UpTrajectory Review

Enrique Dans, writing for Fast Company, makes a point that most small business owners will recognize from their own gut: your AI might be busy without being useful. The piece argues that companies experimenting with AI are stuck measuring the wrong things — how fast it responds, how cheaply it runs, whether it completes the assigned task — when what actually matters is whether the business improved. Revenue, margin, retention, customer satisfaction. The distinction he draws is between outputs (what the AI produced) and outcomes (what changed for the business). It is a simple framing with significant implications, and it arrives at a moment when many operators are being sold AI tools with promises that sound impressive but resist accountability.

For a small business owner, this is not an abstract management theory. If you have deployed or are considering an AI tool — a chatbot for customer service, an automated email responder, a scheduling assistant — you are almost certainly being shown metrics like response time, resolution rate, or cost per interaction. Those numbers feel reassuring. But Dans's point is that an AI agent could be hitting every one of those targets while quietly damaging customer relationships, eroding margins through unnecessary discounts, or churning through your client base with efficient but impersonal service. The metrics that vendors default to are the ones easiest to measure, not the ones most tied to your bottom line.

What is genuinely useful here is the reframe from 'is the AI smart enough' to 'how does the AI know whether what it did actually improved the business.' That second question forces specificity. If you deploy an AI tool to handle customer inquiries, the real question is not whether it answered correctly — it is whether customer retention improved, whether repeat purchase rates held, whether complaint volume dropped. Dans cites OpenAI's enterprise framework of specify, measure, improve as evidence that even the largest AI labs are now recognizing this gap. We agree with the core argument and think it is underappreciated in the small business market, where AI purchasing decisions are often made on demos and promises rather than defined business outcomes.

The second-order effect worth noting is that this shift raises the bar for what it means to deploy AI responsibly. If you accept that outcomes matter more than outputs, then every AI deployment becomes an experiment that requires baseline measurement before you start. You need to know your current churn rate, your current average response time, your current customer satisfaction scores — before the AI enters the picture — or you will have no way to evaluate whether it helped. This is a real cost in time and attention that many small operators skip. It also means that some AI tools, once held to outcome-based scrutiny, may not justify their subscription price. That is an uncomfortable conclusion for vendors and for owners who have already invested.

The practical move is straightforward: before your next AI purchase or renewal, write down the one business metric you want this tool to move. Not a technical metric — a business one. Then measure where you stand today. Set a review date ninety days out and hold the tool accountable. If the vendor cannot tell you how their product connects to that metric, that is a signal. Dans's piece does not offer a template for doing this, and the full article likely goes deeper into the measurement frameworks he references. But the core discipline — specify, measure, improve — is something any operator can start applying this week without a consultant or a new software purchase.

“An agent could be completing its task perfectly as instructed, and still hurt the company by doing so.” — Fast Company

Takeaway: Before buying or renewing any AI tool, define the single business metric it should move, record your baseline, and review results in ninety days.