Following up with leads
Five minutes beats thirty, by a lot.
What was tested. The Lead Response Management study (James Oldroyd with InsideSales, 2007) followed thousands of web leads and how fast companies called them back.
What was found. The odds of reaching a lead were about 100 times higher when the business called within five minutes than at thirty minutes, and the odds of qualifying the lead about 21 times higher.
What we do now: for a service business, the first thing we build is whatever gets a person on the phone fast. For an insurance agency, that was an alert that goes out the moment a lead arrives.
Published study, 2007. Applied in our insurance work, October 2026.
Advertising in Greeley
Local readers trust a face and a free written report.
What we tested. Before placing our own ad, we read every ad in the June 2026 issue of a Greeley magazine.
What we found. The strongest small ads showed a named local person. Readers already respond to a free inspection that comes with a written report, the way a roofer offers one. No technology or AI company advertised at all.
What we do now: our check-up is free, written down, and comes with a person you can meet in Greeley. If you advertise locally, the same pattern probably applies to you.
Noval review, October 2026.
Self-storage income
Most of the money was in decisions, not software.
What we tested. For two storage facilities, we read the industry's own reports on what raises a facility's income and what lowers its costs.
What we found. The income levers were mostly business decisions: pricing for existing tenants, tenant protection plans, fees and promotions. Software showed up mainly on the cost side, in answering calls and cutting staff hours.
What we do now: before buying AI, we look for the money levers an owner already controls, then use AI to watch them, the way our market watch reads competitor prices twice a week.
Noval research, 15 October 2026, built on industry and company reports.
Running a company on records
Urgency is a poor guide to what gets done.
What we tested. We record every decision and every piece of work, and whether it got finished. Then we trained a model on 771 of those records to predict which work actually gets done, and tested it on work it had never seen.
What we found. It ranked the work far better than sorting by urgency, which did worse than a coin flip. Asked to predict how 24 decisions would go before they were made, it matched the real call 19 times.
What we do now: the model's ranking sets the order of our week, and its predictions are checked against what really happened. Any business that writes its decisions down can learn the same way.
Source: Noval OS, read 30 October 2026
Building with AI
Fix the machine that made the mistake.
What we tested. We built a series of games with AI tools and kept a record of every fix.
What we found. Fixing each wrong screen by hand cost days and the same mistake came back. Fixing the tool that produced it, the layout builder or the art instructions, fixed every copy at once.
What we do now: when AI or a person gets something wrong in a client's business, we fix the process or the instructions that produced it, not just the one bad result.
Noval playbook, 25 October 2026. See the games.