Conversational Design•Mar 11, 2024

Saving ~2,800 Man Hours for Inside Sales

How we analyzed over 36,000 live chat transcripts to build a qualification chatbot framework, filtering out low-quality queries.

Simplilearn sales chatbot — pre-qualification nodes, instant PDF fulfillment, and human handoff only for high-intent leads.

Problem

The Simplilearn landing page receives roughly 35,000+ chats every month from users worldwide. While 40% convert into viable leads, the remaining 60% are non-qualified chats and drop-offs. Over 21,400 non-lead conversations consumed more than 2,800 consultant hours monthly, pulling advisors away from high-intent learners.

"More than 56% of total chat initiations resulted in non-leads. Over 21,400 non-lead conversations consumed more than 2,800 man hours monthly."

My Role

  • Conversation Designer — led transcript audit, competitor benchmarking, and chatbot flow design in Figma.
  • Manually classified a 10% sample (3,716 transcripts) after automated NLP parsing proved insufficient.
  • Defined rollout phases with inside sales and regional ops teams.

Constraints

  • Live chat platform: Flow changes had to work within the existing sales chat vendor — no custom backend.
  • Regional rollout: Phase 1 limited to specific Indian regions before nationwide expansion.
  • Data quality: Colloquial, messy chat logs resisted automated NLP — manual audit was required for reliable insight.

Process

I aggregated 36,000+ raw chat logs from September and manually audited a representative 10% sample (3,716 transcripts), classifying each as High Quality or Low Quality:

  • High Quality: Users willing to share contact info and upskilling goals to receive program details.
  • Low Quality: Random inquiries, spam bots, or mid-greeting drop-offs.

Key findings from the audit:

  • Only 2.8% of chat initiations converted into paying, registered program users.
  • 56.79% of chat volume comprised non-leads.
  • Learners frequently requested curriculum PDFs, schedules, and pricing before speaking to an advisor.

I audited 13 ed-tech competitors, mapped 4 qualification chatbot patterns, and synthesized learnings into a custom flow:

Competitor Chatbot FeatureLearnings for Our Flow
Direct transfer to humanCreates instant queues — pre-qualify first.
Multi-choice course selectorHelps users identify interest before transfer.
Automated PDF downloadsFulfills core user needs in-chat, reducing human load.
Chatbot qualification flow with course selector and document download paths
Competitor audit distilled into a flow that qualifies interest, serves syllabus PDFs in-chat, and routes only viable leads to consultants.

Key Design Decisions

  1. Pre-qualification nodes: Ask domain of interest and experience level before notifying a human consultant.
  2. Instant multimedia fulfillment: Syllabus brochures and pricing available inside the chat interface.
  3. CSAT rating trigger: Automated feedback prompt on chat completion for continuous flow tuning.
End-to-end chatbot flow — visitors self-serve syllabus downloads and pricing before a consultant joins.

Measured Outcome

~2800
Man Hours Saved Per Month
3,716
Transcripts Manually Audited
56.8%
Inquiries Filtered Out

What Changed Because of This Work

Inside sales consultants spend less time on syllabus requests, pricing FAQs, and spam chats — the bot handles self-serve fulfillment and qualification first. Phase 1 launched in targeted Indian regions with metric tracking and copy iteration; Phase 2 extended nationwide with multi-language support. Hand-auditing 10% of transcripts (when NLP failed) became the team's standard for validating conversational changes before rollout.