news.today
Hello · Register today | LoginHello · My Account[—] [32]
Business

AI performance depends on data quality, not just model cost

Summarised from 2 outlets · 4 reports · Updated 21 Oct, 11:44 · Archive

Technology leaders often blame AI models for poor results, but outdated data feeding those systems is usually the real problem.

Independent Business reports that technology leaders typically blame AI failures on the models themselves and focus cost reviews on switching to cheaper alternatives and reducing token usage. However, the real issue usually lies with the data feeding those systems, which receives little scrutiny during audits because its cost does not appear as an obvious line item.

A smaller, cheaper model running on current data will usually outperform a larger, expensive model running on stale data. The problem is that outdated data does not announce itself the way token counts do, so it survives budget reviews that would eliminate a slow, expensive model. For example, a pricing agent built to reprice against competitors each morning may work well initially, then gradually start making decisions based on saved snapshots of competitor pages rather than live sites, producing outputs that appear correct but rely on information that is no longer current.

The cost of cutting data spend does not materialise immediately. It typically emerges three months later in degraded outputs, by which time no one traces the decline back to the decisions made during the cost review.

AI performance depends on data quality, not just model cost
Image: TechRadar

Comments (0)

How it is being reported

In this story: Independent Business · Decodo · NHS · UK

This summary was written by AI from the headlines and standfirsts above, and states as fact only what at least two outlets report. How we use AI · Report a problem