The AI Experiment Is Over. Now Marketing Has to Prove the ROI.
- John Brown
- Aug 6
- 5 min read
For the last few years, simply saying you were investing in AI was enough to signal that your company was thinking about the future. That was true in the markets, and it was true inside marketing organizations.
That phase is ending.
Wall Street is beginning to ask a harder question about AI spending: where is the economic return? Marketing leaders should expect exactly the same question from their CEOs, CFOs, and boards.
Wall Street's AI honeymoon is getting more demanding
The shift is showing up in the market right now. On August 6, Reuters reported that Sandisk and Western Digital both fell sharply even after issuing revenue forecasts that beat analyst estimates. The problem was not that the companies were performing badly. The problem was that expectations around AI-driven demand had become so high that strong results were no longer enough.
At the same time, the scale of AI investment is becoming harder to ignore. Reuters reported that Alphabet was seeking up to $25 billion through a new U.S. bond sale as Big Tech increasingly turns to debt markets to help fund enormous AI infrastructure commitments. The same report noted that Amazon, Alphabet, Meta, and Oracle had collectively issued roughly $194 billion in bonds so far in 2026, while projected Big Tech AI-related spending this year is above $730 billion.
The market is not suddenly deciding AI has no value. It is doing something much more normal. It is moving from excitement about capability to accountability for return.
Marketing is about to go through the same transition
The first phase of AI in marketing was about adoption. Teams rushed to deploy writing assistants, predictive models, personalization engines, automated creative, customer-service agents, journey orchestration tools, and increasingly autonomous campaign systems.
A lot of the early reporting reflected that mindset. How many campaigns did AI help create? How much content did the team generate? How many hours were saved? How many customers received personalized messages? How many workflows were automated?
If AI lets a marketing team produce twice as many messages but customers do not buy more, stay longer, upgrade more often, or cost less to serve, then the organization has increased output without necessarily increasing value.
That distinction matters more as execution gets cheaper. AI makes it possible to produce an extraordinary amount of marketing very quickly. It also makes it possible to produce ineffective marketing at extraordinary scale.
The scarce resource is no longer execution
For years, one of the biggest constraints in marketing was production. There were only so many emails a team could build, so many audiences analysts could create, so many creative variations designers could produce, and so many experiments an organization could manage.
AI is removing a meaningful portion of that constraint.
When execution becomes abundant, judgment becomes more valuable. Knowing what to send, who should receive it, when an intervention is actually needed, which outcome matters, and whether the intervention changed customer behavior becomes the
differentiator.
The next competitive advantage in AI marketing will not be the ability to generate more activity. It will be the ability to prove which activity created incremental value.
Lifecycle marketing has an advantage here
Lifecycle and CRM teams are unusually well positioned for this next phase because the work naturally produces repeated customer behavior that can be measured over time.
A lifecycle team can ask whether an AI-driven onboarding experience changed second-order retention. It can measure whether personalized replenishment messaging increased repeat purchase frequency. It can test whether an AI-generated win-back strategy actually brought back customers who would otherwise have stayed inactive. It can compare customer lifetime value, margin, churn, and purchase cadence across treatment and control groups.
That is much more powerful than reporting that an AI engine generated four million personalized messages.
The AI scorecard marketing leaders should use
The measurement conversation needs to move away from whether AI was used and toward whether it changed the economics of the customer relationship. I would start with five questions.
Did the AI-driven experience create incremental revenue, not just attributed revenue?
Did it improve retention, repeat purchase, expansion, or customer lifetime value?
Did it reduce a real cost, such as production time, service expense, or manual campaign work?
Did the result persist after accounting for incentives, discounts, model costs, and implementation expenses?
Can we demonstrate the impact with a credible counterfactual, such as a holdout or control group?
That last question may be the most important. AI makes targeting better, but better targeting can actually make attribution more misleading. If a model becomes very good at identifying customers who are already likely to buy, a campaign can look spectacular while producing very little incremental lift.
AI makes experimentation more important, not less
There is a temptation to assume that better models make traditional experimentation less necessary. I think the opposite is true.
As AI systems become more sophisticated, the gap between prediction and causation becomes easier to overlook. A model can predict who will churn, who will convert, and who is likely to respond. That does not automatically tell you whether the marketing intervention caused a different outcome.
Holdouts, incrementality tests, matched cohorts, and long-term LTV analysis become more valuable because they answer the question executives actually care about: what happened because we spent the money?
The executive conversation is changing
A year or two ago, a marketing leader could get executive attention by showing how aggressively the organization was adopting AI. Today, that is increasingly table stakes.
The stronger executive story is becoming much more concrete: we used AI to reduce the cost of producing lifecycle campaigns by 30 percent, increased the number of credible tests we could run, and improved incremental repeat revenue by 8 percent. Or we used an AI churn model to identify risk earlier, but a holdout showed that the intervention itself did not meaningfully improve retention, so we stopped spending against it.
Both are good outcomes. One scales what works. The other prevents the organization from paying for activity that only looks effective.
Reuters' broader market coverage on August 6 showed the same pressure across chip and software names, with investors punishing even AI-linked companies when performance or guidance failed to clear increasingly high expectations.
The experiment phase is over
None of this means companies should slow down their AI investment. It means AI is becoming mature enough that it should be judged like other major investments.
Capability matters. Adoption matters. Speed matters. But eventually every technology has to connect to economics.
For marketing, that means the question is no longer whether the team is using AI. Almost every serious organization is.
The question is whether customers behave differently because of it.
And if marketing leaders cannot answer that with evidence, they should expect the same reaction Wall Street is beginning to give companies with enormous AI budgets: show me the return.
Sources
Comments