da
AdvisoryCasesInsightsAboutBook a Call
MethodologiesMethodology·5 min read·3 sections

AI in Marketing Analytics: What Works in Practice, and What's Still Hype

AI in marketing is full of promises that do not hold up in practice. Here is what AI actually delivers in marketing analytics today, what still requires a human, and where to start.

New AI platforms for marketing keep appearing, promising to automate strategy, predict the customer journey and think like a marketing director. Many of them solve a problem nobody asked to have solved. Meanwhile the more boring gains from AI in marketing analytics go unused, because they do not look as impressive in a sales presentation.

The dividing line is fairly simple: AI is very good at finding patterns in large amounts of data and weak at strategic judgement. Most of the disappointment with AI in marketing comes from mixing the two up.

01Where AI Makes a Difference Today

01

Monitoring and deviations

AI is far better than people at spotting unusual changes across many campaigns at once. That is exactly the kind of monitoring I described in the article on alerts instead of daily checks. Boring work that the machine does better than a person in front of a dashboard.

02

Bidding

Google and Meta's own automated bid strategies are today better than manual bid management in most cases where there is enough data. The precondition is that they get clean, accurate conversion data, ideally with sales data from the CRM, as I wrote about in the article on offline conversion import.

03

Drafts and variants

AI can write first drafts, create ad copy variants for testing and adapt messages to the roles I also wrote about in the article on buying committees. It does not replace positioning, but it speeds up production once a person has set the frame.

04

Predicting who becomes a customer

Models can find combinations of signals linked to sales that intuition would not find. But they need data. Google Analytics, for example, only shows its SourceGoogle: Predictive metricsAnalytics HelpGoogle Analytics can predict e.g. purchase probability, but only when at least 1,000 returning users have performed the action, and at least 1,000 have not, over a seven-day period within the last 28 days.View source ↗ when at least 1,000 returning users have performed the relevant action over seven days within the last 28 days, and at least 1,000 have not. Few B2B websites reach that, and the same applies to many other models.

02Where AI Is Still Hype

01

AI strategy

A tool cannot set your positioning, choose your target groups or decide whether to enter a new market. That requires knowledge of the organisation, the market and the risk that no model has. A "strategy generator" delivers something that looks like strategy but lacks the judgement that makes strategy valuable.

02

Insight on its own

AI can summarise data, but it does not ask the right question by itself. Valuable insight comes from a person who knows what is worth investigating, e.g. why a particular audience buys less after a price change. Then AI can help find the answer faster.

03

Self-driving marketing

Whether you can hand over set-up, budget allocation and optimisation to an AI system without human control is not primarily a technical question. The platforms themselves are moving towards more automation, and the technology is improving fast. What matters is who is accountable when the system gets it wrong: wrong budget allocation, ad copy that harms the brand, or targeting that breaches GDPR. Accountability cannot be automated, and as long as it lies with you, a person must have the mandate and overview to step in.

03How to Assess a New AI Tool

Ask three questions before you invest. Does the tool solve a pattern problem, where AI is strong, or a judgement problem, where AI is weak? Do you have enough data for the model to learn something, or is the vendor selling a promise that cannot be met with your volumes? And most importantly: who is accountable for the decision when the tool suggests something? If the answer is "the tool", you have handed accountability to something that cannot be held to account.

So should you buy a new AI tool? Start with what you already have. The ad platforms' automated bidding, the built-in alerts in Google Analytics and the AI already in your CRM and office suite cover much of the gain. What is most often missing is not tools but three rules: which AI tools may be used for what, who owns each tool and the decisions it affects, and which data, such as customer data and confidential figures, may be shared with external AI services. A new tool only becomes relevant when there is a concrete pattern problem you cannot solve with what you have, and enough data to solve it. For a quick picture of where you stand, take the AI maturity score and the AI governance check.

The takeaway

AI in marketing analytics is not something you need to wait for. It already delivers a measurable gain when used for what it is good at. The skill is knowing where the line is, and not letting a convincing vendor presentation take people out of the decisions that still require a person.

Want to work with this?

Let's talk about your situation.

Book an introductory conversation