Almost every marketer I meet wants to tell me about their content team.
How much faster it is moving. How many variations they can test in a day. How work that used
to take a week now takes an afternoon. I nod along, because all of it is true and most of it was
hard won.
It is rarely the part of the conversation I am still thinking about a week later.
What I almost never hear, unprompted, is whether AI has changed who in the organisation makes a call. Or how quickly they make it. Or what evidence sits underneath it when they do.
Sit with that for a moment and it stops being surprising.
Automating content asks very little of anyone. The team still briefs the work. Someone still approves it. The approval chain is intact, the reporting lines are intact, and nothing about who answers for the outcome has moved an inch. It is a clean, contained win, and I do not want to undersell it. Speed is real value.
But it is also the kind of win that gets copied within a quarter. The tools creating that advantage are the same tools sitting in everyone else’s stack, priced within reach of every competitor in the category.
The harder shift begins when AI stops helping people produce more and starts changing what they decide to do.
A churn signal that surfaces three weeks before the monthly report would have caught it. A pattern in customer behaviour that quietly contradicts what the account team has been telling leadership for a year.
The technology that produces those findings is no longer the difficult part. The difficult part is that someone now has to own doing something about them, and be the person asked, two quarters later, whether it actually happened.
I have sat in enough of these rooms to notice how carefully most organisations avoid that conversation. It is far easier to approve another content tool than to decide who is accountable when a model surfaces something inconvenient about a client, a channel or a strategy that senior people have publicly backed.
When MMA India ran its AI Marketing Maturity Study with EY and Mobavenue, one finding matched what I had been watching happen in the field.
The highest-impact use case in the study is propensity modelling: using AI to predict who is likely to buy, churn or upgrade. Among marketers using it, 88 percent report high or medium business impact. Close to half the marketers surveyed have not adopted it at all.
The easiest wins are the ones being fought over. The harder, more valuable one is sitting on a shelf.
I do not think that is an accident, and I do not think it is a technology gap. The models are not exotic. The data requirements are not out of reach for a mid-sized business.
What makes propensity modelling harder is what it produces. Content generation gives you an asset. You look at it, you approve it, you ship it. Propensity modelling gives you something closer to an instruction. It tells you this account is leaving, this segment is worth less than you assumed, this spend is not returning. An asset can be reviewed. An instruction has to be acted on or explicitly set aside, and both of those need a name attached to them.
McKinsey’s global research on AI trust arrives at the same place from a different direction. Companies with a named owner for AI governance score meaningfully higher on maturity than companies without one, 2.6 against 1.8 on a five-point scale. The variable there is not technical sophistication. It is whether the responsibility has an address.
AI rarely fails to create value on its own terms. It fails when nobody’s job depends on doing anything with what it produces.
That is not a new problem, if I am honest about it. It is the same reason good research used to die inside a slide deck, and dashboards got built and never opened. What AI changes is the visibility of the gap. The insight now arrives faster than most organisations are structured to respond to it, so a delay that used to be invisible has become measurable.
None of this calls for a new function or another layer of governance. In the organisations I see handling it well, ownership tends to look quite ordinary. Someone senior enough to move budget is named against the model’s output. There is an agreed answer to what happens when the model disagrees with the plan. And someone looks back, periodically, at whether the decisions the model informed turned out better than the ones it did not.
That last one matters more than it sounds. Most companies now measure AI adoption. Very few measure whether it improved a decision.\
Markets are not moving through this at the same pace, and it would be a mistake to read one market’s maturity onto another. But the sequence looks consistent wherever I look.
Organisations reach for the automation win first because it is legible, fast and safe. The decision layer work waits, because it asks something of the people rather than the tooling. And the companies that get there early are usually not the ones with the largest technology budget. They are the ones willing to be specific about who owns what.
If I had to guess what will separate the businesses that look back on this period as a genuine turning point, it would not be how much they automated.
It would be something simpler.
Can anyone in your building tell you which decisions are better today because of AI? And can they tell you whose job it is to keep that true?
Most people I ask go quiet for a second before they answer.
That pause is usually the real answer.
