Insights · Measurement · August 2026
Measuring the ROI of AI, in your numbers, not the vendor’s
Industry benchmarks put average AI returns near $3.70 per dollar invested, but averages are marketing. The only ROI that matters is measured in your own baseline, taken before anything is built.
Vendor ROI is a brochure number
Every AI vendor quotes a return. The number is real somewhere: averaged across companies with different sizes, data quality, and discipline, and it says nothing about yours. Deciding by benchmark is how organizations end up with impressive pilots and unimpressive P&Ls.
The fix costs one week: take a baseline before anything is built. How many enquiries arrive, how many are answered, how long follow-up takes, how many hours the target process consumes. Without a before, every after is an anecdote.
Three numbers per system, agreed in writing
A primary number the system must move: enquiries answered, hours returned, days-to-invoice. Chosen from reporting you already trust, so nobody debates the scoreboard.
A guardrail number that must not degrade: customer satisfaction, error rate, refund rate. Automation that moves the primary number by breaking the guardrail is a failure wearing a costume.
A cost line that includes everything: build, subscriptions, the hours your team spends supervising. Honest denominators make honest ROI.
Check on a date, decide on the result
The scope names the date the numbers are checked. If the system moved its number, expand it. If it did not, fix it or kill it: publicly, without embarrassment, because a program that kills its failures earns the credibility to fund its successes. That discipline, more than any model choice, is what separates organizations that compound value from those that collect pilots.
Decide where AI belongs in your business
The readiness engagement maps your operations, sets the governance frame, and hands you a measured 90-day plan, before you spend on tools.