Most businesses shouldn’t begin an AI initiative by asking where they can use AI.
That question assumes the work already makes sense, the information is reliable, and technology is the missing piece. A lot of the time, those assumptions don’t hold.
I’d start with the operation instead. Find the work that has become unnecessarily hard. Understand why it works that way. Then figure out what kind of change would actually help.
AI may be part of the answer, but choosing the technology comes later.
Start with the work people struggle to get done
If I were talking to a business owner, I wouldn’t begin by asking which tasks they want AI to perform. I’d ask what’s making the business harder to run.
What gets more difficult every time the company grows? Where are customers waiting? Which process depends on one person knowing where everything is buried? What does the team keep doing because nobody has had time to question it?
Employees usually know where to look. They can tell you which report takes hours to prepare and five minutes to use, where they have to check three systems before trusting an answer, and which official process bears little resemblance to what people actually do.
Those conversations reveal the operation beneath the organization chart. They show you where people are quietly connecting systems, resolving conflicting information, filling gaps in a process, and keeping work moving through personal knowledge.
They may also reveal that some of the work shouldn’t exist.
Suppose someone spends two hours every Friday creating a report that nobody uses. Generating it in thirty seconds sounds like a win, but it doesn’t solve the more important problem. Why doesn’t anyone use the report? Do they distrust the numbers? Is it answering the wrong question? Has the decision it once supported disappeared?
The right answer might be a better data source, a clearer metric, or the end of the report entirely.
The visible task is rarely the whole job
At one company, I mapped more than 100 tasks across the go-to-market organization. Many of those tasks contained four or five smaller steps.
Something that sounded simple might require a person to search several systems, compare conflicting information, ask another team for context, make a judgment call, update a record, notify someone, and document what happened.
The visible task was often the last step. Most of the work happened around it.
I’ve seen the same pattern in other organizations. Understanding a customer problem could require support conversations, help content, product analytics, account information, and knowledge spread across several teams. In another role, customer context had to survive handoffs between support tools, project-management systems, design files, team conversations, and product groups.
The information existed. People still had to assemble it before they could act.
This is where many AI projects disappoint. They automate the most visible part of a task while leaving the surrounding coordination untouched.
An employee may get a faster draft, summary, or recommendation. But they still have to find the source material, check whether it’s current, resolve contradictions, move the result into another system, and decide what happens next.
The output arrives faster. The work around it is still there.
Understand why the process became complicated
Most frustrating workflows didn’t start out that way. They were shaped by old constraints, unusual situations, and reasonable decisions made at different points in the company’s history.
Someone copies customer information into three systems because those systems were never connected. A manager approves every request because a few unusual cases once caused problems. Several teams prepare similar reports because each inherited a slightly different process. An employee maintains a spreadsheet because the official system doesn’t reflect how the business actually operates.
AI could help perform all of those activities. That doesn’t mean automating them is the right move.
Before changing a process, I want to understand why each step exists. Some steps create value. Some protect the business from a real risk. Some compensate for a system limitation. Others remain only because nobody has questioned them.
The same is true of the information people use. Where does it come from? Which source should be trusted? What lives only in someone’s head? What happens when two systems disagree?
An AI system can be capable of performing a task and still fail because it doesn’t have the context needed to make a good decision. If the underlying information is stale, incomplete, or contradictory, the system may produce a perfectly plausible answer that an employee then has to investigate.
Now the employee has one more thing to verify.
Use the simplest solution that fits the work
Once you understand the operating problem, choosing what to use usually becomes much easier.
If the rule is exact, use something exact. A scheduled query may be enough to produce a recurring report. A standard integration may stop employees from copying information between systems. A workflow tool may handle predictable routing and notifications.
AI becomes more useful when the work isn’t completely predictable. It can help interpret messy text, compare documents, identify patterns, or assemble information for someone who needs to make a decision.
Some workflows need a combination. Ordinary automation can handle the predictable steps. AI can help with interpretation. People can still own decisions that carry meaningful consequences, require judgment, or depend on a relationship.
That mix is often more dependable than asking one AI system to run the entire process simply because it can technically interact with every part of it.
The goal is to make the business work better. Sometimes an AI-assisted workflow will do that. Sometimes the answer is connecting two systems, fixing the information, simplifying the process, or stopping work that no longer serves a purpose.
A useful operating strategy has to leave room for all of those answers.
Start with discovery, not a technology bet
For most businesses, I think the first phase should be discovery.
Talk to leadership about the outcomes that matter. Talk to employees about how the work actually happens. Follow a few important workflows from beginning to end and look closely at the people, systems, information, handoffs, and exceptions involved.
Pay particular attention to the moments when someone has to search, copy, reconcile, or ask around before the real work can continue. Those coordination tasks are easy to overlook because they’re spread across people and systems. Together, they can consume a remarkable amount of time.
This gives you a clear picture of what’s happening before you commit to a solution. You can see where automation might remove real friction and where an apparent AI opportunity is blocked by unreliable information, unclear ownership, or a process that needs to be redesigned first.
AI can make a sound operation faster and more capable. It can also make a confused operation move faster without becoming any less confused.
Most businesses don’t need to begin with an AI strategy. They need to understand how their business works, where people are struggling, and what kind of change would genuinely make the work easier.
Once that’s visible, AI can compete with every other possible solution on its merits.