AI in Marketing and Analytics. Where It Saves Time and Where It Breaks the Funnel

Every few years, marketing declares another era dead.

I remember when Big Data was announced as the end of the profession. Then the same fate was promised to attribution modeling, programmatic buying, blockchain in advertising, and the metaverse. Now neural networks and autonomous agents are the new headline act. The logic of the enthusiasts remains simple. Tomorrow everyone will be out of a job, from the junior copywriter to the chief commercial officer.

If you have watched this parade of promises for more than a few years, you simply stop reacting.

The current hype around artificial intelligence mirrors the rollout of every corporate software platform in history. Leadership returns from an industry forum where everyone carried the same lanyards and watched the same slides. The word transformation begins echoing through the company. Consultants are brought in, enterprise subscriptions are signed, and calendars fill with endless Zoom meetings. Six months later, this massive machine produces two measurable outcomes. A summary of an hour-and-a-half call where nobody decided anything. And pages of flat, lifeless prose that everyone feels awkward showing to clients.

The technology does have practical value. It just lives in an entirely unromantic place.

A language model is brilliant at the kind of dull mechanical grunt work that used to ruin a Friday evening. Moving raw data from one messy export to another without manual spreadsheet wrangling. Sifting through a massive dump of search queries to cluster them by intent. Translating dense technical documentation into readable language. Spotting a broken tracking script or a misplaced tag in a campaign URL.

In those tasks, the algorithm works exceptionally well. It never gets tired, never argues about creative vision, and delivers in sixty seconds.

The trouble begins the moment you ask the machine to think for you and draw conclusions.

The danger of large language models is not that they make mistakes. Everyone makes mistakes. The danger is the absolute, unshakeable confidence with which they deliver complete nonsense. A model will gladly fabricate an interpretation of your campaign performance, confuse correlation with cause, and attach a clean chart to make its fantasy look rigorous.

When I first started testing these systems, I fed one a full campaign dataset. It eagerly spat out optimization recommendations. The logic sounded bulletproof. These specific ad groups drive cheap purchases, so pour more budget there. Those upper-funnel campaigns have zero direct bookings, so kill them immediately.

The machine had no concept of first-click versus last-click attribution. To the software, the customer journey simply did not exist. It failed to grasp that the awareness campaign introduced the airline to someone who was not even planning a trip yet, while branded search simply collected them at the checkout page. If a marketer followed that advice and slashed the top of the funnel, new customer flow would evaporate within two weeks. The bottom of the funnel would dry up on its own.

The model will never admit this blind spot. It will never say that it lacks the full path to conversion. It simply drops this garbage on your desk as a finished result.

If an analyst does not deeply understand how their numbers reconcile in Google Analytics or Yandex Metrica, they take that bad advice and proudly present it upstairs.

In the travel industry, the cost of that ignorance is calculated immediately. Fail to verify the logic, trust the algorithm's tidy summary, and cut prospecting on a route. In a few days, your aircraft takes off half-empty, the ad spend is gone, and you face a very grim conversation with the commercial director about why Thursday's flight operated at a loss when the dashboard looked so green.

Algorithms do not pay for corporate losses. The machine does not care whether a route turns a profit or sinks the monthly financial plan. Responsibility always rests on the person who approved the campaign and pressed the button.

My rule for any new tool remains simple.

If software reliably saves me two hours of manual labor a day, I use it. If it forces me to spend three hours auditing its hidden hallucinations, or presumes to teach me strategy based on industry averages, it goes into the trash.

The evangelists will move on to the next shiny thing soon enough. The ability to look at numbers with a cold, sober eye will remain the only skill that actually matters.