How to pick your first three AI use cases
List candidate processes with the people who do the work today, then score each on business value and complexity, weighting value higher so technically elegant projects with no metric don't survive. Scale the high-value, low-complexity ones, pivot or set a deadline for high-value, high-complexity ones, and stop the rest. The best first use cases are often boring: frequent, text-heavy, already measured, and reviewable by a person.
Why the first pick often goes wrong
The first use case chosen is often the most visible one, a customer-facing chatbot, which is also the riskiest and hardest to measure. Document processing, internal search, drafting, routing and compliance checks are less exciting, and they're frequent, measurable and low-risk when a person reviews the output.
Value and complexity
Value is real business impact: revenue, cost reduction, strategic position or risk reduction. Complexity is what it takes to deliver: whether the data is available, compliance exposure, whether the team can build it, and how many months it needs.
| Low complexity | High complexity | |
|---|---|---|
| High value | Scale: put resources here first | Pivot: name the blocker, set a 90-day deadline |
| Low value | Absorb or stop: replace with an off-the-shelf tool | Stop first: burning money on something that can't pay back |
Scorecard
Score each criterion from 1 to 5, where 5 is better.
| Criterion | Weight |
|---|---|
| Business value (hours saved, cost, revenue, risk) | 3 |
| Data ready and accessible | 2 |
| Team can build and run it | 2 |
| Risk if wrong (5 = low risk) | 2 |
| A person can review outputs | 1 |
| Strategic alignment | 1 |
| Needs AI at all (5 = clearly yes) | 1 |
The weights are a starting point to adjust, and the last row matters: if a script or an existing product does the job, use that.
Pick three
- One quick win: back-office, frequent, reviewable, with data ready.
- One with larger value and moderate risk.
- One strategic bet with a clear blocker to remove.
Where quick wins usually are
Customer service and financial operations are common starting points: support queries, invoice processing, reconciliation and fraud checks are slow, error-prone and measured already, so a change shows up in numbers quickly.
From the floor
My first question to any client is: show me how you do this today and tell me what exactly you want to improve.
In the founding session of the AI leadership cohort, we scored ideas with a weighted sheet where business value is multiplied by 3, data readiness and similar factors by 2, and strategic alignment by 1, and a participant pointed out that without weights everything sits on the same scale and the score stops being useful.
In the webinar on AI agents for business leaders, I recommended starting with customer service and financial operations as quick wins, and keeping agents simple, each solving one specific problem, because that's where they work well.