“Where can we add AI?” is often the wrong first question. It encourages a catalogue of tools and demonstrations before anyone has identified a business constraint. A more useful question is: which repeated activities consume time, depend on organising information or generating variants, and can be accelerated without losing control of quality?
AI’s strongest contribution to a marketing team is not necessarily replacing the team. It can reduce friction: prepare a research map, compare source material, structure a brief, create test variants, draft a report or help a salesperson find relevant context. That leaves more time for diagnosis, customer conversation, creative judgement and decisions for which someone must remain accountable.
Start with a workflow and a test
Before choosing a model, describe the existing process: where the input comes from, who uses the output, how errors would be detected and what a successful change would look like. A good early workflow has clear boundaries and reviewable results. For example, AI might turn verified product documentation and interview notes into a first draft of an FAQ. An editor then checks each claim against its source before publication.
Compare this with instructing a model to publish dozens of unsupported articles about a client’s industry. Both are “AI content”, but only the first has a credible route to reliable work. The same distinction applies to analytics: a model can find anomalies and summarise a dashboard, but cannot infer a causal business explanation from incomplete data without assumptions being tested.
Good candidates for assistance
Research synthesis, meeting-note organisation, brief preparation, first-pass classification, repeated formatting, variant generation and routine quality checks are promising candidates. They have a human-readable output and can often be compared with source material or acceptance criteria. AI can also support sales by grouping common objections or surfacing questions raised in customer conversations — provided personal information is handled appropriately.
The more sensitive the action, the more important the boundaries become. Drafting an internal summary is different from sending a message to a client, changing a production website or making a public claim. Define which actions require approval and what evidence a reviewer should see. Automation should be reversible where possible.
What should remain with people
Positioning, reputation, ethics, client relationships and the interpretation of ambiguous commercial evidence should not be outsourced to an unaccountable tool. A model can propose alternatives or pressure-test a strategy. It does not own the consequence of a misleading claim, a poor campaign or an insensitive response. The responsible person needs both the authority to reject the output and sufficient context to do so.
This is not anti-technology. It is the condition for using technology well. When people retain responsibility for goals, sources, risk and final sign-off, AI can make the production loop faster without making the organisation less thoughtful. Read From SEE to THINK for the larger shift from visual control to intent-driven work.