Writing · Essay

Diagnosis before tools

The instinct in most companies right now is to reach for AI first and ask what problem it solves second. The more useful skill, and the harder one, is knowing when the right answer isn't AI at all.

The same rule runs through my operations work and my projects: name the problem before choosing the fix.

A symptom is not a diagnosis

For example, a company may say that too many customers are calling and immediately assume it needs a chatbot, but call volume is a symptom rather than a diagnosis. The first question should be why customers are calling. If three questions account for most of the volume, a better FAQ or clearer confirmation email may solve most of the problem before AI is needed.

Building the chatbot first could leave customers with the same unanswered questions while adding another system the company has to pay for and maintain.

Before I'd recommend AI for something like this, I want to know

1.What's driving the behavior, not just how much of it there is
2.Whether a simpler, non-AI fix already gets most of the benefit
3.What's left over once the simple fix is in place, and whether that remainder is actually where prediction or personalization would help

Should this use AI at all?

Work through a real or hypothetical problem. Each question can end the process early, which is the point: most problems should exit before they reach a model.

Question 1 of 8

Can you state the business problem clearly?

The better question, by function

In each function, the usual question asks how to make existing work faster. The better question asks whether the work, or the problem behind it, could change: preventing an issue earlier, improving the decision, or changing when the work happens.

Function	The usual question	The better question	What changes
Finance	Can we process invoices faster?	Can we prevent bad invoices from entering the workflow at all?	The check moves from after the fact to before the payment happens.
Customer service	Can we answer tickets faster?	Can we predict and prevent the issue before the customer has to call?	The goal changes from answering customers faster to preventing avoidable calls in the first place.
HR	Can we screen resumes faster?	Can we predict role fit while reducing bias?	The focus moves from filtering résumés faster to improving the quality and fairness of the hiring decision.
Supply chain	Can we forecast demand better?	Can we redesign ordering and inventory around live prediction?	Forecasting becomes a continuous input instead of a quarterly exercise.
Sales	Can we write outreach faster?	Which customer needs which message at which moment?	The goal moves from producing more outreach to reaching the right customer with the right message at the right time.
Operations	Can we inspect defects faster?	Can we predict defects upstream before they happen?	The focus moves upstream from catching defects to preventing the conditions that cause them.

The same decision pattern applies across AI systems

Fraud detection, demand forecasting, predictive maintenance, computer vision, dynamic pricing, and recommendation systems all begin with the same basic structure. They use available data to predict something the business does not know yet, and then the business decides what action should follow. The examples above apply that structure to six different functions, but the value comes from the decision and workflow around the prediction rather than the technology alone.

For the full framework behind this pattern, read AI strategy starts with the decision, not the model.