AI applied to marketing is sold as a lever for efficiency. But there's a cost that almost no one accounts for: continuous rework, model maintenance, and human supervision that never completely disappears. These are the same mistakes made with programmatic advertising years ago, when the promise was to cut costs and automate everything.
Those managing advertising budgets in an SME or mid-market company need to answer a concrete question: do the savings declared by AI take these hidden costs into account? If the answer is no, the real ROI is much lower than what appears in the reports.
This article analyzes the parallelism between the promise of programmatic and that of AI, and suggests how to honestly evaluate an investment before budgeting it.
The broken promise of programmatic: a costly déjà vu
Programmatic advertising — the automated buying and selling of ad space in real-time — entered the market with a clear promise: less waste, more targeting, and plummeting operational costs. Companies invested. Then they realized that managing the platforms required specialized skills, that data needed cleaning and updating, that ad fraud was eating into budgets, and that creative control had been lost along the way.
Today the same narrative is repeated with AI applied to marketing. This is reported by Search Engine Journal : before declaring efficiency gains, you need to count the hidden costs of rework and continuous maintenance. If you don't, you risk repeating the same mistake and weakening your long-term strategy.
Where the costs really hide
When a company adopts an AI tool for marketing — whether it's for generating copy, optimizing campaigns, or managing attribution — the visible cost is the license or subscription. The invisible one is everything else:
- Rework : AI output needs to be reviewed, corrected, and adapted to the brand's tone. Who does it, how long does it take, how much does that hour cost?
- Prompt and model maintenance : the instructions given to AI age. Products change, the market changes, platforms change. Someone has to update everything.
- Human supervision : automation does not eliminate responsibility. An AI-generated error on an active campaign can cost much more than the savings achieved.
- Integration with existing systems : CRM, ad platforms, analytics. They rarely connect without friction.
These costs don't show up in the demos. They appear in the invoices of subsequent months and in the time taken away from the team.
The parallel with campaign automation today
It's not a theoretical problem. Anyone managing campaigns on Google or Meta already knows that automation is never truly autonomous. We discussed this when analyzing how to stay in control when Google and Meta Ads automate decisions : platforms optimize for their own goals, not necessarily for the advertiser's. Governance remains a human job.
The same applies to attribution. Models that distribute conversion credit among channels can be misleading. We analyzed the topic in detail by discussing Advertising attribution and risks for SME budgets : a wrong model shifts budget towards channels that seem to perform, not towards those that actually perform.
Adding AI to an already imperfect attribution system doesn't solve the problem. It hides it better.
How to evaluate real ROI before signing
Before budgeting any AI solution for marketing, it's worth doing an honest calculation. The platform cost alone is not enough.
- Monthly supervision hours : estimate how many hours the team will spend checking, correcting, and updating outputs. Multiply by the real hourly cost.
- Cost of rework : how many times a month is an AI-generated content or creative rejected or rewritten? Every rejection is a cost.
- Initial integration cost : development, configuration, team training. Usually underestimated by 40-60%.
- Cost of error : what happens if the tool produces the wrong output on an active campaign? Do you have a control plan?
Only after estimating these numbers does it make sense to compare them with the savings declared by the vendor.
What to do in the coming days if you are considering an AI investment
If you have already adopted AI tools for marketing, or are about to, there are some concrete actions to put on the agenda this week.
First of all, ask your team how many monthly hours they already dedicate to supervising and reworking AI outputs. If nobody knows, that's the first problem to solve: you can't measure what you don't track.
Then, it maps the integration points: where the AI tool connects with your ad platforms, your CRM, your analytics. Each connection is a potential friction point.
Finally, consider the issue of non-human traffic. Those managing performance campaigns already know that a portion of the budget is consumed by bots and automated traffic. Adding AI to the mix without a protection system can amplify the problem: we wrote about it analyzing how to protect your ad budget from bots and AI traffic in 2026 .
For those who want a broader picture of current trends in paid media, the starting point is the section dedicated to advertising, paid media and video from our blog, where we gather the most relevant analyses for the Italian market.
AI in marketing isn't a scam. It's a tool that works when adopted with calibrated expectations and a complete cost calculation. The problem isn't the technology: it's the tendency to buy the promise instead of verifying the numbers. Programmatic already taught us this lesson once.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.
