humility to know the difference between what I bring and what they bring.
What is actually rare
Plenty of people know AI. Plenty of people know business. The shortage is people who have both at operating depth: who build with these tools every single day, hands on the wiring, and who have also spent seventeen years in rooms where work had to survive clients, budgets, and the market, not just a code review.
That combination is what lets an outsider move fast in your domain without breaking things. The builder half ships the loop. The operator half knows which loop matters, what the output is worth, and when good enough is good enough. Either half alone fails in a predictable way: the pure engineer automates something nobody needed, and the pure strategist writes a deck about automating something nobody built.
The first ninety days
So when a payments company, or a gaming company, or any company asks what I would do inside their walls, the answer does not start with their industry. It starts with an audit of loops. Weeks one and two: find the three bottlenecks, pick the one with the best ratio of pain to wiring. By week six: the first system is running in production on real work, owned by a named person, measured on throughput. By day ninety: two or three loops live, standards written down, and the company knows how to do the next one without me.
Your industry is special to you, and it should be. Your bottlenecks are not. That is not an insult. It is the best news available: it means the path out is already mapped, and someone has walked it before, just wearing a different uniform.