Automate Your Marketing Agency's Proposal Workflow

Many agencies rely on tools like PandaDoc or Proposify. These are template libraries, not generative systems. They store content blocks, but an account executive still manually selects which services to include. The software cannot read discovery call notes and reason about which service package is the best fit for a new lead. An AE spends 30 minutes dragging and dropping the same five service blocks for every new web design client.

Others use the quoting feature in their CRM, like HubSpot Quotes. This tool is a line-item editor designed for product sales, not custom services. It can pull a price for "SEO - Gold Package" but cannot generate a scope description that mentions the client's specific competitors or business goals. The output is static and requires heavy manual editing to feel personalized.

Consider a 15-person agency. A new lead's discovery call notes are logged in HubSpot. The AE opens Proposify and spends the next 90 minutes toggling between windows, copy-pasting notes, rewriting service descriptions to use the client's vocabulary, and manually calculating pricing. This low-value work happens 5-10 times per week, delaying sales velocity and introducing a high risk of copy-paste errors.

The structural problem is that these tools are document assemblers that manage content, not context. They lack a reasoning layer, like an LLM, to interpret unstructured client needs from call notes and map them to a structured service catalog. Their architecture is built for static templates, preventing true, context-aware automation.

We would begin with a thorough audit of your service catalog and 10-15 past proposals you consider high-quality. The goal is to map every service, its pricing logic, and its standard scope language into a structured format. We've built document processing pipelines for financial services using the Claude API, and the same pattern of extracting structured data from text applies directly to your sales process.

The core of the system would be a FastAPI service powered by the Claude 3 Sonnet API. The service would ingest unstructured text, like call notes, along with a client's HubSpot ID. The LLM then identifies the client's specific needs, matches them to the correct services in a Supabase database, and generates a complete proposal draft. Pydantic schemas would validate the output to ensure every SOW has the required sections before it is finalized. A single proposal generation would complete in under 60 seconds.

The final deliverable is a secure, internal API. Your team could access it through a simple web form or we could build a custom action inside your HubSpot workflows. The generated draft can be pushed to Google Docs or back into PandaDoc, ready for a 5-minute human review. The entire system would run on AWS Lambda, typically costing less than $50 per month to operate.