All Case Studies
Manufacturing & Industrial United Kingdom

Building an AI Sales Agent That Could Prepare Quotes Without Setting Its Own Prices

Our client is a UK-based industrial manufacturer handling a high volume of quote requests from distributors and business customers. Preparing each quote involved checking customer records, product availability, pricing and discount rules across multiple systems.

As quote volumes grew, sales representatives were spending too much time on these repetitive tasks. The company wanted to introduce AI into the process without giving it authority over pricing or commercial decisions.

Brainium built an AI sales agent that could interpret quote requests, retrieve relevant information, prepare quotes using existing pricing services, and route exceptions for human approval. Pricing rules and approval controls remained within the underlying business systems.

IndustryManufacturing & Industrial
CountryUnited Kingdom (UK)
Tech StackPython, FastAPI, OpenAI API, PostgreSQL, AWS
Key Result~60% Faster Quote Preparation
01

The Challenges We Faced Along the Way

The challenge was giving the agent access to the information and systems needed to prepare a quote without changing the rules behind it.

Fragmented Sales Data

Customer details, product information, stock levels and commercial terms lived across the CRM, ERP and product systems. The agent needed to retrieve consistent information from each without duplicating it.

Complex Product Requirements

Some products required specific configurations, quantities or supporting information before they could be quoted. The agent needed to recognise when a request was incomplete rather than filling gaps with assumptions.

Pricing Had to Stay Deterministic

Customer-specific prices, volume discounts and negotiated terms were already handled by the company's pricing service. The agent could request a calculation, but it could not calculate or override the price itself.

Tool Access Needed Clear Boundaries

The agent required read access to several systems and controlled write access for quote creation. Actions such as changing customer terms or modifying pricing remained outside its permissions.

Exceptions Could Not Be Treated as Errors

Requests outside the standard quoting rules needed to reach a salesperson with the relevant context intact. The workflow therefore had to support escalation rather than forcing every request through an automated path.

02

Our Primary Approach for This Project

Brainium designed the agent around the client's existing sales systems, keeping pricing and commercial rules outside the AI layer.

Mapping the Quote Workflow

We mapped the existing process from the initial customer request through product selection, pricing, approval and quote creation. This helped define which steps the agent could handle and where existing systems or sales staff needed to remain involved.

Connecting the Required Systems

The agent was connected to the CRM, product catalogue, availability data and pricing service through defined APIs. Each integration exposed specific actions and returned structured information for the next step in the workflow.

Keeping Pricing Deterministic

The existing pricing service remained the source of truth for customer-specific prices, discounts and commercial terms. The agent could request a calculation using the required inputs, but it could not calculate, alter or override the resulting price.

Defining Agent Permissions

The agent could retrieve customer and product information and create a draft quote. It could not change customer terms, modify pricing or approve exceptions. These permissions were enforced at the application and API level rather than through instructions to the model.

Building the Approval Path

When a quote fell outside the standard rules, the agent stopped the automated workflow and passed the relevant details to the appropriate salesperson or approver. This allowed routine quotes to move faster without removing human control from non-standard decisions.

03

Our Primary Solution

Brainium built an AI sales agent around the client's existing CRM, product and pricing systems. The agent handled interpretation and coordination, while existing services remained responsible for business rules and commercial decisions.

Python FastAPI OpenAI API PostgreSQL AWS
Structured Quote Requests

The agent converted natural-language requests into structured quote data, including customer, product, quantity and configuration details. Missing or ambiguous information was flagged before processing continued.

Controlled System Access

We exposed dedicated tools for customer lookup, product validation, availability and pricing. The agent could access only the functions required for the quoting workflow, keeping the underlying systems as the source of truth.

Deterministic Pricing

The agent sent validated quote details to the existing pricing service and used its response to prepare the quote. It had no ability to modify the returned price or apply unauthorised discounts.

Quote States and Approvals

Standard quotes could be prepared automatically, while predefined exceptions moved into a pending-approval state. The agent could not approve its own exceptions or bypass the existing commercial workflow.

Traceable Execution

Tool calls, validation results and approval actions were recorded against the quote, giving the sales team a clear record of how each quote was prepared.

04

The Results

The new quoting workflow reduced repetitive sales work while giving the client a controlled way to introduce AI into an existing commercial process.

~60% Faster Quote Preparation

Standard quote preparation time fell by around 60%, as the agent handled routine customer, product and availability lookups.

Consistent Pricing

Quotes continued to use the client's existing pricing service, keeping calculations and commercial rules outside the AI layer.

Controlled Automation

Routine requests could progress automatically, while non-standard quotes were routed to sales staff for review.

Traceable Agent Actions

Tool calls, validation results and approval events were recorded against each quote, providing a clear record of the workflow.

Designing AI Agents Around Real Business Rules

Brainium helped the client introduce AI into its quoting process without moving core commercial decisions into the AI layer. By connecting the agent to existing customer, product and pricing systems, routine quote preparation could be automated while pricing rules, permissions and approvals remained under application control.

The result was a faster quoting workflow that gave the sales team more room to focus on customers and exceptions, without treating the AI agent as the authority over the transaction.

Building an AI Agent for a Real Business Workflow?

Talk to Brainium about designing agentic solutions that work with your existing systems and business rules.

Leaving Already?
Let us help you find the right services for your business!

Our expert will help you in:

  • the right solution for your business
  • A ballpark estimate
  • An estimated delivery time

Start the Conversation!

Reach Out to Our Team