GLOSSARY
Multimodal
Multimodal AI systems process or generate several forms of information, such as text, images or audio. The supported combinations depend on the model and application. Additional input formats expand the available evidence and introduce additional places where errors can occur.
Combining customer statements and documents
A proposal might depend on a meeting record, a photographed sketch and a pricing table. Multimodal processing can help bring the requirements together. The result should still identify whether a quantity was spoken, read from a table or inferred from an image.
A misread figure can change the price calculation and delivery scope. An unclear speaker attribution can turn a question into an apparent commitment. The organisational task is therefore to trace consequential statements to their original location and flag conflicts for clarification. Someone must remain responsible for resolving those conflicts before the offer is issued.
Test each information format
Acceptance cases should include poor recordings, ambiguous tables and contradictory statements. Measure missed requirements and human review effort alongside speed. Recording and processing conversations require appropriate permission and data protection arrangements. Across all formats, the common quality criterion is preserving meaning, even when the presentation looks convincing.