Conversational designSimplilearn

Designing out 2,800 hours of avoidable sales-chat load

I audited 36,000+ conversations, studied how comparable education platforms qualified visitors, and redesigned the chatbot around intent, course discovery, and a deliberate human handoff.

Role: Conversation Designer
Data: 36,000+ conversations
Partners: Inside sales & regional operations
Deliverable: Qualification flow & rollout plan

Qualification and course-discovery states from the redesigned chatbot flow.

The sales team was absorbing low-intent traffic

More than half of incoming chat sessions did not become viable leads. Consultants were still handling syllabus requests, pricing questions, spam, mid-greeting drop-offs, and visitors who were only beginning to explore.

That created two connected problems: learners could not complete common tasks without waiting for a person, and consultants had less time for conversations where their judgment mattered.

36,000+

monthly chat logs in the research set

56.79%

chat volume classified as non-leads

~2,800

consultant hours tied to addressable monthly load

What the 2,800-hour number means

It sizes the consultant time attached to low-intent and self-service conversations. It is the opportunity the redesign targeted, not a claim that every one of those hours was recovered after launch.

Manual review made the data usable

Automated language processing was not dependable enough for the shorthand, incomplete sentences, and inconsistent intent in the raw transcripts. I manually reviewed a 10% sample—3,716 conversations—and classified the behaviors behind the headline volume.

3,716

chat transcripts manually monitored

40.85%

higher-quality conversations in the sample

59.15%

lower-quality conversations in the sample

Higher intent

Visitors willing to share contact details, describe an upskilling goal, or ask for specific program information.

Lower intent

Spam, unrelated requests, early drop-offs, and information needs that could be resolved without a consultant.

Figma audit of the existing Simplilearn chat entry points and callback paths — Existing-state audit from the Figma file. Chat appeared inconsistently across pages and offered multiple paths to a callback or live chat.

Benchmarking exposed the recurring failure modes

I mapped qualification and handoff patterns across education platforms including Hero Vired, Emeritus, Great Learning, and UpGrad. The visual audit showed the same issues repeating in different forms: contact capture came too early, course discovery was shallow, and the promised counselor was not always available at the end of the flow.

Figma audit of UpGrad chatbot entry points, qualification questions, course discovery, and handoff branches — One competitor study from the Figma file: UpGrad's entry points, qualification paths, course discovery, and callback branches.

Qualify with purpose

Ask only for information that changes routing or gives a consultant useful context.

Support exploration

Let visitors browse by learning goal and domain instead of forcing an immediate lead form.

Set handoff expectations

Keep the human route visible and avoid promising an instant counselor when none is available.

The redesign starts with intent, not contact capture

The proposed flow first asks what the visitor is trying to accomplish. A learner who wants to upskill can name themselves, browse a curated set of subject areas, expand the list, or admit they are unsure. Each answer advances the conversation without hiding the route to a person.

Chatbot asking the visitor's name after they say they want to upskill — Start with the visitor's goal, then ask for the minimum context needed to continue.

Chatbot presenting course domains and an option for visitors who are unsure what to choose — Offer concrete domains, progressive disclosure, and a route for people who have not chosen a program.

Four decisions shaped the full flow

Route by intent. Distinguish program discovery, syllabus and pricing needs, callback requests, and visitors who want to talk now.
Progressively disclose the catalogue. Show a useful first set of domains, then let people reveal more rather than facing a wall of options.
Resolve common requests in chat. Program information and supporting material should not require a consultant to join.
Preserve a deliberate human handoff. Automation prepares the conversation; it does not remove the consultant from the experience.

Rollout paired the new flow with measurement

The experience moved into a phased rollout so the team could review behavior and copy before expanding coverage. The shipped flow gave visitors a clearer way to self-serve and gave consultants more context when a conversation reached them.

Confirmed output: a shipped qualification and self-service flow backed by transcript analysis and competitor research.

Measurement boundary: the ~2,800 monthly hours remain the addressable workload estimate. Realized savings require post-launch handling-time and lead-quality data.

What I would measure next

I would instrument each branch against consultant handling time, callback completion, lead quality, and downstream enrollment. That would separate three different outcomes—traffic filtered, time actually recovered, and sales conversations improved—instead of compressing them into one headline metric.