When AI Costs Drop, Which Tasks Become Worth Automating?
A cheaper AI query changes the math on low-value tasks. Here is how to find which ones just crossed the line into profitable.
The Task You Shelved Six Months Ago
A year or two ago, someone on your team ran the numbers on automating something small: sorting inbound emails, drafting first replies to support tickets, tagging invoices, summarizing call notes. The idea got tested, the cost per request came back higher than expected, and the project got quietly shelved. Not because the task didn't matter, but because the math didn't work. The time saved was real but modest, and the cost of running an AI model against thousands of low-value requests every month didn't clear the bar.
That shelved project is worth reopening. Not because AI got smarter in some abstract sense, but because the price of running it dropped. When the cost per query falls, tasks that used to sit below the profitability line move above it. The problem is that most businesses don't have a system for noticing when that shift happens. They evaluated once, filed the result, and moved on. Meanwhile the underlying economics keep changing every few months, and nobody is checking back.
Why Cost Per Query Is the Number That Matters
Most conversations about AI in a business focus on capability: can it write, can it summarize, can it answer a customer question correctly. Capability matters, but it's not what decides whether a task is worth automating. What decides it is a simple ratio: the value of the task divided by the cost of doing it with AI, compared to the cost of doing it by hand.
A task that saves an employee three minutes but costs eight cents per run at high volume can still be worth automating if it happens ten thousand times a month. A task that saves an hour but only happens twice a week might not be, even at the same per-run cost. The volume and the value per instance both matter, and they don't move in lockstep with the price of the model. This is why a blanket decision — "we looked at AI and it wasn't worth it" — ages badly. It was a decision made against one price point, for one set of tasks, at one moment.
How to Find the Tasks That Just Became Worth Automating
The fix isn't to re-run a full AI strategy review every quarter. It's to keep a short list of operations you've already ruled out, and revisit the arithmetic on a fixed schedule rather than by memory or gut feeling.
- List every repetitive task you evaluated for automation and rejected on cost grounds — include the volume per month and the estimated cost per run at the time.
- Recalculate cost per run using current pricing for the model tier you'd realistically use, not the flagship tier — cheaper mid-range models often cover routine tasks like triage, tagging, and drafting just fine.
- Multiply the new cost per run by monthly volume and compare it against the fully loaded cost of the manual alternative, including error correction and management time, not just the hourly wage.
- Set a threshold before you start: if the automated cost falls below a fixed percentage of the manual cost (many teams use somewhere around half), it moves to a pilot.
- Pilot on a slice of real volume for two to four weeks before committing, and track cost per run and error rate side by side, not capability in isolation.
This approach keeps the decision anchored to numbers you already have, rather than requiring a new evaluation from scratch every time a new model gets announced.
What Changes Once the Threshold Moves
When a task crosses into profitable territory, the change is usually not dramatic on its own. One ticket triage workflow, one invoice tagging step, one first-draft reply — none of these transform a business by themselves. What changes is the cumulative effect across a dozen small tasks that were each individually too marginal to justify the effort of setting them up. As the cost per query keeps falling, more of these small tasks clear the bar at the same time, and the operational load they take off your team compounds.
The risk on the other side is treating every price drop as a reason to automate more. Cheaper doesn't mean free, and a task with genuinely low volume or low value still isn't worth the setup and monitoring cost, no matter how cheap the model gets. The discipline here is the same in both directions: run the arithmetic, don't run on instinct.
What to Watch to Know It's Working
Once a task is automated, track two numbers monthly: the actual cost per run against your projection, and the error or correction rate against the manual baseline. If actual cost creeps above what justified the automation, or errors push work back onto your team, the task hasn't earned its place yet — pull it back and reassess rather than letting it run on faith.
Reassess Your Shelved Automation Projects
ArkonLabs builds and measures AI workflows for PMEs — pricing them against real task volume, not against a demo. If you have a project that was ruled out on cost a year ago, get in touch through www.arkon-labs.com and we'll help you re-run the numbers.