AI is changing marketing and IT analytics. The numbers still need an owner

Rate this article ↓

Marketing teams buy attention. Analytics teams try to determine whether that attention produces results. AI puts pressure on both kinds of work by making first drafts, variations and explanations cheaper to produce. The employment question is whether a business uses that capacity to do more, needs fewer people or discovers that checking the output consumes much of the saving.

All three outcomes are plausible. A useful analysis starts with particular workflows instead of treating every marketer, analyst and IT specialist as interchangeable.

Evidence of gains, with boundaries

A working paper listed by Columbia Business School reports randomized experiments at a large online retailer during 2023–24. Seven consumer-facing workflows were tested; reported sales effects ranged from zero to 16.3%, depending on the application. Four applications produced positive effects. The experiments support a business case for some uses, while the variation argues against assuming every deployment will pay. They measure outcomes at one platform, not employment losses throughout marketing. Columbia research summary.

A separate Harvard–BCG experiment involved 758 consultants performing tasks involving writing, creativity and analysis. Researchers found substantial gains on tasks within the model's capabilities and emphasized that those capabilities were uneven. Consulting tasks are not a direct measure of marketing departments, but the study supplies a useful warning: success on one assignment does not establish reliability on the next. Harvard research summary.

Where the workforce pressure falls

For marketing, plausible candidates for automation include producing initial copy variants, reorganizing approved product information and drafting campaign summaries. A small team could test more alternatives without hiring the same number of production specialists. Alternatively, a company with fixed demand could reduce outsourced volume. These are possible business responses, not measured forecasts from the cited studies.

In IT analytics, the pressure concerns the path from data to decision. An assistant might draft a database query or explain a chart. Someone still has to establish which records belong in the calculation, whether a join duplicates customers and whether the result answers the business question.

Consider a hypothetical campaign report. Sales rose after a new advertisement began running, and an automated summary attributes the increase to the campaign. But the same period included a discount and a holiday. The chronology alone cannot distinguish those effects. Producing the explanation faster does not fix the comparison.

That distinction changes the value of skills. Teams should reward people who can define a credible test, inspect the data and challenge an attractive conclusion. Fluency in prompting helps, but it cannot substitute for knowing what a valid result looks like.

Cheaper production changes the career ladder

Entry-level employees often learn through the routine work that organizations would like to automate. If a firm removes those assignments, it should deliberately provide other ways to acquire competence: supervised experiments, independent checks and responsibility for explaining discrepancies.

Otherwise, it risks retaining senior reviewers while shrinking the pipeline that produces their successors. That is an organizational risk inferred from the proposed workflow, not proof that AI has already caused such a shortage.

The strongest case for aggressive adoption is competitive. A firm that insists on manual production while rivals improve conversion and speed could lose customers. Protecting every existing task is not a sustainable strategy. But neither is declaring success because a team produced twice as many reports that nobody can defend.

CivitasPoint assessment

Evaluate a campaign tool against conversions and profit after its full costs. Evaluate an analytics tool against correctness, reproducibility and the time required for review. Record the model-assisted work, preserve the underlying query or calculation and give a named employee responsibility for the conclusion.

Staffing decisions should follow those measurements. AI can plausibly reduce the labor needed to produce marketing material and routine analysis. Whether it creates a better business depends on the quality of the decisions that follow. The person who can explain why a number should be trusted will still have a clear job to do.