The easiest thing to do with AI right now is add another tool. Sign up for something new, test a feature, buy a licence someone recommended. It feels like progress.

But more tools don’t automatically make the work better. The harder question is simpler: what are we actually trying to improve? And finding a useful application for AI isn’t the same as making it work inside an organization. Identifying a use case is the beginning. Implementing it is where many organizations stall.

In McKinsey’s State of Organizations 2026 survey of more than 10,000 leaders, 86% said their organizations were not very prepared to adopt AI in everyday operations. The issue isn’t necessarily access to technology. It’s whether organizations have the people, processes and ownership in place to make it useful.

Start with the work, not the tool

If the problem is reports that take too long to prep, repetitive partner follow-up, documents that take forever to review, or meeting notes that need turning into next steps, the first question shouldn’t be “what AI tool should we buy?” It should be “what would a better version of this work look like?”

Specificity is what makes that possible. “Use AI for marketing” isn’t a use case. “Turn a raw partner recap into a structured first draft in our usual format” is. “Use AI for leadership” isn’t a use case. “Pull the five key points, risks and questions from a 40-page report before a meeting” is.

Good candidates share a pattern: work that happens often, takes longer than it should, follows a repeatable structure, and involves a lot of reading, writing or organizing, with a person reviewing the final result.

A use case isn’t an implementation plan

Take sponsorship reporting, the kind of work partnership teams in hockey organizations do every month.

Someone spends hours collecting performance numbers, checking which deliverables have actually run, gathering screenshots and building the report a sponsor will read.

AI can help with parts of that: organizing the information it’s given, drafting the summary, formatting the report in the team’s usual structure. What it can’t do is fix information that lives in five different places.

If performance data sits in one spreadsheet, deliverables are tracked in someone’s inbox, and screenshots live on an employee’s phone, no tool fixes that on its own. Someone still has to decide where the information lives, who keeps it current, what the report includes, and who checks it before it goes to a sponsor. That’s the work most organizations skip.

The 2026 AI and Data Leadership Executive Benchmark Survey found the same pattern: 93.2% of respondents named cultural and change management challenges, not technology, as the biggest impediment to adoption. McKinsey also identified unclear executive ownership as another challenge to AI adoption.

Decide what success looks like

First, write down what the work costs today and what better would look like.

Say the monthly sponsor report takes a partnership manager six hours to prepare. The goal might be to cut that time in half without losing accuracy. Then measure the whole workflow, not just the part AI touches: total time start to finish, time spent reviewing and correcting the output, and the quality of what the sponsor actually receives.

A draft produced in five minutes isn’t a productivity gain if someone spends three hours fixing it. I wrote about that gap in AI Adoption Is Not AI Productivity. The short version: measure whether the work improved, not how fast the tool produced something.

Start small, and build something repeatable

Pick one meaningful workflow and build it properly. Start from the problem in plain terms, decide what success looks like before you test anything, and try it with tools you already have where that’s realistic. Expand only when there’s evidence the change is helping.

For the workflow to stick, a few questions need owners, not just answers. Who maintains it? Does the team actually know how to use it? Is the information it relies on accessible, accurate and properly protected? Who checks the output and makes the final call? And how will you know people are still using it a month from now?

Sports organizations run lean: small teams, people covering multiple roles, no spare department to manage a complicated rollout. One workflow done properly beats five disconnected experiments. I looked at which hockey workflows are actually worth automating in AI Agents Are Coming to Sports Marketing.