I love a detailed process map more than is probably healthy. This one makes me nervous. If they complete it exactly as drawn, they may have vibe-coded an entire martech stack: the same campaigns moving through a new collection of tools, after a lot of hours spent rebuilding work the team already does. And still no answer to the only question that matters: did pipeline, margin, or customer value improve?

One person on the team put it more plainly: “I’m using AI for everything and I still haven't closed any gap on my pipe gen goal.” That is The “Do the AI” Trap. AI activity becomes the goal, the roadmap gets very impressive, and business value is left to chance.

The fix is not to stop using AI (obviously). It is to stop starting with AI. Start with the number the business needs to move, work backward to one valuable loop, and give one operator enough authority to change it, measure it, and keep improving it. That is the 30-day process I call The AI Value Reset. In the rest of this issue, I’ll show you how to run it next week.

The expensive part is not the Figma board. It is what happens when the team completes it and the economics stay the same.

For an agency, that means rebuilding the delivery model without improving client results, renewal odds, or scope margin. The work looks different. The business does not.

The pattern is bigger than one ambitious Figma board.

Deloitte's 2025 survey of 1,854 executives across Europe and the Middle East found that 85 percent of organizations had increased AI investment and 91 percent planned another increase. One executive described the mood as: “Everyone is asking their organization to adopt AI, even if they don’t know what the output is”.

The AI Value Reset is designed to close that gap between investment and value. Over 30 days, the team chooses the business result, cuts the roadmap down to one valuable loop, and gives an accountable operator enough authority to change how the work gets done.

When working with new teams, I map delays, repetitive work, painful handoffs, and decisions made with inadequate information. The first pass usually produces 50 to 100 ideas for applying AI (no team has ever suffered from an idea shortage). A big process can easily generate a 250-step build plan.

The length of the plan feels like rigor. It often hides the absence of a business thesis.

A team can automate a monthly report and draft every campaign asset faster. Those gains lower production costs, create capacity, and help the company outproduce slower competitors. I use AI for those reasons every day. The ROI gap appears when the underlying campaign strategy, approval chain, offer, and measurement system stay the same. Faster production sends more work through the same machine. Rebuilding the same car with different tools may help you sell a few more at the same price.

You can recognize the trap in the way the team reports progress. Licenses assigned and weekly users counted should lead to a measured change in revenue, margin, cost to serve, or customer value. Hours saved on individual tasks should show up as capacity redirected to a named business priority. More content and campaigns produced should improve campaign performance or attributable pipeline.

The same test applies farther down the stack. A pilot launched successfully only matters once the workflow is adopted, measured, maintained, and improved. A polished AI-generated deliverable earns its keep when it helps someone make a faster or better decision with appropriate human judgment. If tool adoption and completed tasks dominate the quarterly update, the team has an adoption program without a value-creation system. Other warning signs include a roadmap organized by tool, disconnected pilots, no baseline, and a power user who has become the unofficial help desk.

That power user can look 10 times more productive while the P&L barely notices. Their analysis and decks are better, so everyone asks how they did it. They hear “Claude” or “ChatGPT,” try the same tool, and get... a longer report... devoid of reasons why the numbers are, and what will be done to improve them. Owning Excel did not turn any of us into investment bankers (deeply unfair, given the price of Microsoft 365).

McKinsey describes the same bolt-on pattern in its July 2026 operating-model analysis: activities get faster while approval layers, roles, and constraints remain. In a July McKinsey Global Institute conversation, Tanguy Catlin points to proprietary data, human judgment, customer embedding, and a faster “metabolic rate of learning” as stronger sources of advantage.

The fastest way I know to cut a 100-row roadmap down to size is to ask what the business has to accomplish this year. Then every possible AI investment has to earn its place against that answer.

Start with the number the business needs to move

In one engagement, the answer was painfully specific: the marketing team had no additional media budget, no additional headcount, a 30 percent pipeline gap, and six months left to close it. Once we put that constraint at the top of the roadmap, several popular ideas dropped immediately. Automating a monthly report was useful and unlikely to close the gap. Campaign-impact and revenue-creation opportunities moved up.

Pull the outcome from the operating plan, P&L, or customer scorecard. Record the baseline, target, owner, constraint, and one leading indicator the team can observe within 30 days. Efficiency belongs here when cost or capacity is the binding constraint. Growth, retention, decision quality, customer experience, and speed of learning may create more value in a different business.

This step also forces a decision about where saved time will go. Time saved becomes business value only when someone redirects that capacity toward a named priority.

Score the roadmap before the loudest idea becomes the priority

I rank each idea on outcome impact, frequency and reach, reuse potential, strategic proof, learning value, build simplicity, integration stability, existing-platform leverage, human-judgment fit, and data readiness. That score gives the team a shared way to compare a high-frequency campaign loop with a fragile automation that saves two hours every quarter.

A first build should have a clear owner, use stable data, ship a useful version within one or two weeks, and produce a measurable before-and-after result. Configure an existing platform when it can perform most of the job. Use custom code for work that is differentiated, repeatable, and valuable enough to maintain.

Turn the winning use case into a learning loop

An AI system that writes email copy produces an input. A useful growth system uses customer and performance context to write the copy, gets it into market, measures what happened, learns which variants worked, and changes what it produces next. It makes more of the winners and less of the losers. The marketer reviews the judgment calls and feeds that learning back into the system.

A real loop has five parts: an input, AI-enabled creation or decision support, distribution into live work, a returned signal, and reinvestment of that signal into the next cycle. The returned signal matters most. Without it, the team built a faster one-shot.

Over time, the loop accumulates customer history, performance data, offer knowledge, brand judgment, and review decisions (assuming someone actually captures them). That operating context is harder to copy than a prompt.

McKinsey's July analysis of its 2026 survey reinforces the redesign point. Leaders at organizations in the earliest enablement stage were 5.3 times more likely to report enterprise value when workflows had been redesigned, 32 percent compared with 6 percent when workflows stayed unchanged. The findings are self-reported and associative, so I read them as directional evidence.

Give an operator authority and put senior judgment in the loop

Most marketing jobs are still defined by inputs such as running the email program, writing the blog, or building the campaign. Very few people can describe their job as “increase attributable pipeline from email by 20 percent this year.” Changing a program may require eight approvals and three months, until the live work barely resembles the original idea. A roadmap workshop cannot overcome that operating design.

The power user invested nights and weekends learning how to structure context, build repositories, test tools, and review output. Leadership sees the result and quietly assumes they will teach everyone else. Their title, compensation, authority, calendar, and full-time workload remain unchanged. They have no mandate to redesign a peer's workflow or require a team to use what they built.

Scaling the work requires an operator with protected time, organizational authority, a deployment playbook, and responsibility for adoption and results. That person also needs enough senior judgment around them to improve the model's first answer.

Broad, senior marketing generalists can direct several functions across a demand gen motion and judge the output against a commercial result. A practitioner with 15 or 20 years of experience can spot the generic assumptions in a base-Claude response and make it 20 percent better. I use that 20 percent as a rough operator's heuristic. Publishing every plausible first response creates a whole-company slop cannon at extraordinary speed.

My working hypothesis for some production-heavy organizations is that 40 to 50 percent of junior production capacity may eventually move into AI systems, with part of the savings reinvested in fewer, more expensive senior generalists. I offer the range as a scenario; using it as a restructuring target would be irresponsible. Attrition, redeployment, work complexity, and team quality change the answer.

McKinsey's readiness research helps explain why licenses cannot create this shift. Seventy percent of respondents felt personally ready to use AI, while 27 percent of leaders believed their organizations were ready for the required people and operating changes. Employees were moving faster than the institution around them.

Measure the live result and decide what to scale, change, or stop

Run the loop in live work for 30 days. Compare the leading indicator and available outcomes with the baseline. Review quality, cycle time, adoption, and capacity created. Save the context, decisions, review rules, and open problems. End the review with one of three decisions: scale it, change it, or stop it.

The scorecard can stay almost embarrassingly small: baseline, intervention, target, owner, review date, decision. IBM reported in February that 90 percent of 1,510 technology leaders in an Apptio survey struggled to measure ROI. A CFO.com guest analysis makes the practical problem clear: pilots without a financial owner, baseline, and measurable economic delta leave success impossible to determine.

I pulled together the tools I use at each stage so a team can run this process without inventing its own planning system first. The full guide and downloadable toolkit are available to everyone.

AI Workflow Prioritization Rubric. Use the downloadable spreadsheet below when the workshop produces more ideas than the team can fund. It gives you a 10-factor score, build-path gates, score bands, and a live roadmap formula.

Get the Rubric

Build vs. Buy Decision Prompt. When the plan starts drifting toward a custom martech rebuild, use the published decision framework and prompt to compare configuration, purchase, custom build, and hybrid paths.

Get the Prompt

The AI Value Reset itself is intentionally demanding. It will not accept “use AI more” as an outcome, and it forces low-confidence assumptions into the open before they become expensive builds.Run the reset on one live marketing workflow next week

Use the five sessions below as a planning outline for one test. Allow more time if the baseline, data access, or approvals are still unresolved.

Monday: name the outcome. Pull one business result from the operating plan. Record the baseline, target, constraint, owner, and 30-day leading indicator.

Tuesday: map the current work. Capture the inputs, decisions, delays, handoffs, approvals, systems, failure points, and the weird workarounds everyone stopped mentioning.

Wednesday: score the opportunities. List the possible AI interventions, classify the build path, and run them through the prioritization rubric. Pick one.

Thursday: design the loop. Define the live output or decision, the returned signal, the human review gate, the operating context the system needs, and the person accountable for improving it.

Friday: approve the 30-day test. Set the launch date, review date, quality threshold, scorecard, and the conditions for scaling, changing, or stopping the work.

One valuable loop, one accountable operator, and a decision system that turns evidence into the next investment will teach the company more than 250 completed steps.

Start there, then make the roadmap earn its way back in.

I’m Hanna Huffman. Marketer in the Loop is my weekly letter about doing good marketing work as AI changes how we do it. Expect useful ideas, people worth knowing, and things I’m figuring out in my own work.

Published by MultiplAI Growth Partners.