Before choosing an AI tool, find the bottleneck

Marketing does not need another AI tool, at least not before understanding what is making the work difficult. The supply of tools grows every week, and each one’s demo is usually convincing. What the demo does not show is whether the problem it solves is the problem the team has.

Two problems that look the same

Consider two common situations.

In the first, a piece of content goes back for revision several times because no one clearly agreed on the audience and the message before production. A tool that generates the first draft faster does not solve this. It only brings the rework forward, and may even increase it, because now there are more versions to discuss.

In the second, a piece of information has to be copied manually between two systems every day. Here there is a well-defined, repetitive task with a result that is easy to check. This is the kind of situation where automation tends to work.

These are different problems. The first is about alignment and process. The second is about repetitive execution. Treating both with the same solution usually disappoints expectations in both cases.

What research indicates about value

McKinsey’s annual survey on AI in companies offers a useful data point for this discussion. In the March 2025 edition, among 25 attributes tested, redesigning workflows was the factor with the biggest effect on an organization’s ability to see financial impact from generative AI. At the same time, only 21% of respondents said their organizations had fundamentally redesigned at least some of their workflows (McKinsey, The state of AI, 2025).

The survey covers large organizations and aggregate results, and should not be read as a recipe. Its indication, however, is consistent with everyday experience: the gains come less from the tool on its own and more from how the work is reorganized around it.

Three observations before comparing tools

Before comparing options, I would start by observing the work as it happens today.

Where does the deliverable get stuck? Mapping the steps of a deliverable and noting where it waits, and for how long, often reveals that the bottleneck is not in production. It may be in approval, in missing information or in a dependency between teams.

What needs to be repeated? Tasks that repeat with the same format and the same rule are good candidates for automation. Tasks that repeat because something was poorly defined earlier are candidates for a process fix.

Who can tell whether the result is correct? If no one can evaluate the output with confidence, automating that step transfers the risk without reducing the work.

The solution may not be a new tool

Only after these observations would I compare tools, and I would include alternatives that are not AI products in the comparison. A briefing template can solve rework caused by lack of alignment. A well-structured spreadsheet can handle a consolidation task. An adjustment to an existing process can eliminate the step that seemed to require automation.

None of this rules out using AI. Often, AI comes in later, with a better-defined task and, because of that, a better chance of working. I wrote about this transition in From test to recurring use.

A justification that survives the demo

I am interested in experimenting with technology and I enjoy testing new tools. What I look for is a choice that comes with a justification that still makes sense after the demo. Three questions help check this: which task improves, who keeps the solution running and how the work continues when it fails. If these answers are not clear, the problem probably has not been described well enough yet to choose any tool.

References

• McKinsey & Company. The state of AI: How organizations are rewiring to capture value. March 12, 2025. Link