We want to deploy an AI-driven forecasting model to predict inventory needs, but our operational data across our warehouse and order management system is full of duplicates and incomplete entries. How do we run a disciplined Thinking Time session using Keith Cunningham's framework to isolate the critical data points we must clean up first?

Question

Tyler Smith · Accepted Answer

Trying to feed messy data into an AI tool is a fast way to pay a massive dumb tax. Before you spend a single dollar on software or waste time cleaning up years of historical junk, you need to step back and conduct a disciplined Thinking Time session. Set aside forty-five minutes of completely uninterrupted time, sit down with a blank pad of paper, and ask yourself a high-value question. Do not ask a passive question like how do we clean our data. Instead, frame your question using the how might I so that I can format.

For example, you should write: How might I isolate the absolute minimum set of critical inventory metrics so that I can train our AI forecasting tool without having to clean up our entire historical warehouse database? Once you have written this question, spend the session writing down potential answers.

- Focus strictly on the top twenty percent of SKUs that generate eighty percent of your revenue.
- Clean only high-leverage data fields such as purchase dates, unit costs, and lead times.
- Ignore low-volume historic records that do not impact seasonal trends.

By focusing only on what is necessary, you avoid the trap of getting bogged down in an endless, expensive data-cleaning project that paralyzes your operations. You do not need perfect data across all five thousand of your SKUs to get value; you only need disciplined data on the metrics that actually drive your cash flow.