Count the Hours That Moved, Not the Hours You Saved. Change Futurist | September 19, 2026
Name the step your AI investment made faster. Now name the step that absorbed the work it displaced. Most marketing leaders answer the first question in two seconds and the second one never.
Two days ago this column argued that authority is the constraint on AI programs, and that teams need to write down who can approve an agent's output and ship it. That argument assumes something worth examining. It assumes you know where the work sits. Research published inside the last 48 hours says you probably do not, because AI moves work between steps faster than anyone updates the map.
Jana Retkowsky, Ella Hafermalz, Marleen Huysman, and Daan Odijk spent a year following senior creatives on television and streaming campaigns and found the displaced work landing on production teams who never saw it coming. Paula Goldman, Salesforce's chief ethical and humane use officer, argues on HBR IdeaCast that the useful question has become where to withhold AI rather than where to apply it. Josh Stephenson reports on a B2B marketing function rebuilt around the business instead of around channels. Matthew Valentine asks B2B CEOs what actually gets a budget signed.
Four vantage points, one variable. Where the work went.
I have watched this play out in enough content operations reviews to recognize the shape. Draft time collapses, and everyone celebrates. Six weeks later, however, the production calendar has slipped, legal is three days behind, and nobody connects the two events because they live in different dashboards owned by different people.
Count the hours that moved, not the hours you saved
Retkowsky of Rotterdam School of Management, Hafermalz and Huysman of the KIN Center for Digital Innovation at Vrije Universiteit Amsterdam, and Odijk of RTL Nederland published Research: Gen AI Is Collapsing Creative Processes on September 16. Their opening case is exact enough to use in a meeting.
A photographer reviews a concept image the client has already approved. A man wears a baseball cap that shadows his face, and his eyes are lit anyway. No camera setup on earth produces that lighting. A senior creative generated the image, the client loved it, and the production team now has to build something reality cannot match.
The mechanism the authors describe runs through polish. A single person equipped with a generative tool produces output finished enough that clients fix on details that cost a fortune to reproduce or cannot be reproduced at all. Approval happens against an artifact that carries no signal about its own feasibility. Rework follows, and it lands on people who were not in the room.
Their remedies are structural. Bring downstream specialists into early development. State plainly what a generated concept represents and what it does not. Preserve options instead of converging early. Set client checkpoints before the polish does the persuading.
Read that list again and notice what it costs. Every item adds time upstream. That is the trade, and teams reporting a 60% drop in draft time while their content supply chain slows down have already made the trade without pricing it.
It reminds me of what Jay Combs, VP of Marketing at ModelOp, said when I interviewed him on The Agile Brand podcast: "When governance is bolted on at the end and fragmented across teams, it slows everything down. The work finishes, and only then do missing documents, new approvals, and open compliance questions surface."
My take: Retkowsky and her coauthors document a cost transfer, and marketing leaders can act on it immediately by measuring cycle time across the whole chain and not only at the step where the tool sits.
Decide where AI does not go before you decide where it does
Goldman appeared on HBR IdeaCast on September 17, in the first of a four-part series with Adi Ignatius. She traces the phrase human in the loop back to Cold War missile detection, where the question was who makes the consequential call. Her argument is that marketers have flattened it into "AI drafts, people approve," which fails once agents reason through many tasks at once. She proposes human at the helm: put human judgment at the points where it changes the outcome.
She is direct about the limits of the standard heuristic. AI handles routine work and people handle complex work "gets you about 70% of the way there," in her words, and the remaining 30% turns on customer emotion, employee preference, and sensitivity.
Her sharpest example runs the other direction from most AI case studies. IKEA's innovation team wanted a couch that broke the boxy, cushioned stereotype. Every generative attempt reverted to the mean. So the team front loaded the brief, brainstorming around campfires and gathering spaces until the breakthrough ideas were explicit, and only then gave the model direction. The result was a ten-pound couch in a box that ended up in a Copenhagen museum exhibit.
That is the same remedy Retkowsky's team prescribes, arriving from a different discipline. Do the expensive thinking before the cheap generation, because cheap generation anchors everyone who sees it.
Velocity without coherence produces motion you pay for twice. It also maps to the principle I keep returning to: humans stay accountable for direction even when the analysis runs itself. Deciding what a campaign should be is a normative act. Deciding what a model can produce is not.
My take: Goldman gives CMOs a reusable decision frame, since naming the three or four steps where you will deliberately withhold AI forces the conversation about judgment that most AI roadmaps skip entirely.
Put the constraint in front of the people who sign the budget
Stephenson reported on September 16 that Khalid Aziz rebuilt Canon's B2B marketing function into what Marketing Week describes as the command centre of the business. Valentine followed on September 17 with interviews asking B2B CEOs what earns a budget sign-off and what changes when marketing gets full flexibility.
Both pieces circle the same requirement. A marketing function that owns an outcome can move work between steps. A function that owns a channel cannot, because the step that absorbed the displaced work sits in somebody else's budget line.
That is why the measurement problem and the operating model problem are one problem. If your agency holds production, your legal team holds review, and your brand team holds approval, then the cost transfer Retkowsky documents is invisible by construction. Three ledgers, no consolidated view, and a CFO who sees a drafting tool that paid for itself.
It reminds me of what Tambi Younes, VP of E-commerce at Newell Brands, said when I interviewed him on The Agile Brand podcast: "Freeing up the teams to be able to do that due diligence and help inform the teams and refine our process further upstream. It's kind of our next step, but you can't do that if you're bogged down with just that compliance effort in the day-to-day."
Younes describes automation earning its keep by funding upstream judgment. That framing survives a CFO conversation. Faster drafts do not.
My take: Stephenson and Valentine document the authority that makes end-to-end measurement possible, and marketing leaders who cannot see cost across production, review, and approval should make that visibility the ask in their next budget cycle.
Run the numbers on a chain you recognize
A consumer health brand runs product content with eleven people across copy, design, regulatory review, and retail syndication. Last year the team licensed a generative suite for concepting and first drafts. Concept-to-first-draft time fell from eleven days to two. The team reported a 5.5x improvement and the CMO put it in a board deck.
Here is what the board deck omitted. Regulatory review queue time rose from four days to nine, because the volume of concepts reaching review tripled and each one now arrived with claims language a model had inferred rather than a writer had sourced. Retail syndication rework rose from 6% of assets to 19%, because generated product imagery carried lighting and packaging details the photography vendor could not match on set. Total concept-to-live time moved from 34 days to 31.
Three days. On a 5.5x drafting improvement.
Now apply both remedies. Front load the brief: regulatory and the photography vendor join concepting for one 90-minute session per campaign, which costs roughly 14 hours a quarter across both teams. Label every generated concept with what it represents, so review reads it as direction instead of specification. Withhold generation entirely from claims language, where the lightest mechanism that clears the accuracy bar is a writer working from the approved claims library.
Regulatory queue falls back toward five days because volume drops and inputs arrive sourced. Syndication rework falls toward 8% because the vendor shaped the concept. Concept-to-live lands near 19 days.
Same eleven people. Same tool. The difference is that somebody counted the whole chain and then spent time upstream on purpose.
Do these four things before Q4 planning is complete
Measure one workflow end to end this month. Pick your highest-volume content chain and instrument every step from brief to live, including queue time in review functions you do not own. Expect the number to embarrass the tooling narrative.
Name the three steps where you will not use AI. Claims language, pricing, and net-new positioning are the usual answers. Write them down and tell the team why, because an unstated exception gets automated by the first person who does not know about it.
Move one specialist upstream. Whoever absorbs the most rework joins concepting. Fourteen hours a quarter buys back weeks of production time, and the math is easy enough to show a CFO.
Take the consolidated chain to your budget conversation. A 5.5x drafting gain that produces three days of end-to-end improvement is a story about where the work went. A CEO who sees that diagram will fund the redesign.
I have written more than a few books, and drafting has never been the expensive part. The time goes to fact-checking, permissions, editing passes, and the index nobody thinks about until the end. When drafting got faster, none of that got faster. The manuscript just arrived at the slow part sooner.
Name the step that got faster. Then go find the one that absorbed the difference, and measure that one for a quarter before you buy anything else.
Featured insights:
Research: Gen AI Is Collapsing Creative Processes. Jana Retkowsky, Ella Hafermalz, Marleen Huysman, and Daan Odijk, Harvard Business Review, September 16, 2026
How AI Is Changing Talent, Not Just Tasks: Rethinking Where Human Judgment Matters Most. Paula Goldman with Adi Ignatius, HBR IdeaCast, September 17, 2026
'Bring your helmet': How Canon rebuilt its B2B marketing function. Josh Stephenson, Marketing Week, September 16, 2026
What do B2B CEOs really think about brand building?. Matthew Valentine, Marketing Week, September 17, 2026