A water utility wants form OCR enrichment before an agent answers billing questions from scanned statements. Where do Foundry Tools fit in that plan?

A water utility that wants form OCR enrichment before an agent answers billing questions from scanned statements should place Foundry Tools OCR enrichment upstream in the pipeline. Scanned statements are images until text and fields are extracted; OCR converts those pages into content agents and retrieval can use. Planning Tools for extraction before generation is the correct sequencing for civic document enrichment. Placing Foundry Tools OCR enrichment in the pipeline so extracted fields build the evaluation set after answers are generated fails because enrichment must feed extracted fields to the agent before answers are generated, not assemble test data afterward, so the current billing questions would still be answered without grounded content. Running Foundry Tools OCR enrichment in the pipeline so extracted fields update the search index for future billing queries fails because indexing for future queries does not feed the current extracted meter and amount fields into the agent before generation, so the immediate billing answers would still lack grounded searchable content. Placing Foundry Tools OCR enrichment in the pipeline so extracted fields fine-tune the base model on billing forms fails because enrichment is meant to feed extracted fields to the agent for grounded answers, not to fine-tune base model weights, which would not make the current statements usable for generation. Exam caveat: OCR quality, form templates, and human review thresholds for low-confidence fields matter as much as 'using Tools.' Operational check: run sample scanned statements through OCR, validate extracted meter and amount fields against ground truth, then confirm the agent answers only from extracted/searchable content—not from hallucinated readings.