Quote Agent

The bottleneck after a site visit.

At Paul Kick, preparing renovation quotes took up a substantial share of the office team’s time. A specialist visited the property and collected measurements, plans, notes, the scope of work, and customer preferences. The office then had to interpret that information, identify materials, check suppliers, and prepare the quote before the job could move forward.

I built this workflow into a broader cloud-based operations platform connecting field and office teams. The goal was to reduce the manual effort between a site assessment and a quote ready for the customer.

Understanding the decisions before automating them.

I started by studying how the office team prepared quotes manually: what information they needed, which steps followed repeatable rules, and where experience or clarification changed the answer. I mapped the decision points and the consequences of getting them wrong to decide how much autonomy each stage should have.

The recurring work was a good fit for automation, but material suitability, purchasing trade-offs, and the final customer offer still needed judgment. I split those responsibilities across requirements analysis, supplier comparison, and quote drafting.

How the quote took shape.

LangGraph coordinated the agents. Customer and site information was normalized first. A requirements agent then worked out material quantities, specifications, tools, and associated components, using the company’s internal knowledge base and additional web research to check the selections.

A sourcing agent compared purchasing options. After a person reviewed and corrected the proposal, the approved selection formed the basket and fed a separate agent that drafted the quote using the company template. The completed quote passed through a final human check before reaching the customer.

The quote included the cost of work, markup, and taxes as well as materials. The agent used the company knowledge base to calculate these components, with code-based sum checks and human review before release.

Working with suppliers without APIs.

The company worked with around five major suppliers. Current catalogs and availability were accessible through their account portals, without an API for the workflow we needed. I used web parsing to make that supplier information available to the sourcing process.

The agent compared candidate items by price, availability, warehouse location, delivery cost, and lead time. It also checked internal inventory and opportunities to consolidate orders. A predefined decision matrix guided the purchasing recommendations.

This made the basket a purchasing decision with several constraints. A lower item price could be offset by delivery charges, fragmented orders, or a delivery date that did not fit the work. Comparing those factors together was central to the design.

Checking the recommendation.

I built in code-based recalculation of sums and checks on units and packaging. These checks addressed a practical failure mode in supplier comparison: a plausible product match can still produce an incorrect estimate if the quantity, unit, or pack size is interpreted incorrectly.

The decision matrix guided the trade-offs between price, availability, delivery, and order consolidation. Arithmetic checks covered the numbers; human review covered the suitability of the purchasing proposal and the commercial offer.

More capacity to prepare quotes.

The system substantially reduced the office effort spent assembling quotes. Staff could review a prepared proposal instead of repeating the same requirements research, supplier checks, and document drafting for every job.

Reducing that workload creates room to handle more inquiries with the same team and respond to customers sooner. Earlier clarity on materials and delivery options also supports purchasing and scheduling once a job is approved.