GLOSSARY
Training
Training adjusts an AI model’s parameters using data and a learning objective. The patterns it acquires depend on the examples, target outcome and training method. Attaching a document to an individual request does not normally change those parameters.
Deciding what sales records should teach
A business might use previous proposals to improve particular response formats or classifications. Those records also contain outdated prices, individual working habits and different levels of negotiating authority. Treating every successful deal as a good example can distort the learning objective: unusually low prices or intensive support may explain the outcome.
Data rights, reliable labels and the desired behaviour therefore need clarification before training begins. Current delivery conditions can often be supplied more directly through retrieval-augmented generation. Training may be appropriate when enough suitable examples exist to improve a recurring behaviour.
Work continues after the initial run
The economic commitment includes selection, cleaning, independent assessment and subsequent updates. Better performance on familiar examples is insufficient evidence. Evaluation needs previously unused cases and a comparison with the existing process. Changes to products or customer groups can affect suitability. Responsible teams should therefore document the model version, assessment and approval, with a clear route to renewed validation when conditions change.