Zolve × Nbyula · call-content analysis

What are John & Swatika
asking that the AI isn't?

The short answer: almost nothing. We read the transcripts of both sides — the two advisors' loan calls and the Nbyula loan AI agent's calls to the same kind of leads — and scored what each actually talks about. On every qualification question, the AI covers more, not less.

Basis 127 human deep calls · 70 AI deep calls (5-min+), 6 Aug Source call transcripts + AI call summaries, Nbyula CRM

The finding

John and Swatika's calls are qualification interviews — country, university, course, loan amount, co-applicant, income, collateral, prior rejection. The AI runs the exact same interview, more completely: it asks the co-applicant's pension amount, existing EMIs and CIBIL score on nearly every call, where the humans usually stop short. The only thing the humans do that the AI doesn't is after qualification — issuing the document checklist and giving lender-processing updates. So the qualification call is fully AI-doable, and the AI already does it — at far higher coverage.

Same questions, scored on both sides

Share of substantive calls (5-min+) where each topic came up. Blue = human advisors, green = AI agent.

Human advisorAI agent

Qualification questions — the AI asks these more often

Loan amount / quantum

Human

41%

94%

Co-applicant / guarantor

Human

43%

93%

University / course / intake

Human

54%

91%

Collateral / secured-vs-unsecured

Human

38%

76%

Country

Human

28%

63%

Co-applicant income / ITR

Human

37%

57%

CIBIL / credit score

Human

44%

Offer / admit letter

Human

16%

39%

Prior lender rejection

Human

10%

29%

After qualification — the one area humans do more

Document checklist / collection

Human

68%

36%

Lender-processing / login status

Human

69%

26%

Next steps / follow-up

Human

76%

57%

Moratorium / repayment terms

Human

14%

27%

Bank / lender names

Human

17%

24%

Exhibit A — the qualification call, side by side

Real excerpts. Lightly cleaned of speech-to-text noise; the wording and questions are as spoken.

Yash · loan AI agent9 min · Germany, unsecured

AIWhat country are you planning to study in?

UserGermany.

AIAnd what course are you looking at there?

UserInternational management.

AIWhich university in Germany are you targeting?

UserNot finalised yet — I have one offer letter, waiting on another.

AIWhen are you planning to start? Which intake?

UserSeptember 2026.

AIHow much loan are you looking to borrow — tuition only, or tuition plus living expenses?

UserTuition plus living, around 40 lakh rupees.

AIDo you have any collateral, like a property in India, for a secured loan, or would you rather go unsecured?

UserI'd rather go unsecured.

AIFor unsecured we need a co-applicant, typically a parent. Do you have someone in mind?

UserMy mom, or my uncle.

AIIs your mom salaried or self-employed?

UserShe has a pension.

AIWhat's her approximate monthly pension?

UserAround 24–25 thousand.

AIDoes she have any existing EMIs or loans she is paying off?

UserNo EMI.

AIAnd does she know her CIBIL score, or is that something we need to check?

UserWe need to check.

John Stanley4 min · prior rejection

AdvisorYou are going for masters in embedded systems… you have applied in Credila and Avanse. What is the process right now?

StudentCredila got rejected — twice — due to the loan. My father has three active loans.

AdvisorWhy did it get rejected, could you inform me?

StudentMy father has three active loans and…

John Stanley6 min · co-applicant

AdvisorYou are going for MS in International Business and Management, right? University of Kent, September intake. Who will be your co-applicant?

StudentThat is the problem — there is no co-applicant, their CIBIL is too low.

The AI's chain runs unbroken — country → course → university → intake → amount → collateral → unsecured → co-applicant → pension amount → existing EMIs → CIBIL check — in a single call. The human calls cover the same ground but usually stop at the co-applicant, and rarely reach the pension figure, EMIs or CIBIL that actually decide the case.

Exhibit B — the AI also handles the hard cases

A rejected applicant, diagnosed and re-routed on the call.

Maya · AI agent7 min · diagnosing a rejection

AII can see you are targeting an MSc in Applied AI & Business Analytics at University of East Anglia, and you have been working on your loan — a few things to sort out on that front.

UserI have accepted the offer and paid the deposit, but I have been rejected by 2–3 banks for the living-expenses loan.

AIRejections from a couple of banks don't mean the door is closed. When they rejected you, did they give a specific reason? Was it about your co-applicant?

UserYes — my father was co-applicant, his CIBIL is around 655, that is why they declined.

AIYour father's 655 is the main blocker — that makes sense. But you don't have to use your father. You mentioned a brother — have you considered using him instead?

UserMy sister used him and got her loan sanctioned…

Swatika Reddy7 min · basic qualification

AdvisorFor which country are you planning?

StudentItaly.

AdvisorHave you applied to the universities, and did you receive the offer letter?

StudentYes, offer letter received — Eastern Piedmont, Masters in Medical Biotechnology.

Exhibit C — what the humans uniquely do

This is the genuine human value today: it happens after qualification — issuing the document checklist and updating the student on where their file sits with the lender. These are real call summaries.

John Stanley

Document checklist sent via WhatsApp for the necessary submissions to proceed with loan processing; student asked to review and respond.

John Stanley

Processing discussed with SBI and PNB, plus a better rate of interest available through Union Bank; next step is document submission.

John Stanley

Agriculture-income certificate and other documents outlined with a 3–4 day timeline; profile to be shared with Credila and Poonawala for processing.

The recommendation: give the qualification call to the AI — it is more thorough and reaches ~78% of leads vs the team's ~52%. Keep the humans for document collection and lender-processing follow-up, the one place they add value the AI doesn't. Even the checklist step is partly automatable.