The business was producing complex operational quotes and compliance documents in a category where quality depends heavily on historical context, repeatable rules, and clear document structure.
One of the most persistent problems was inconsistency. Different staff members would approach the same job type and arrive at quotes that varied significantly in scope, structure, and pricing assumptions. There was no single source of truth to anchor the quoting process — experienced team members carried the institutional knowledge in their heads, and that knowledge was not transferable in any reliable way.
New staff had no reference point. Without access to prior work, they were effectively starting from scratch on every quote, making decisions based on incomplete context and producing outputs that looked nothing like what the business had delivered before. The cost of a wrong assumption was not just a rework cycle — it was a compliance risk and a margin risk on every affected job.
Generic AI output was not enough. The workflow needed to reference prior jobs, reflect domain-specific constraints, and generate outputs that looked consistent with how the business already operated. The challenge was to combine retrieval, structured form inputs, and document rules into a process that was useful in production rather than just interesting in demos.