Model Drift

Model drift is when an artificial intelligence’s (AI’s) performance declines over time, “drifting” from its original intended purpose.

What Is Model Drift?

Most AI solutions are inspired by the human brain, and as such, they can similarly decline in performance over time. Model drift describes exactly that: when an AI’s predictive ability loses its accuracy or power over a given period of time. Model drift occurs because the data that informs its predictive ability is no longer working to help the AI achieve its original goal. For example, AI solutions trained on small amounts of data may observe gaps or nuances with the data. They may try to fill in those gaps by assuming or hallucinating information, feeding incorrect data back into the inference pool and, therefore, diminishing its quality. The problem can also occur if a brand using a large language model (LLM) shifts its goal but doesn’t update the inputs used to train the AI. The inputs won’t match the intended output, rendering the AI insufficient.

See Salsify in Action

Bring your team, tools, and data together to launch high-impact inbound marketing campaigns in less time.

Model Drift

What Is Model Drift?

Related Terms

Related Resources

The AI Path to Purchase Continues: Executing Your AEO & GEO Strategy with Salsify and DetailPage

AI Agents Assist Humans to Enhance Digital Commerce Performance

See Salsify in Action