A2A Research · Report 12 · National · Agency Operations · August 2026
One long interview with an owner-operator whose AI stack actually works, read against the national data on why most do not. The difference was never the model.
01 · The objection
If you are tired of being told you are using AI wrong, that is a reasonable place to stand. The tools are in the building. The work still takes the same number of people.
You bought the seats. Your team drafts with it daily. You have watched a demo compress a week of work into a morning, and watched your own margin stay exactly where it was. Between those two facts sits a suspicion that the problem is you — that a better prompt, model or vendor is the missing piece. It is not. The gap between agencies getting leverage from AI and agencies getting faster drafts is measurable, and has little to do with which tools they bought.
Ninety-nine percent of agencies use AI. Their number one operational complaint is inefficient process. Both are true at the same time, and the second explains the first.
02 · What came first
This report is built on a long interview with the owner of a done-for-you agency running roughly fifteen accounts in a single construction-adjacent vertical. His stack works. Client onboarding produces a complete brand and campaign foundation — positioning, offers, audience profiles, creative direction — from a fixed sequence of prompts run against an intake form, the client’s site and their customer list. Campaign build time went from about a week to half an hour. Then the part that matters. Before any of it, he spent six and a half years on the phone at an average of 250 dials a day. He can describe the contact sequence he later automated hour by hour, because he ran it himself several million times. The machine did not discover his process. It inherited one.
He was blunt about the ordering: the AI did not make him good at follow-up. It let him stop personally being the follow-up. Only one of those is available to an agency that never wrote its process down.
03 · Knowledge is not the constraint
Lead follow-up is the most heavily measured process in commercial life. Its optimal shape has been published and taught for fifteen years. It is still almost never executed.
Velocify, The Ultimate Contact Strategy (~3.5M leads) and XANT, Lead Response Report (55M interactions, 2021) — vendor platform data, labeled as such. Separately, a Harvard Business Review audit of 2,241 US firms found 23% never replied to a test lead at all.
A process nobody sustains is indistinguishable from a process nobody has. The machine’s advantage is not that it is smarter on any single call. It is that it makes the fourth one.
04 · The precondition, tested
The obvious objection: well-run firms adopt technology well, so process discipline may be a symptom of competence, not a cause. One clean test exists. Bloom, Eifert, Mahajan, McKenzie and Roberts ran a randomized field experiment on large Indian textile plants, giving free management consulting to a randomly chosen treatment group and comparing it against controls. Adopting structured practices raised productivity 17 percent in the first year. The third recorded effect is worth pinning to a wall: computer use went up, because the data collection the new process demanded created the need for it. The process came first and pulled the tools in behind it. The authors also asked why firms had not adopted profitable practices on their own. The answer was not capital or willingness. It was information — owners did not know what good looked like.
That finding cuts both ways. It supports the ordering, and warns that “write your process down” is not sufficient by itself: writing a bad process down faithfully buys a machine that runs a bad process faithfully.
05 · Where the agency dollar lands
The shape is not an accident. Ideation and research are forgiving: a strategist edits everything, and no workflow has to change to accommodate them. Media planning and buying strategy are where an agency’s judgment converts into a client outcome — and they are the two lowest-adoption uses on the list. The forgiving work got automated. The valuable work did not — automating it would first require writing down how it is done.
Agencies are not underusing AI. They are using it precisely where using it changes nothing.
06 · The five signs — a five-minute mirror
Five honest answers about your own firm. Nothing you tap here is stored or sent anywhere — it runs entirely on your screen.
Five honest answers. That’s all the mirror needs.
07 · What the working stack had underneath
Not the operating manual. The single sequence that converts attention into revenue at your shop — new business, onboarding, or account renewal. One document, written by whoever actually does it well, at the level of detail where a competent stranger could follow it without asking a question.
Pick one number that would move if the process improved — response time, hours per campaign build, accounts properly reviewed per quarter — and measure it for thirty days while changing nothing. Fewer than one adopter in five does this, which is why most AI spend can never be argued about in either direction.
Redesign ranked first among twenty-five organizational attributes tested against bottom-line impact. Acceleration leaves a workflow’s shape intact; redesign asks which steps stop existing altogether. Do it once, all the way through, before doing it anywhere else.
Not the work you find boring — the work that is done well the first time and badly the ninth. Follow-up cadence, account review, the quarterly thinking every client is owed and only the loudest ones receive. That is where consistency is worth more than brilliance.
None require a new vendor and three cost nothing but an afternoon. That is deliberate: if a move only works after a purchase, it is a sales argument wearing a recommendation’s clothes.
08 · The honest limit
This is one interview, and a successful operator’s account of why he succeeded is the least reliable evidence in business — filtered through survival, assembled after the fact. It appears here as illustration, not proof. The national data is stronger and still not clean: McKinsey’s model ranks workflow redesign first of twenty-five attributes and explains roughly a fifth of the variance. Redesign is the best single lever on the list. It is not destiny.
The causal arrow may run backwards. Firms disciplined enough to document a process may simply be good at everything, in which case documentation is the tell rather than the cause.
The randomized experiment is the counterweight, and it was run on textile plants rather than agencies — a real limit on how far it carries. So here is the test, applied to yourself rather than anybody else: if you cannot describe in writing the process that produced your best account result last year, no finding on this page tells you what to buy. It tells you what to write.
09 · The problem underneath
It is hard because it must be repeated — for the next account, and again next quarter for all of them at once. The process is not mysterious; most owner-principals could describe theirs out loud. It simply does not survive being performed forty times a year by a tired person. What gets skipped is never the client complaining. It is the quiet, healthy account nobody has looked at properly in eight months — the one that leaves without warning.
The constraint on strategic quality in a small agency is not talent and it is not knowledge. It is repetition without decay — the one thing a person cannot supply and a system can.
That gap is the reason A2A exists: an account intelligence layer that runs the same disciplined read across every account in a book, on a schedule, whether or not anyone remembers to ask. We would rather have that claim tested than believed. Report 04 covers where the agency AI dollar actually pays.
If you disagree with any of this, that is a conversation worth having in the open. The argument gets better under pressure, and the report gets corrected when it is wrong.
One useful next step — and it isn’t a sales call
Pick the sequence that produces revenue and document it well enough for a stranger to follow. If you want a second pair of eyes on what you wrote — or an argument about whether it is the right one — that is a conversation worth having.
Request a conversation