Skip to content
Brandon Burkman
Back to work

Beta Systems Software · Case study

AI-Assisted Lead Routing

Cutting speed-to-lead from more than four hours to under 60 minutes.

My role
Designed the routing framework and refined the AI prompt and workflow logic.
Focus
Marketing Operations · AI · Automation · Lead Management
Systems
HubSpot · ChatGPT · CRM workflow automation
Speed-to-lead
<60 minPreviously more than four hours
Lead assignments automated
~90%Approximate share of assignments
Estimated routing accuracy
85–90%

Problem

Inbound lead routing was heavily manual. When someone submitted a form, a digital marketing specialist in Germany reviewed the inquiry, determined which of roughly six business units it belonged to, and identified the appropriate sales owner.

That process introduced unnecessary delay. Average response time was more than four hours, and some inquiries could wait considerably longer before reaching Sales.

Many inquiries were ambiguous. Determining intent sometimes required understanding what the prospect wrote, what content they had engaged with, and the context surrounding the submission.

Context

At the same time, we were building the North American marketing operations infrastructure in HubSpot. That created an opportunity to rethink routing as a system.

The goal was to automate decisions that could be made reliably, use AI where interpretation was useful, and retain a human fallback when confidence was low.

Approach

I designed a routing framework that evaluated multiple signals:

  • Form fields
  • Website and page context
  • Submission context
  • Known business-unit indicators
  • Free-text comments from the prospect

Deterministic rules handled straightforward assignments first. When those signals were not enough to confidently determine intent, ChatGPT served as a classification layer to interpret the prospect’s comments and help determine the appropriate business unit.

I refined the AI prompt and routing logic over roughly two weeks, testing how different types of inquiries were categorized and adjusting the instructions where classifications were inconsistent.

Unclear or ownerless records were intentionally sent to a manual fallback rather than forcing an automated decision.

System

The workflow combined known routing signals, business-unit logic, AI classification when needed, sales-owner assignment, and CRM tasks and notifications. Manual review handled exceptions.

The key decision was when to use each layer. Clear signals could bypass AI; ambiguous language could benefit from classification; unresolved intent or missing ownership required a person.

From inquiry to sales handoff

  1. Capture the inquiry

    Inbound form submission brings together form fields, page context, business-unit indicators, and the prospect’s comments.

  2. Choose a routing path

    Evaluate known signals before asking AI to interpret free text.

    • Clear business-unit signals

      Apply rules

      Use known indicators to identify the business unit and sales owner.

    • Ambiguous intent

      Classify with AI

      ChatGPT interprets comments. A usable classification proceeds to ownership; unresolved cases go to review.

    • Unclear or ownerless

      Human review

      Resolve the exception manually before completing the handoff.

  3. Assign a confirmed owner

    Complete sales-owner assignment after the rules, classification, or manual review resolve the inquiry.

  4. Complete the CRM handoff

    Create the task and notification so Sales can act on the inquiry.

A simplified view of the routing logic. The three paths are alternatives: AI is used only when known signals do not resolve intent. Unclear or ownerless records go to human review.

Outcome

Speed-to-lead dropped from more than four hours to under 60 minutes. The workflow automated approximately 90% of lead assignments while maintaining an estimated 85–90% routing accuracy.

Sales received inbound inquiries substantially faster, while Marketing spent less time manually interpreting and assigning submissions.

The workflow demonstrated that AI could be integrated into an existing marketing operations process while preserving structured rules and human exception handling.

Lessons

AI was most useful for the part of the workflow traditional automation struggled with: interpreting ambiguous human language. Structured rules remained better for predictable decisions, and human review remained valuable for exceptions.

Rules where rules work. AI where interpretation helps. Humans where judgment is still required.

Get in touch