From ad spend to AI spend: lessons from marketing at scale.

I spent years doing big-data marketing science at Fashion Nova, inside paid media at planet scale. Ads ran around the world, around the clock, and every dollar left a trail. That seat teaches you one lesson fast. Budgets do not fail because people are careless. They fail because money flows to what is easy to buy, not to what is proven to work.

Buying is easy. A contract, a subscription, a pilot. One signature and you feel motion. Proof is hard. Proof means wiring the work to a number and waiting for the number to move. So budgets drift toward easy. Every quarter. Every company. Every size.

I watched that law run at ad scale. Now I sit in AI budget reviews and watch the same law wearing a new badge. Seats, tools, and pilots, all bought on promise, none wired to proof. The fix comes straight from marketing science, and it transfers further than you’d think.

Marketing science, in plain language

The name sounds complicated. The job is three moves.

Attribute every dollar. That means tying each dollar of spend to the sales it caused. Kill what does not convert. Convert just means turn a viewer into a buyer. Double what does. That third move matters most, because cutting alone never grew anything. Reach lives in the winners. Protect it there.

Emotion out. Arithmetic in. We never asked which ad we loved. We asked which ad paid for itself, and the answer insulted our taste constantly. The ugly ad won all the time. Nobody argued, because the number outranked everyone in the room.

Attribution at that scale was real work. Tracking tags on every link. Holdout groups, which means people we deliberately showed no ads, so we could measure what the ads truly added. The rule underneath was plain. If we cannot trace it, we do not trust it.

At Fashion Nova scale there was no other option. A wrong guess repeated millions of times is not a small mistake. It is a crater. My Google years taught the same lesson from the reliability side. Systems at scale punish opinion and reward measurement.

People assume this discipline needs big data and a science team. It does not. It needs the will to let arithmetic outrank feelings. A ten-person company can run the entire system on one spreadsheet. The math does not care how big you are.

The same waste, new labels

Put an ad budget and an AI budget side by side. The leaks match line for line. Both budgets buy promises about attention and output. Both leak in the dark. I keep a list, and the pairs are almost embarrassing.

• Impressions you cannot tie to sales. An impression is one showing of an ad. Buying showings with no path to revenue is the oldest leak in marketing. The AI twin: seats nobody opens. A seat is one user license. Open the admin panel and count last month’s logins. That count is the truth about your rollout. If logins are low, you did not buy a capability. You bought shelf decoration.
• Creative refreshed on a feeling. Ad teams swap images because the old ones feel stale. AI teams rewrite prompts because the output feels off. A prompt is just the written instruction you give the model, and it deserves the same rigor as an ad test. Version it, measure it, keep the winner. Nobody checks a number first when feelings run the process. Hours burn, and everyone calls it progress.
• Agency retainers without performance clauses. A retainer is a flat monthly fee. Without a performance clause, the agency gets paid whether sales move or not. The AI twin: subscriptions without an owner. If no one’s name sits next to the tool, no one answers when usage hits zero. Ownership is a name, a number, and a review date.
• Channel sprawl. Marketing teams keep adding platforms because adding feels like growth. AI teams do the same with tools. Five overlapping tools do the work of two and bill like five. Count how many of your tools write text. More than two means you have sprawl.

That line should sting. Ad platforms push charts at you whether you ask or not. Most AI line items sit on a card statement with no chart at all. Same waste, less visibility. And the fix is not a smarter purchase. The fix is a system. Run the audit before the renewal dates arrive, not after.

The transfer playbook

Run the three moves on your AI budget, in order. This is the exact sequence I use inside companies.

First, attribute. Every AI dollar gets a job and a number. Not “productivity.” A job reads like this: draft the first reply to every support ticket. A number reads like this: minutes saved per ticket, counted weekly. If a tool cannot name its job, it is not a tool. It is decoration. I published the metric set I use for this in my guide to measuring AI ROI. Five numbers. One page. Give every dollar thirty days to show its number. No number, and the dollar moves.

Second, cut. Kill the unused seats and the duplicate tools, fast. Not at renewal. Now. In paid media we shut off losing campaigns every week, because every extra week was pure loss. An unopened seat is a losing campaign. Same math. Same move. Nobody mourned a dead campaign, and nobody should mourn a dead subscription. Finding duplicates is simple. Group your tools by the job they claim. Two tools, one job, one survives.

Third, reallocate. Cutting is not the goal. Feeding winners is. Take everything you just recovered and move it into the one workflow that already shows payback. One workflow with proof beats ten pilots with applause. Proven payback means a number you already measured, not a number from a vendor’s slide.

The order matters. Cut before you attribute and you are killing blind. Reallocate before you cut and you fund the new thing with money you do not have. Attribute, cut, reallocate. Keep the order.

Here is why this cuts cost without cutting reach. The dead spend was never producing the reach. It was producing comfort. You are not shrinking the machine. You are removing parts that were never connected to it.

Why I built Claresto

Ad operations taught me one more lesson, and this one cost the most. I lived the problem long before I solved it. The numbers usually exist. Nobody can see them in one place. Spend lives in one dashboard, results in a second, creative tests in a third. The waste hides in the gaps between screens.

So I built Claresto, the Ad Operations Command Center for paid media. It puts every campaign’s numbers in one place, spend next to results, account by account, day by day. When all the numbers share one room, bad spend has nowhere to hide. I did not build it because dashboards are fun. I built it because I was tired of watching sharp teams lose money to blind spots they could not even see.

AI spend needs that same command center discipline. Command center is not a fancy phrase here. It means the place where every decision can see every number. One page. Every tool on it. Each row holds the tool, its job, its number, and its owner. At iExcel, my production agent systems live under that rule. An agent shows its number or it gets turned off. Review the page monthly and sprawl never grows back. You do not need software for this on day one. A spreadsheet works.

Start where the spend is zero

Ads never offered this option. There is no free tier on a billboard. AI has one, and it is good. The cost of testing an idea has never been this low, and most budgets still behave like every test needs a purchase order.

Today’s free tiers can run a real workflow end to end. Writing, images, research, code. I wrote the $0 generative brand stack playbook to prove it, and it is the order of operations I follow with every team I advise.

Free first is not about being cheap. It is attribution made easy. When spend is zero, any gain is pure win, and you learn which workflows actually matter before the first invoice lands. Then a paid tool has to beat a working free baseline to earn its line item. That is a high bar. It should be.

The free stack also trains the habit at no cost. Attributing, cutting, reallocating, all practiced where mistakes cost nothing. By the time real money enters, the discipline already runs. Free tools also fail cheap. A failed test that cost nothing is research. A failed rollout that cost a year of seats is a write-off.

Earn the paid tools with results. That sentence sounds backwards in a world where buying feels like progress. It stops sounding backwards the first time you watch a company pay a full year for software nobody opened.

Spend follows proof

Everything marketing at scale taught me compresses into three words. Spend follows proof. Buy nothing on a promise. Fund whatever shows its number. Check the loop weekly, because budgets drift the moment nobody looks.

Start tonight if you want. List every AI charge on the company card. Next to each one, write the job, the number, and the owner. The blank cells are your answer. Cut them without ceremony, and move the money to the row that earned it. The first pass takes an evening, not a quarter.

This is the work I do inside companies now, the same arithmetic pointed at newer tools. I laid out the operating model in my essay on the fractional Chief of AI. Embed, ship, hand off.

If you run a budget and suspect it leaks, you are right. Bring your line items and I will bring the playbook, builder to builder. I’m the Fractional Chief of AI for businesses that know they’re behind. I take you from watching AI happen to running on it in 90 days. Book a strategy call.