₹299 consultation — paid vs walked away

1. Is there a recorded reason for the loss?

Almost never — and that is the first finding. The CRM's last lead stage is operational noise for the overwhelming majority: Other AI Alerts (67), Paid Lead (35), Requested callback (15), Escalation (13). Only about 10 of 135 carry anything resembling an actual loss reason — Cold/Exploring 6, Personal Situation 2, Not eligible 1, Hung up on pitch 1. Nobody is writing down why a paying lead walked. Until a mandatory loss-reason field exists at closure, any "top reasons" list is inference, not record. What follows is inference from behaviour.

2. The observed drivers, ranked

Categories overlap — a lead can be both under-followed and cold.

3. Segments that convert at zero

Two of these are counter-intuitive and worth arguing about. The richest budget band and the people who say they are ready to apply both convert at zero. Likely reading: 40L+ buyers shop elsewhere or self-serve, and “ready to apply” means they already have an agent. Both look like qualification questions that are pointing the wrong way. n is small (22 and 27) — treat as a strong hypothesis, not a settled fact.

4. Are these leads being sufficiently followed up?

No — and it is not close.

38.5% of paying leads got one live conversation or none. The three shaded rows are all recoverable failures of process, not of the lead. And 76.3% have had no contact of any kind for over a month — 43% for over two months.

5. What conversion actually takes

Not one sale closed in two connects or fewer. Every one of the 18 conversions took at least 3 live conversations; 14 of 18 took 5 or more; the range runs 3 to 26 with a median of 8. Every converted lead also had at least one G-Meet actually held. The single cleanest statement available from this data: a paid lead is not a lost cause until it has had eight connects and three meetings — and 92.6% of the non-converted never got there.

The 10 leads who received 8+ connects and still did not buy are the honest denominator for “the pitch does not land” — 7.4% of the non-converted pool. The other 92.6% is a coverage problem, not a persuasion problem. Talk time is call_insights.actual_talk_time on outgoing calls only.

No. The AI safety net does not exist for the people who paid. Median AI attempts on a paying lead is zero — for converted and non-converted alike. Paying moves a lead onto a human queue and simultaneously off the robot, so when a counsellor goes quiet there is nothing behind them.

Combined AI + coach: was anyone trying?

Median total dial attempts (AI + coach combined): 5 for a non-converted lead, 15 for a converted one. Only 28.9% of the non-converted ever received 10 or more attempts from any channel.

The sharpest number in this card: 42 of the 135 non-converted — 31.1% — had no AI attempt at all and one coach connect or fewer. For nearly a third of the people who paid us, neither channel did meaningful work. That is not a lead-quality problem and it is not the pitch; it is an unassigned queue.

Every field below is captured on the form itself, so this can score a lead the moment it arrives, before anyone spends a rupee or a phone call on it. Base conversion across the 153 payers is 11.8%; lift is against that base.

Red flags — four answers that have never once converted

40.5% of the paid pool carries at least one red flag and none of them has ever bought a plan. Two are counter-intuitive — the richest budget band and the people who say they are ready to apply. The likely reading is that both already have an agent or intend to self-serve, so they buy the ₹299 for a second opinion rather than as a first step.

Positive signals, ranked by lift

Country count is a decisiveness proxy and it works. One target country converts at 16.0%; four or more at 3.0%. Meanwhile the form's existing isQualified flag gives 12.8% against 10.2% for unqualified — a lift of 1.1×, which is to say it is doing almost nothing. Replacing it with the rule below is the single highest-value change to the form logic.

The proposed qualification rule

Read this as a triage order, not a gospel. It is fitted on 18 conversions, so the positive tiers are near-certainly over-fitted — the exact 30.6% will not hold on new data. The red-flag rule is the durable part: 62 leads, zero conversions, and a probability of about 0.04% that this is chance if the true rate were the 11.8% base. Use tier D to stop spending counsellor hours, and validate tiers A–C prospectively over the next 60 days before wiring them into anything automated.

This is the most important finding in this deck. The people who dismissed the ₹299 form are not a worse population than the people who paid. On every profile dimension we capture they are the same. What differs is entirely what we did to them afterwards.

Are they actually as qualified? Yes.

Three near-identical populations, a 10× spread in outcome. Nothing in the profile explains it.

What we actually did to each group — median per lead

The two groups are mirror images. Pay ₹299 and you get a human and no robot. Don't pay and you get a robot and no human. The median non-payer has had zero live conversations with a person and ten attempts from the dialer.

Human contact is the whole game

A non-payer worked to eight connects converts at 26.7% — higher than the ₹299 cohort's overall 11.8%. The same shape holds for meetings: 0 meets held → 0.09%, one meet → 2.37%, two or more meets → 32.8% (21 of 64), a 38× lift. 28 of the 31 non-payer conversions had at least one meeting held, and 28 of 31 had 3+ connects. There is no cheap path to a sale in this data.

More AI does not help — the dose-response runs backwards

Piling on AI calls is associated with lower conversion. Read it correctly: heavy AI volume is a marker of a lead nobody is working, not a cause of failure. But the practical conclusion stands either way — the 858 leads who received 20+ AI calls produced 3 sales. That is where the dialer capacity is being burned.

Red flags, validated on an independent 3,593-person sample

This is why validation matters. Two of the signals I reported from the 153-person paid cohort do not survive contact with a larger sample. Budget 40+ Lakhs converts at the base rate among non-payers — the zero in the paid cohort was small-sample noise, and it should be dropped from the rule. Country-count reverses sign entirely. The three that survive — Bachelor's, intake 2028+, and “ready to apply” — now rest on 1,577 people and 2 conversions between them. Those are safe to act on.

Where the inefficiency is, and how big it is

Do not multiply 1,677 by 26.7%. Leads reached eight connects because a counsellor judged them worth it — that decision carries most of the signal, so the observed 30.6% for worked no-red-flag leads is an upper bound contaminated by selection, not a forecast. The honest statement is narrower and still large: 2,177 non-payers have never had a single live conversation, they are indistinguishable on profile from people converting at 11.8%, and the one intervention that moves this number in both cohorts is a human on the phone eight times. Run the 459-lead slice as a measured pilot — roughly 90 counsellor-days at 40 connects a day — and read the real incremental rate off that rather than off this table.

The AI is not the problem here. It is the only thing working. 98.2% of non-payers were dialled by the AI, 85.8% actually connected with it, and 37.7% had a conversation longer than five minutes. The dialer did its job and then the lead fell into a hole.

AI coverage — the exact inverse of the paid cohort

50,314 AI dials, 11,937 connects (23.7%), 402 hours of AI talk time spent on non-payers — against 512 dials and 8 hours on payers. 99.0% of the entire AI dialer effort is pointed at this cohort. Nobody can argue we did not try to reach them.

The AI conversation predicts conversion — but only if a human picks it up

A long AI conversation doubles the odds — but 1.70% is still nowhere near a business. Now split that same group by what happened next:

Identical AI outcome. 25% versus zero. The only variable is whether a human followed up. The handoff happens for 55 of 865 leads — about 6% of the time. This is not a lead-quality problem, an AI-quality problem, or a pitch problem. It is a handoff that does not exist.

The list to work on Monday

This supersedes my earlier recommendation. I previously suggested piloting on 459 leads chosen by profile alone. These 126 are strictly better: same profile filter, plus proof they will talk to us for five minutes. And the fix is not more dialling — the AI already reached them. It is a rule that says any lead whose AI call runs past five minutes gets a human within 24 hours. Today that happens 6% of the time.

The 0.00% on 810 leads is a real zero, not a rounding artefact. Caveat in the same breath: leads reaching 8+ coach connects were selected by counsellors, so 25.45% is an upper bound, not a forecast for the 810. The zero is the finding; the 25% is the reason to test.

• This is correlation, not proof that the ₹299 causes the sale. People who pay ₹299 are self-selecting as serious buyers. Some of the 11.7× gap is the filter working, not the paywall creating intent. The honest read is that ₹299 is an extremely strong qualifying signal — not that charging it converts people.
• The paid cohort also gets far more human effort — 96% get a coach dial versus 58%, and 84% get a G-Meet held versus 10%. Effort follows payment, so the outcome gap is partly a resourcing decision you already made, not purely lead quality.
• Plan attribution is current-state. A lead's Premium/Ultimate status comes from their entitlement today; if someone upgraded then lapsed, or bought via a counsellor payment link rather than self-checkout, the plan still shows. Counsellor-link sales are not in the orders table at all, so the true plan count in both cohorts is a floor.
• Cohort membership is by phone. 149 of 157 successful ₹299 orders matched a form by phone; the 8 that did not are excluded from the paid cohort and will be sitting in the unpaid one, slightly flattering it.
• A person can submit the form more than once. Rows are forms, not unique humans, so someone who filled twice counts twice in the base. Paid leads re-submit more often, which mildly inflates the paid denominator and therefore understates the conversion gap.
• “Started ₹299 checkout” is a form-level flag. It sits on the specific form row where checkout opened, so a payer who submitted the form twice carries it on only one row — which is why the paid cohort shows 149 checkouts against 234 forms. Use it as a directional signal, not as a denominator for the payment step.
• April is excluded — zero completed ₹299 payments and Spaces on only 59 of 1,042 forms.

Sources — virtualofficeconsultations (form, filtered to page_link containing virtual-office), orders (₹299 success), gameplayers (phone bridge), conversationmessages (author = lead's gamePlayerId), ai_voice_conversation (userId = gamePlayerId, source ai_outbound_call), argonaut_counsellors_call_history (outgoing only, call_insights.actual_talk_time), argonaut_task (task_type appointment; held = has meet_participants), entitlements (activePlan.planName). Dates in IST. Pulled 27 Aug 2026.