Turning call notes and a rate card into a first-pass proposal inside the real Word template

The challenge

Putting together a client proposal meant a rep manually drafting fee tables, payment milestones and a consultant roster from scratch, or copy-pasting them from whatever similar proposal existed previously. Rate cards and current consultant availability largely lived in people's heads rather than in the document itself.

Case knowledge needed for a proposal, call notes, briefs and scanned documents, was scattered across files, with a rep typically having to re-read a transcript before every draft rather than working from a single reference brief.

The tool that solved this had shipped quickly on a pragmatic stack, using an external database provider and a mix of cloud and office-based hosting to get it into reps' hands fast. Once it became part of daily workflow, that setup sat outside the Microsoft tenant's compliance boundary, with an external vendor and a machine outside IT's direct control handling case data.

The business needed the tool brought fully inside its Microsoft and Azure environment, with proper single sign-on and centralised access control, backup and audit, without losing the drafting quality that came from the underlying AI models.

Our approach

1. 01

   Building a conversational drafting assistant

   We built a conversational agent reps could talk to like a colleague, taking a client job request, tone and brand guidance, and call-transcript notes, and turning them into a first-pass proposal drafted directly into the client's real Word template rather than a generic summary.

2. 02

   Connecting fee tables and rosters to live data

   We connected the assistant to the current rate card and consultant list, so every draft's fee tables, payment milestones and 'Meet Our Experts' roster reflect live figures and availability, rather than whatever was copy-pasted from the last similar proposal.

3. 03

   Synthesising case knowledge automatically

   We built the assistant to accept whatever a rep already had on a case, transcripts, briefs, even scanned PDFs and images, and fold them automatically into a single reference brief, so the case's context is always on hand rather than requiring a rep to re-read source material before drafting.

4. 04

   Shipping fast on a pragmatic initial stack

   To get the tool into reps' hands quickly, we launched on a pragmatic stack outside the core Microsoft tenant, validating the workflow and drafting quality in daily use before committing to a larger migration.

5. 05

   Migrating fully inside the Microsoft and Azure tenant

   With the tool proven and mandated for migration by IT, we moved the system onto Azure SQL and Azure App Service, added Entra ID single sign-on, and replaced direct calls to the AI provider's API with Microsoft Foundry, which hosts the same underlying Claude models natively inside Azure, so drafting quality was preserved while access control, backup and audit moved under the same identity system as every other internal tool.

The outcome

What used to be a manual, multi-round drafting and formatting job is now one guided conversation, with the assistant producing a first-pass proposal in the client's real Word template rather than reps building fee tables and rosters by hand.

Every draft now pulls from the same live rate card and current consultant list, removing the reliance on rate and roster information living in people's heads or being copy-pasted from older proposals.

With the migration into Azure SQL, Azure App Service, Entra ID and Microsoft Foundry complete, the external vendor and office-based hosting setup have been retired, bringing access control, backup and audit under the same tenant identity as the rest of the business's systems, without giving up the Claude-powered drafting quality reps had come to rely on.