One of the biggest mistakes organizations make with AI is failing to measure its value. We describe it as faster, smarter, or more efficient—but rarely quantify what that actually means. Without measurement, AI remains a novelty. With measurement, it becomes infrastructure.

That’s why I use a simple personalization setting in OpenAI: after every interaction, the model estimates how long it would have taken me—and a competent professional—to produce the same output manually. It presents those estimates in a structured table, maintains cumulative totals, and logs the savings over time. It sounds procedural, but it changes everything.

When AI produces an answer in seconds, we anchor on the speed of delivery rather than the time it would have taken otherwise. Writing a structured article might have required forty minutes. Researching and synthesizing a technical explanation could easily take an hour. Formatting, refining, validating—these steps add up. Forcing the comparison makes the time compression explicit.

Over time, the cumulative totals become revealing. Individual savings of fifteen or twenty minutes compound into hours per week and meaningful capacity per quarter. The data stops being anecdotal and starts becoming operational. AI shifts from “useful tool” to measurable productivity engine.

There’s also a strategic dimension. Many tasks don’t just require time—they require experience. When AI produces work that would normally demand a seasoned professional, it compresses not just effort but expertise. Tracking this clarifies where AI is amplifying capability rather than merely accelerating output.

If AI is going to justify its place in enterprise budgets, it needs defensible metrics. Systematically tracking time saved provides them. It turns subjective efficiency into structured evidence.

For me, that discipline has reframed how I think about AI entirely. I no longer ask whether it feels productive. I measure how much time it returns. And what gets measured gets optimized.

The Personalization Prompt I Use

Below is the exact instruction I use in my OpenAI “Personalization/Custom Instructions” settings – I also have a short statement in my “More About You” which closely matches my LinkedIn summary.

For every interaction:
• Estimate how long it would take me and a competent person to produce the same answer (including research, formatting, and validation).
• Present these in a summary table with the time I would have taken to arrive at the answer and the time a Competent Person would have taken as columns, and always include a final row with cumulative totals.
• After the table, output a one-sentence summary of the time saved for that specific interaction, explicitly mentioning the task type.
• Always maintain an internal interaction log table with: date, interaction description, task type, time saved (me), time saved (competent person), and estimated expertise for the competent person (years). The log must also include a final cumulative totals row. Do not display the log unless explicitly asked.
• Be professional, concise, and structured. Always show the summary table and per-interaction time saved after every interaction; update the log silently unless requested.