When to Stop an AI Tool

By Paige Gilmore, Founder, NetLift · Published 2026-07-28 · Updated 2026-08-07

An AI tool should be stopped when its current net value—the labour value of realised time saved minus all costs—is consistently negative and the evidence for future value is weak. Decisions are based on deterministic value models and evidence quality grades rather than sentiment.

Stop an AI tool when the total cost of ownership exceeds the labour value of the time it saves. This decision should be driven by a deterministic value model that subtracts licences, usage, training, and rework from the realised time savings multiplied by the loaded hourly staff cost.

Finance and IT leaders use five decision states to manage AI portfolios: Expand, Continue, Review, Improve, or Stop. A tool enters the 'Stop' state when net value is negative, payback periods are indefinite, and the evidence quality for potential improvement is low.

How do you calculate the point of failure for AI spend?

An AI tool is a financial liability if the labour value of time saved is lower than the full cost of the tool. To find this, calculate the time the work would take without AI and subtract the time taken with AI. Multiply those saved hours by the loaded hourly cost (NetLift uses a default of $75 per hour unless specified). If this figure, after subtracting all costs, remains negative without a clear path to future volume increases, the tool should be decommissioned.

What costs are often missed in the 'Stop' decision?

Many organisations only look at licence fees, but the 'Stop' decision requires a view of the full AI cost. This includes implementation, training, and usage fees. Crucially, it must also include the cost of review and rework where tracked. If an AI saves a developer two hours but requires three hours of manual review and debugging, the net value is negative, and the tool is an immediate candidate for the 'Stop' or 'Review' state.

Why is evidence quality critical for decommissioning?

Decisions to stop a tool should not rely on anecdotal feedback. Evidence quality grades range from 'Estimate Only' to 'Verified.' Objective baselines, such as historical data or cohort comparisons, carry more weight than self-estimates. If a tool shows a high cost but the only evidence for its value is a small sample of self-reported 'estimates,' the risk of waste is high. High-quality, verified data provides the confidence needed to cut spend without affecting actual output.

How does future value impact the decision?

While current net value looks at realised savings, future value accounts for expected recurring savings and projected volume. A tool might have a negative net value during its first month due to high implementation and training costs. However, if the payback period—the time it takes for net value to cover the total cost to date—is short and the future value is high, you would 'Improve' or 'Continue' rather than 'Stop.'

NetLift provides the deterministic framework to distinguish between useful tools and sunk costs. By measuring the labour value of realised time saved against a baseline, NetLift assigns each tool a decision state like 'Stop' or 'Expand' based on objective evidence quality. This is done by measuring work and value, never through employee surveillance or keystroke logging.

Frequently asked questions