Part 2 of 3 — The AI ROI Series
In Part 1, we looked at the conditions where AI genuinely earns its keep. High volume, consistent tasks, low error stakes, and labor costs that clearly exceed tool costs. When those green flags line up, AI can be transformative.
But not all processes present a perfect green flag scenario. I’ve talked with clients that feel pressured to use AI; from owners, from peers, from employees.
That pressure is where red flags get missed. And missed red flags are expensive. So, let’s talk about the 4 red flags that may muddy the waters.
Low Volume
Fixed AI tool costs don’t care how much you use the tool. The subscription fees are often the same whether the process runs 500 times a month or 15 times a month.
At low volume, that fixed cost never amortizes across enough instances to generate meaningful savings. You end up paying more for the AI solution than you were paying for the human doing the job, sometimes significantly more, once you factor in setup and management costs.
Volume isn’t everything. But without enough of it, nothing else matters.
High Variability
Let’s say AI handles the predictable 80% of a task flawlessly. What about the other 20%? What about the exceptions, the edge cases, and the situations that require judgment.
High variability creates two costs that rarely show up in vendor projections. First, you need more human-in-the-loop oversight, because you can’t fully trust AI output when inputs are unpredictable. That oversight eats into your savings rate. Second, the AI will need more ongoing management; retraining, prompt adjustments, validation as new scenarios emerge. AI is not set it and forget it.
The more a process relies on what’s in an experienced person’s head rather than a documented set of rules, the harder it is to automate reliably. Which brings us to the most important lesson I’ve learned in thirty years of working with business automation:
AI gives you more flexibility than traditional automation — you don’t need a rule for every edge case. But if the process isn’t documented, standardized, and understood, you will have the same problem you’ve always had. You’ll just have it faster, and it will cost more.
High Stakes Errors
Ask yourself one question: if the AI gets this wrong, what does it cost? It may not be able cost effective to quantify, but you should make the a judgement call, especially if the ROI is low.
For a customer service email, the answer might be “minor inconvenience, easily corrected.” For a sales quote, a compliance document, or anything that touches a customer contract, the answer is very different.
High error stakes don’t automatically disqualify AI. But they do require more human review of every output. More review means lower effective savings rate. Lower savings rate means worse ROI. And if the review process is extensive enough, you haven’t automated the task, you’ve added a step to it.
Costs
You may be using ChatGPT for free, but the free version will not support automation. Getting AI up and running also costs real money. Integration work, prompt engineering and configuration will take time. Addition costs may include staff training, parallel testing and ongoing education.
All that upfront investment must be recovered before the project has a positive ROI. If the monthly savings are too thin because volume is low, the savings rate is modest, or the tool costs are high that recovery point either arrives very late or never arrives at all.
Red Flags Don’t Always Mean Stop
Identifying red flags isn’t about killing the AI project. It’s about going in with realistic expectations, and sometimes, about doing the work to eliminate the red flags before flipping the switch.
A high-variability process with poorly defined exceptions isn’t necessarily a bad AI candidate forever. It might be a process that needs to be standardized first. Get the documentation right, reduce the variability, define the edge cases, and suddenly the ROI picture looks completely different.
Sometimes the best AI advice isn’t “use AI.” It’s “let’s get this process ready for AI.”
Next up — Post 3: Finish Line? How to Know Before You Commit