Skip to content
Aplora
EdTech
Aplora Sales

We validated the system inside our own operations first

Conversation review in an admissions funnel: how we validated the system on ourselves

An online school selling through consultation: application, advisor call, contract, payment, delivery. Intake runs in waves, and at peak the team physically cannot work through the volume.

Published: 2026-08-24 · updated: 2026-08-24

The period-comparison screen in Aplora Sales: before and after funnels and conversion side by side, with the change at each stage.Demo
The showcase build of the product deployed in this case — not a screen from the admissions funnel itself. Aplora Sales demo interface. The figures on screen are sample data, not a client result.

  • Conversion from hot leads

    30%Baseline: 9%

  • Share of calls that went nowhere

    20%Baseline: 60%

  • Share of second contacts

    55%Baseline: 25%

  • Time to send the quote and contract

    a day → 10 minutes

In short

  • Problem: conversation quality was known from a sample, and the reasons for a no stayed a hypothesis.
  • Solution: every conversation reviewed against the school's own sales book, the result written to the CRM, a follow-up draft handed to the advisor.
  • Result: the metrics are in internal review and will be published once the evidence audit is complete.
  • Systems: telephony, CRM, LMS.

Context

  • An online school in IT education, selling through a consultation call.
  • Admissions team: advisors plus a head of the intake campaign.
  • Process volume: application flow is uneven, with pronounced peaks at the start of each intake.
  • Systems: telephony with call recording, CRM, LMS, payment system.
  • The constraint became visible as volume grew: adding advisors increased cost without addressing conversation quality.

Baseline

Baseline metrics with their sources. Without them, any later result has nothing to be compared against.
  • A few recordings a month are reviewed out of hundreds
  • The reasons for a no stay a hypothesis
  • Commitments are captured from memory
  • The CRM is filled in after the fact and only partially
  • Before the work started we fixed: the share of conversations the manager actually listened to, completeness of required CRM fields, time from conversation to follow-up, and conversation-to-contract conversion.
  • Data source: CRM and telephony exports covering a comparable intake period.
  • The values are in evidence review and will be published together with the measurement window and the calculation formula.

Diagnosis

Which hypotheses were considered, why this one was chosen, what was assumed, and the condition under which we would have stopped.
  • Three hypotheses were on the table: not enough applications, not enough advisor time, or quality lost inside the conversations.
  • The third was chosen: application volume was sufficient, and the conversion spread between advisors pointed at how the conversation was run rather than at the funnel.
  • The assumption: the school's sales book describes practice that works, and departing from it genuinely degrades the outcome. That assumption was tested against historical recordings.
  • Stop criterion: if criterion-based review cannot separate conversations that closed from those that didn't on historical data, the hypothesis counts as unconfirmed and the work stops.

What we implemented

  • Data sources: call recordings from telephony, deal records from the CRM, programme details.
  • AI components: transcription with speaker separation, criterion-based review against the methodology, extraction of commitments and next steps.
  • Business rules: qualification definitions, required questions, next-step routing rules.
  • Integrations: telephony, CRM.
  • Human checkpoints: the advisor sends the follow-up; the head of admissions reviews contested cases.
  • Monitoring: share of conversations escalated to manual review, and the gap between the system's score and the manager's on a control sample.

How the process changed

Before

5 steps
  1. The conversation happens; the recording is stored in telephony.
  2. The advisor enters part of the information into the CRM from memory — usually later, between calls.
  3. The follow-up is written by hand whenever there is a gap.
  4. Once a month the manager listens to a few recordings, chosen at random.
  5. The reasons for a no are discussed at a standup, based on impressions.

After

5 steps
  1. The conversation happens; the recording goes to review automatically.
  2. The system fills the CRM fields and creates the next-step task.
  3. The advisor receives a ready follow-up draft and sends it after checking.
  4. The manager sees every conversation, with the ability to open the specific excerpt.
  5. The reasons for a no read as a recurring pattern rather than as separate anecdotes.

What was stuck

Conversation quality was judged from a few recordings a month. The reasons for a no stayed a hypothesis, commitments were captured from memory, and the CRM was filled in after the fact and only partially.

  1. 1The conversation happens; the recording goes to review automatically.
  2. 2The system fills the CRM fields and creates the next-step task.
  3. 3The advisor receives a ready follow-up draft and sends it after checking.
  4. 4The manager sees every conversation, with the ability to open the specific excerpt.
  5. 5The reasons for a no read as a recurring pattern rather than as separate anecdotes.
  6. Measured result

Results

Conversion from hot leads

×3.39% → 30%

Reported by: LearnIT figures, CommaCRM export

Share of calls that went nowhere

−40 pp60% → 20%

Reported by: LearnIT figures, CommaCRM export

Share of second contacts

+30 pp25% → 55%

Reported by: LearnIT figures, CommaCRM export

Time to send the quote and contract

a day → 10 minutes

Reported by: LearnIT figures, process observation

Economic impact

  • Method: the value of management time freed from manual QA, plus the change in conversation-to-contract conversion multiplied by the average programme price.
  • Implementation cost and ongoing cost of ownership — recognition, compute, support — are subtracted.
  • Actual figures will be published once the audit confirms the baseline and the measurement window. Publishing the method without the numbers is acceptable; publishing the numbers without the method is not.

Adoption

  • Used by the admissions advisors and the head of the intake campaign.
  • The advisor's role shifted: less time entering data, more time preparing for the next conversation.
  • Onboarding: walking through several of their own conversations with the manager at the start.
  • Individual scores were not shown to staff initially — the criteria were calibrated first.
  • Solution owner: the head of the intake campaign.

What's next

  • Scaling: the same criteria applied to repeat sales and programme renewals.
  • Next initiative: churn signals from the LMS and attendance data.
  • Decided against: unattended e-mail to students without advisor confirmation — in admissions the cost of an error outweighs the gain in speed.

Have a similar workflow? Let's check whether the hypothesis transfers

Another company's result is not a promise. It does show where to look.

All cases