The impossible job at the centre of modern business.

I set out to build marketing software and accidentally discovered I was building for the most structurally impossible job in the boardroom.

Every other executive is measured against one definition of success.

The CMO is measured against two that contradict each other.

Be the guardian of the brand. Prove the return to the penny.

Everything I learned building Bertha began there — and everything the industry’s resistance to AI is trying to tell us lives in the same place. Because the arrival of AI does not soften that contradiction. It makes it sharper. AI can, for the first time, genuinely answer the ledger — measure, attribute, prove. The question nobody in the industry agrees on, the one every argument about “AI in marketing” is really a proxy for, is whether in doing so it destroys the magic.

I am a technologist by trade, and I walked into that room with a technologist’s quiet arrogance: here is a domain full of soft language and unfalsifiable claims, and I am going to render it into functions, endpoints and a schema. Marketing, I assumed, was mostly inefficiency wearing a nice suit.

I was half right, which is the most dangerous thing to be.

The mystical world I walked into

Software says: here is what it does, here is the latency, here is the price. Advertising says nothing so vulgar. It runs on theatre — the pitch, the reveal, the room going quiet, the idea that lands like a magic trick and cannot quite be explained afterwards. I went in expecting a factory and found something closer to a séance.

And the people humbled me. The CMO who could recite her churn to two decimal places and then, in the same breath, kill a technically perfect campaign because it “didn’t feel like us” — and be right. The founder running his own marketing on instinct and a phone full of screenshots, who understood his customer better than any dashboard I could have built him. The old-school creative director, all contempt for measurement, whose thirty-second idea I watched move a market.

I came to systematise these people. Instead they taught me that the thing I most wanted to automate — taste, timing, the read on the culture, the nerve to say the unexpected thing — was the one part that refused to sit still long enough to be typed into a schema. Every time I thought I had captured “the magic” as a feature, it had already moved. That is the mystical world, and it is not a con. It is a genuine skill that hides from the people trying to price it.

But every magician in those rooms answered to someone holding a ledger. And that is where the job actually lives.

The contradiction, up close

I said it at the top, but it is worth proving, because the impossibility is structural and not a matter of talent.

The CFO is asked to be precise. The CRO is asked to close. The CTO is asked to make it work and not fall over. Each is a hard job, but a coherent one — a single definition of success. The CMO alone is handed two that actively fight each other. The ledger wants marketing to behave like every other line in the P&L: spend in, revenue out, a return you can attribute and defend. The magic — the brand, the reason a customer pays more for the same molecules, the trust that survives a bad quarter — is the most valuable thing the company owns and is almost entirely unmeasurable in the timeframe the ledger demands. John Wanamaker said it more than a century ago and it has never been retired: half the money I spend on advertising is wasted; the trouble is I don’t know which half. The CMO is the person paid to stand inside that sentence.

Every other executive is measured against one definition of success. The CMO is handed two that contradict each other, and asked to be both people in the same meeting.

This is why the CMO has, for years, carried one of the shortest tenures in the C-suite. Spencer Stuart’s 2025 study puts the average at about 4.3 years — below the roughly 4.9-year C-suite average, and a long way under the CEO’s seven-plus — and it sank to around three-and-a-half years at the start of the decade. Not because marketers are worse at their jobs, but because the job is structurally impossible to win cleanly. Deliver the magic and the CFO asks where the return is. Deliver the return and the brand guardians say you have hollowed out the thing that made the return possible. Stand in that crossfire long enough and you are eventually asked to leave, having done nothing wrong except hold a contradiction that was never yours to resolve alone.

So I stopped designing for “a marketer” and started designing for a person under a very specific pressure — someone who walks into a demo not wanting features but carrying a two-headed fear. One head: I cannot prove what I do, and the people who control my budget are losing patience. The other: if I make everything provable, I will have to stop doing the unprovable thing that actually works — and then what am I? You do not sell to that person with a longer feature list. You are not competing with other tools. You are competing with their fear.

What Bertha was actually for

I thought I was building a content machine — take a brief, produce more marketing, faster and cheaper. The market is full of those, and they are almost entirely beside the point.

More content was never the CMO’s problem. By 2026 the cost of producing marketing has collapsed to roughly zero. Regular generative-AI use had already passed two-thirds of organisations, with marketing and sales among the first functions it reached; anyone with a subscription can now generate a thousand competent ads before lunch. Abundance was never going to be the scarce thing. What the CMO is starved of is not output. It is the two things the contradiction denies them: proof on one side, protected judgement on the other. A machine that just makes more stuff pours water into the half of the bucket that was already overflowing.

AI made content abundant. It did not make judgement abundant. The value moved.

That pattern appears everywhere AI touches. The cost of producing an answer collapses, and the value shifts to asking the right question. The cost of generating options collapses, and the value shifts to selecting between them. The cost of execution falls, and the value moves upstream into judgement. Marketing is not unusual in this regard. It is simply one of the first industries being forced to confront it in public.

So the parts that mattered were never the generators. To the ledger, the honest answer to Wanamaker is not a prettier dashboard but a closed loop: publish, track the real outcome, attribute it, feed it back, let the next decision be measurably better than the last. Marketing stops being a spend you defend and becomes a system that visibly compounds — a circulatory system rather than a bonfire. That is a thing a CMO can carry into a board meeting, because it finally answers the CFO in the CFO’s own language.

To the magic, the answer is subtler, and it is the part I got wrong first. The creative is not the output — the words, the image, the film. It is a carrier. What it carries is a decision about who this brand is, what it has permission to say, and why anyone should believe it. The magic the creative director was protecting was never the thirty-second idea. It was the judgement underneath it. Good software does not automate that away; it gives it leverage and gets out of its way, letting one person with real taste operate at the scale of a department without diluting the one thing that made them worth listening to.

That is the whole product, really — less a factory than an argument that the two masters can be served at once instead of traded off.

And this is the optimistic part, the part I want to be loud about, because it is the truest thing I learned: marketing has never feared new tools. It is, and always has been, the first room in the business to pick them up. Print, broadcast, digital, search, social, programmatic — every wave, the craft absorbed the technology and used it to move the work forward, not to shrink it. AI is simply the next wave, and the marketers I met are already reaching for it with both hands. A system like Bertha does not stand against that instinct; it stands under it — handing the commodity work to the machine so the judgement has somewhere better to go.

Here is what that argument looks like once it is built into software: an entire agency, compressed into a system, ending in the closed loop that answers the ledger.

Product discovery is temporary empathy

Here is the part I did not plan and only saw in hindsight, and it is the part I most want to say plainly.

Product discovery is usually described as gathering requirements. In reality it is an exercise in temporary empathy. You borrow another person’s mental model, build against it, then let reality show you where your simulation was wrong. You cannot build software for a job you have not inhabited, because the requirements are the symptom — and if you build to the symptom, you build the wrong thing.

So to make Bertha at all I had to do the thing that does not come naturally to a technologist: climb inside these people and feel the pressure they feel. Sit where the CMO sits and feel both heads of the fear at once. Sit where the creative director sits and feel the insult of being measured. Sit where the agency CEO sits and feel the floor moving under a business model. You do not get a working strategy layer by reading about strategy. You get it by borrowing the mind of the person terrified of being unable to prove it, until their constraints become your data model.

And here I have to correct something, because it is the easiest thing in the world to misdiagnose — and I was the one most tempted to get it wrong. It would be convenient to cast this as the technologist crossing into foreign territory: the measurer learning, humbly, to respect the magicians. That framing is flattering and it is false, and it was false from the start. A technologist — an innovator, a builder, someone who makes a thing that did not exist that morning — is not the opposite of a creative. He is a creative working in a different medium. The blank editor and the blank page are the same blank. The architecture I sketch and the campaign the creative director sketches are both attempts to drag an idea out of nothing and into a form that can survive contact with the world. Different discipline, different material, different vocabulary — creative nevertheless. The reason I could climb inside those minds is that I was never standing as far outside them as the org chart pretends. We are separated by domain, not by kind. That is the quiet thing the whole industry keeps getting wrong when it seats “the creatives” and “the technical people” at opposite ends of the table: it mistakes a difference of medium for a difference of nature.

And then the thing folded back on itself. Those same minds — the ones I had to inhabit to build it — became the feedback that corrected it. The CMO who killed the perfect campaign taught me the system had to protect the “doesn’t feel like us” veto, not override it. The founder with the screenshots taught me my audience model was too clever and not close enough to how he actually saw his customer. The creative director’s contempt taught me exactly where automation would be resented and where it would be welcomed. I would build to my model of their mind, show it to them, watch their real mind reject the part I got wrong, and correct. The people I was modelling were also the feedback correcting the model.

There is a symmetry there I noticed too late. Bertha’s whole thesis is a loop — publish, measure, feed the result back, get smarter each round. But the machine only has that loop because building it was the same loop, run on humans: inhabit the mind, ship to the mind, let the mind correct you, go again. The circulatory system inside the product is a copy of the circulatory system that made it.

The pushback, taken seriously

You cannot write honestly about AI in marketing in 2026 without the resistance, because it is everywhere — and I have learned the hard way that the resistance is not the problem to overcome. It is information about a fear I have not yet addressed.

There are three real objections under the noise, and each is at least partly correct.

The creative’s objection: you are going to automate the soul out of it. Not sentimentality — a genuine risk that a tool optimised for plausible output floods the world with marketing that is competent, on-brief, and completely dead. They are right; most AI marketing on the market does exactly that. The mistake is aiming the fear at the technology instead of the implementation. A system that amplifies the human’s judgement produces something very different from one that removes it — but that distinction has to be built in on purpose, and usually isn’t, so the suspicion is well earned.

And there is an irony sitting on top of that objection that I never fail to notice in the room. The creatives are the loudest resisters — technology is “not their wheelhouse,” it is the thing the IT department does to them — and yet they are already using it every day, and have been for years. The moodboard pulled in seconds, the reference found instantly, the rough cut, the mockup, the ten variations of a headline explored before lunch. Technology does not threaten ideation; it accelerates it — it lets a good idea be turned over, stress-tested and shown ten ways in the time it used to take to sketch one. The same person who says the machine will kill the craft has already quietly let it collapse the distance between having an idea and seeing it. Use it and fear it, exactly the knot I wrote about before.

And the age we are in tightens the irony further. No idea reaches production on taste alone any more. A campaign now has to arrive at the table already carrying data points that say it might work — a signal, a test, a proof of appetite — before anyone will fund it. So the very technology the creative resents is also the thing that earns their idea its hearing: the machine that supposedly threatens the magic is the same machine that gets the magic past the ledger and into the world. Resist it entirely and you do not protect the idea. You leave it undefended in the one meeting where it has to be proven before it is allowed to exist.

The agency’s objection, the one nobody says aloud: this disintermediates me. If a lean team with the right operating system can run the full lifecycle of an agency, a great deal of what agencies charge for is exposed. That is a real threat to a real business model, and pretending otherwise insults the people feeling it. What survives the disruption is what was always the actual product — the judgement, the client trust, the nerve to recommend the uncomfortable thing. The commodity layer erodes; the taste layer becomes more valuable, because it is now the scarce part. Cold comfort to anyone whose living was in the commodity layer, and I won’t pretend otherwise.

The brand guardian’s objection, the deepest: AI erodes the one thing we cannot afford to lose — trust. When the feed fills with synthetic, competent, infinitely-generated content, the scarce asset stops being attention and becomes trust — and a brand that floods its own channels with machine-made marketing can burn, in a quarter, credibility it took twenty years to earn. I believe this so strongly I wrote a whole essay about it. The guardian who resists AI on trust grounds is not a luddite. They are the person in the room correctly pricing the risk everyone chasing efficiency is ignoring.

So no — the pushback is not wrong. It is the most useful signal in the category. Every objection is a specification: amplify the judgement, don’t replace it; make the value provable so the disintermediation lands on the commodity and not the craft; treat trust as the thing to protect, not the thing to spend. Build against the fear and it stops being resistance. Ignore it and you ship exactly the soulless, disintermediating, trust-burning machine the industry is right to refuse.

Everything, in a moment, for nothing

There is a fourth pressure, and it does not come from inside the craft. It comes from the client, the boss, the board — and it is quietly the most corrosive of all.

AI has reset what people believe things should cost and how long they should take. Because the one visible, cheap thing — generating the content — is now instant and very nearly free, everyone downstream assumes the entire chain is too. If the machine can do it in a moment for nothing, why does this take two weeks, why does it cost what it costs, and — the unspoken part — why am I still paying you at all? Every CMO and every agency I know is now managing that sentence, in some form, from someone who controls their budget.

And here is why it perpetuates the problem rather than solving anything. The part that collapsed in cost was never the part that was the work. Generating a competent ad was always the cheap step; the expensive, slow, scarce layer is the judgement, the strategy, the proof that it will land, the governance that keeps it on-brand and trustworthy. But that layer is invisible to the person paying, and the cheap layer is sitting right on top of it in plain view — so the commodity’s price now sets the expectation for the craft’s. Clients price the judgement by the cost of the content.

The cheapest part of the work has quietly set the price for all of it. And the cheapest part was never the part that mattered.

The cruelty is that this squeezes fees and timelines precisely on the work AI did not make cheap — starving the very judgement that stops AI producing slop in the first place. Expect everything in a moment for nothing, and you defund the only thing standing between your brand and the dead, competent output the creatives were right to fear. The CMO feels it as a fresh turn of the ledger screw: the CFO now expects marketing to be instant and free too, “because it’s all just AI now,” which makes the impossible half of the job harder still. The expectation collapse does not lower the cost of good marketing. It just makes good marketing harder to fund, harder to defend, and easier to replace with something that only looks the same.

But this is exactly where the craft has always earned its keep, and it is where I am most optimistic — because the same technology that manufactures the “something for nothing” expectation is also the sharpest answer to it. Let the machine harvest the low-value, high-volume work that was never worth a human’s afternoon — the routine variations, the always-on filler, the thousandth resize — and the marketer is handed back the hours to do the thing that actually compounds a brand. Find the small, high-value cohorts that genuinely matter and reach each of them with something meaningful, precisely, at a cost and a granularity that used to be impossible. A message that once had to be broad enough for everyone can now be exactly right for the few hundred people it was truly for. That is not the industry being hollowed out; that is the industry raising its own game — trading busywork for relevance, volume for precision, spray-and-pray for something a customer can actually feel was made for them. The pressure is real. But marketing has met every pressure like it before, absorbed the tool that caused it, and come out sharper. It will meet this one the same way: by getting better, not smaller.

The next seat to break

I argued not long ago that the CTO was the first executive role to break — that one title had quietly fractured into five jobs, then six, not because CTOs were weak but because technology had escaped the technology department and become the substrate of the whole business. The crack showed there first. It was never going to stop there.

The CMO’s is the seat where it spreads next, by exactly the same mechanism. Marketing has escaped the marketing department. It is now a software and data discipline wearing a brand’s clothes: the seat holds the brand storyteller and the growth engineer and the owner of the customer-data-and-model stack that quietly decides who sees what, when and why — and now the AI-governance seat on top of all of it. The creative chair and the engineering chair are merging into a single job, and most marketing org charts have not noticed yet. The agency CEO feels the identical tremor from the other side: the thing they sold — the work — is commoditising underneath them, and what survives is the seat that sells judgement, governance and translation instead of deliverables.

Which is the same reframe the CTO role was forced through, arriving now at the CMO’s door. The CTO stopped being the person who understood the technology and became the person who helps the organisation understand what technology makes possible — architect of outcomes, not systems; translator, not specialist. The CMO is being asked to make the mirror-image move: stop being the person who owns the brand and become the person who helps the organisation understand what the brand — now instrumented, now measurable — makes possible. Not the keeper of the magic. The translator between the magic and the machine.

The CTO was the first seat to break. The CMO is the seat where the two halves of the job — the magic and the ledger — were already the furthest apart. Which is exactly why, when it fractures, it will hurt the most.

And notice what the fracture actually is, underneath the org-chart language. It is the contradiction I opened with, finally made visible in the structure of the role itself. For years the magic seat and the ledger seat were loosely coupled — a brand team over here, an analytics function over there, a comfortable distance between them. AI collapsed the distance. Now both are pressed into one title, one person, one meeting, in real time — and whoever holds it is asked to be the translator between two languages the last twenty years let stay politely separate. That is not a role getting harder. It is a role becoming a different role while keeping the same name.

The counter-case, honestly

It would be too tidy to leave it there, with the enlightened builder having answered every objection. The steelman is stronger than my thesis, and my thesis needs it.

The uncomfortable possibility is that AI will expose how much of marketing was never magic in the first place. Every industry contains genuine expertise. Every industry also contains stories people tell themselves about why their expertise cannot be measured. Not every “it doesn’t feel like us” is hard-won taste; some of it is a highly-paid person resisting measurement because measurement is threatening. The technologist’s original arrogance was not entirely wrong — there is real vagueness here used as cover, and AI genuinely will, and should, strip a lot of it out. Lean too hard on “protect the judgement” and you hand every unaccountable marketer a permanent excuse. Some of the magic deserves to be automated away, and pretending all of it is sacred is just the mirror image of the technologist who thought none of it was.

Harder still: everything I have said about amplifying judgement assumes there is judgement worth amplifying. Most organisations do not have a creative director whose idea moves a market. They have a stretched team producing adequate work under deadline — and for them, a machine that produces slightly-better-than-adequate work at a fraction of the cost may simply be the right answer. The premium, protected-judgement model I have described is real, but it may be a smaller market than I want it to be. A fair reader should hold both open, because I honestly do not yet know which is bigger.

It was never really about marketing

The fracturing seat is the visible symptom. It is not the deepest layer.

Step back further and the CMO stops looking like a special case at all. Because the contradiction underneath the fracture is not a marketing problem, and not even a C-suite problem. It is the defining problem of senior judgement in the age of measurement: modern leadership increasingly consists of proving the value of things before they become measurable. The investor asked to justify a thesis before the returns exist. The doctor weighing a clinical read against what the protocol will reimburse. The hiring manager who can feel the wrong candidate but cannot put it in the scorecard. The architect, the intelligence analyst, the product leader — every one of them handed the same impossible instruction the CMO gets: prove something whose value appears before it becomes measurable. The CMO is merely the most visible person standing in it, because marketing is where the contradiction is loudest and the tenure is shortest. But the shape is universal, and AI presses on all of it at once, because AI is, at bottom, a measurement machine arriving in every room where judgement used to be trusted precisely because it could not be measured.

And here is where the two worlds I spent three years between finally collapsed into one.

Technologists see the world as a measurement problem. Marketers see it as a perception problem. The longer I spent between them, the more I realised they were describing the same thing from opposite sides.

Trust.

The ledger is simply trust rendered as numbers. The brand is trust rendered as feeling. The CMO sits between them not because the job was badly designed, but because the job is where a company’s trust gets converted from one form into the other. The technology that wins will not choose the ledger over the magic, or the magic over the ledger. It will recognise that they were always the same asset, viewed through different lenses.

The magic was always real. So was the ledger.

The challenge was never choosing between them.

It was understanding they were measuring the same thing all along.

Epilogue: The Machine and the Magic

I walked through that door a technologist expecting a factory, and I enjoyed everything on the other side of it more than the tidy version of this essay would admit. It left me with a deeper respect than I arrived with — for the industry, for the people who hold its impossible seat, and for the pressure they carry into the uncharted territory of tomorrow. Will tomorrow be better? I have no hesitation: yes. But what it looks like will be decided the way it always has been — not by the ledger, not by the machine, but by the creative magic.

And I want to name what these four years actually were, because “building a product” is far too grey a phrase for what they gave me. It was a dance — with the most magnetic partner in the ballroom. Marketing is the boardroom’s darling: the glamorous one everyone watches and almost no one truly understands, courted for her reach and mistrusted for her mystery. For four years I got to learn her steps — where she leads, where she lets you lead, the timing that cannot be taught from a page. She showed me that taste is a discipline and not an indulgence, that a room can be read as precisely as a dashboard, and that the line between persuasion and manipulation is a single, deliberate step. I arrived to build her a machine. I am leaving having been taught to dance.

I won’t pretend the thing I built was perfect. It wasn’t. It was, frankly, too much — a whole agency compressed into one operating system is intimidating to sell and more intimidating to buy, and the completeness that makes it powerful is the same thing that makes a client take a step back. It was expensive to bootstrap in exactly one place, the video rendering and the tuning around it, where the compute bill bites hardest — but that is a partner problem, not a flaw in the design. So today it sits much as I described it in The Giant Under the Snow: built, tested, ready to deploy, sleeping while it ponders which road to venture down. Perhaps it wakes for someone else. Perhaps I simply point it at my own shingle and run the agency of the future myself — one operator, one machine, the whole lifecycle from CRM to campaign to measured outcome.

But here is the honest shape of it, and it is the same shape as everything above. Everything I built answers the ledger. It measures, attributes, proves and governs — and it was only ever half the equation. The other half, the magic, was never mine to supply alone. I can build the machine that makes the magic provable, scalable and safe. I cannot be the magic. That takes a different kind of creative from me — the taste, the nerve, the read on the culture I spent this whole essay learning to respect.

The machine is built. The question I keep returning to is simpler.

Where is the creative leader willing to combine their magic with it?

Sources

• Spencer Stuart — CMO Tenure Study 2025 — average CMO tenure of about 4.3 years, trailing the roughly 4.9-year C-suite average, and recovering from a pandemic-era low near 3.5 years. spencerstuart.com
• Comparative C-suite tenure — CEO ~7.2 years and CFO ~5.7 years against the CMO’s ~4.2, compiled from Korn Ferry and Russell Reynolds data. Behind the CMO
• McKinsey — The State of AI (2025) — regular generative-AI use reaching roughly two-thirds of organisations, with marketing and sales among the largest value pools for deployment. mckinsey.com
• John Wanamaker — “Half the money I spend on advertising is wasted; the trouble is I don’t know which half” — widely attributed to Wanamaker (and, in the UK, to Lord Leverhulme).