Cheaper AI can still raise your bill. Here’s how to check

Lower AI prices do not guarantee lower bills: businesses may spend more by using AI more often, or by paying people to correct cheap but unreliable results. Lindsey Witmer Collins, founder of AI company WLCM.AI, offers a practical test for deciding whether the work is worth its cost.

Watch Desk analysis

What happened

In a Forbes Technology Council article, Collins says businesses should count more than the price per token, the chunks of text a model processes. Include repeat attempts and the staff time needed to check or fix the result, then compare that total with the cost of doing the work without AI. A more expensive model that gets a task right first time could cost less overall than a cheaper one that needs correction.

She also advises testing different models and reasoning-effort settings on real tasks. Routine extraction or formatting may work at a lower setting, while harder analysis may need more effort, which can mean more billed tokens even if the final answer is short.

Why it matters

A falling unit price can encourage much heavier use. Collins points to the Jevons paradox: if a task gets cheap enough to run many more times, total spending can rise. Her example is simple: cutting a task’s cost from $1 to 50p but increasing its volume from 100 runs to 1,000 takes spending from $100 to $500.

She also flags a less visible cost: the instructions, documents and conversation history sent with each request. Long threads can carry stale material into new work, so a fresh conversation with a short briefing may be cheaper. For automated tasks, she recommends capping retries and checking that the results are actually used and improve decisions.

Our read

This is useful advice because it shifts the question from “Which model has the lowest rate?” to “What does this job cost when it is finished?” That means including human review, repeated attempts and how often the task runs, not just the price on a model menu.

Collins is offering practitioner guidance, not a universal cost audit. Still, the next time someone celebrates cheaper tokens, ask whether the organisation’s total bill fell too. The answer may be hiding in the retries, the review queue or that very long chat nobody had the heart to close.

What to watch

• Whether teams compare total task costs, including review and correction time.
• Whether lower model prices lead to more frequent or automated use.
• Whether businesses set retry limits and trim unnecessary context.

Discussion spark: When assessing AI’s value at work, should businesses prioritise the lowest cost per task or the best result after human checking and repeat attempts?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.