KYC Automation with Voice AI: Partial KYC to 95%+
Most people discuss KYC automation as a verification problem: OCR, face match, database checks. However, for most Indian apps, verification isn’t where users drop off. Instead, they slip through the gap between starting KYC and finishing it. For example, a rider uploads Aadhaar, gets distracted before PAN and never comes back. Similarly, a loan applicant’s selfie fails silently. Meanwhile, a user thinks they’re done when they’re not.
This guide covers KYC automation for that gap: the partial KYC problem. First, it explains how a voice AI agent recovers those users. Then, it shows how we used it to take KYC completion past 95% in Shiprocket Quick’s rider onboarding. Finally, it helps you calculate what partial KYC is costing you today.
What is partial KYC?
Partial KYC (or incomplete KYC) is when a user starts an identity verification flow but stops before submitting every required document and finishing every step. For example, a user uploads Aadhaar but not PAN, or submits documents that fail review and never re-uploads them. As a result, the user exists in your system but can’t transact, earn or go live.
A clarification for fintech readers: “minimum KYC” or small PPI wallets are a specific RBI regulatory category with their own limits. This article, however, is about the operational problem of users who abandon a KYC flow midway. In fact, that problem exists in every industry that onboards people with documents: gig platforms, NBFCs, lending apps, insurance, broking and wallets.
What is KYC automation?
KYC automation is the use of software to collect, verify and follow up on customer identity documents with minimal manual work. Most KYC automation tools focus on verification (OCR, Aadhaar offline XML, PAN validation, liveness checks). However, the less-solved half is follow-up automation: getting users who stalled to come back and finish.
Why users abandon KYC midway: the Stall-Point Taxonomy
Before automating follow-ups, you need to know where each user is stuck. After all, “KYC incomplete” is not one status. In our experience, it breaks into five distinct stall points, and therefore each needs a different message.
| Stall point | What the user is thinking | What actually fixes it |
|---|---|---|
| 1. Started, nothing uploaded | “I’ll do it later.” | A timely nudge plus a direct link |
| 2. One document done, one pending (e.g., Aadhaar done, PAN pending) | “I thought I finished.” | Telling them exactly which document is left |
| 3. Uploaded but rejected (blurry, cropped, wrong side) | “Why isn’t my account active?” | Explaining the rejection reason in plain language |
| 4. Data mismatch (name on PAN ≠ name on Aadhaar) | “This is too complicated.” | Reassurance plus clear next steps, or escalation to a human |
| 5. Submitted, pending verification | “Is something wrong?” | No call needed. Calling here wastes money and annoys users |
The insight: most KYC reminder systems send the same message to all five groups. So a user at stall point 3 who gets “Complete your KYC now!” opens the app, sees nothing obviously wrong and leaves again. In short, recovery depends on specificity: telling each user their exact pending item, in real time.
Why SMS, push and email reminders don’t recover partial KYC
Most teams already send reminders. However, they plateau for three reasons:
- They’re static. A templated SMS can’t say “we couldn’t read your PAN photo because it was blurry.” A dynamic one could, but even then it can’t answer the follow-up question (“Can I use my PAN e-card instead?”).
- They’re easy to ignore. Users swipe away push notifications. Also, users in tier-2 and tier-3 markets often treat promotional-looking SMS as spam.
- They can’t handle objections. “Why do you need my Aadhaar?” or “I don’t have my PAN with me right now” are conversations, not notifications.
Human callers solve all three, but they don’t scale. For instance, a human calling team can’t call every stalled user within minutes, at 9 PM, in volumes that spike with every hiring drive.
Timing matters too. Harvard Business Review’s study The Short Life of Online Sales Leads (Oldroyd, McElheran and Elkington, 2011) found that firms contacting leads within an hour were nearly 7x more likely to qualify them than firms that waited even one hour longer. Likewise, a user who abandoned KYC 10 minutes ago behaves like a hot lead. That’s because intent decays fast.
How KYC automation with voice AI works on Twenty2
Twenty2 lets you build a voice agent as a flow graph: a set of conversation steps, decisions and tool calls. Here’s how a partial KYC recovery flow works end to end.
Step 1: Trigger on stall
First, your backend fires a webhook when a user’s KYC status stays incomplete beyond a set window (e.g., 30 minutes after their last KYC action). Also, the call can trigger at any point in the flow, not just at signup.
Step 2: The agent calls and checks live status
During the call, the agent makes a tool call to your KYC API. It then queries the user’s current status in real time, not a stale snapshot from when the system queued the call. So if the user uploaded PAN two minutes ago, the agent knows.
Step 3: It names exactly what’s left
Instead of “please complete your KYC,” the agent says something specific:
Shiprocket Quick · rider registration callAgent: Hi Ravi, this is a call from Shiprocket Quick about your rider registration. We’ve verified your Aadhaar. Only your PAN card is pending. Do you have it with you?
Rider: I uploaded it yesterday.
Agent: I can see you uploaded it, but the photo was too blurry to read. So you’ll just need to re-upload a clearer photo. I’ll send you a WhatsApp link right after this call.
Rider: Okay.
Agent: Done. Once we verify it, you can start taking deliveries.
Illustrative call flowStep 4: It handles objections
The graph includes branches for common questions: why the app needs a document, what to do if a card is lost, how long verification takes. Meanwhile, the agent escalates complex cases, such as data mismatches, to a human.
Step 5: A WhatsApp upload link follows the call
After the call, Twenty2 triggers your WhatsApp Business provider’s API to send a message with a direct link to the exact pending step. As a result, the user doesn’t hunt through the app. Instead, one tap takes them to the right screen.
Step 6: Re-check and retry
If the status is still incomplete after a set window, the flow can schedule a follow-up call or message. Finally, the flow drops users who complete automatically, because it checks the status live.
Want to see this flow running on your own KYC API? Book a demo with Twenty2.
Case study: Shiprocket Quick rider onboarding
The context: Shiprocket Quick onboards delivery riders at scale. Before a rider can take orders, they must upload Aadhaar and PAN for KYC.
The problem: 30–40% of riders left KYC midway. These riders had downloaded the app, signed up and shown clear intent to earn, but they remained in an incomplete state. Consequently, every stalled rider meant lost delivery capacity and wasted acquisition spend.
The solution: We built a partial KYC recovery flow on Twenty2:
- Twenty2 triggers a call whenever a rider’s KYC stays incomplete
- A live tool call to the KYC API identifies the exact pending document (Aadhaar, PAN or a rejected upload)
- The agent explains what’s left and why
- The flow then sends a WhatsApp message with an upload link after the call
The result: KYC completion rose to 95%+ across the app.
Why it worked:
- Specificity. The agent told riders exactly what was missing, not just that something was.
- Real-time data. The agent never asked for something the rider had already submitted.
- Channel pairing. Voice built urgency and trust; meanwhile, WhatsApp removed friction for the actual upload.
- Fit with the audience. India’s gig workforce is large and mobile-first. For example, NITI Aayog’s report India’s Booming Gig and Platform Economy (2022) estimated 77 lakh gig workers in 2020–21, projected to reach 2.35 crore by 2029–30. Therefore, for this audience, a phone call is often the most trusted channel.
The KYC Recovery Math: what partial KYC is costing you
Before buying any tool, first use this formula to size the problem:
Stalled users/month = Monthly KYC starts × Midway drop-off rate Recovered users/month = Stalled users × Recovery rate Idle acquisition spend = Stalled users × Cost per signup
Worked example (illustrative numbers; plug in your own):
| Input | Value |
|---|---|
| Monthly KYC starts | 10,000 |
| Midway drop-off | 35% → 3,500 stalled users |
| Cost per signup | ₹300 |
| Idle acquisition spend | ₹10.5 lakh/month |
| Completion before (65%) → after (95%) | +3,000 completed users/month |
So if each activated user is worth even ₹500 in first-month contribution (a delivery rider’s output, a loan’s processing fee, a funded trading account), that’s ₹15 lakh/month in recovered value from users you had already acquired.
Rule of thumb: if your midway KYC drop-off is above 20% and your cost per signup is above ₹100, then partial KYC recovery is likely one of your highest-ROI automation projects.
Voice AI vs SMS vs WhatsApp vs human callers for KYC follow-ups
| Factor | SMS / Push | WhatsApp only | Human callers | Voice AI agent (Twenty2) |
|---|---|---|---|---|
| Gets the user’s attention | Low | Low (easy to leave unread) | High | High |
| Tells user the exact pending document | Limited | Limited (fixed templates) | Yes | Yes, live via API |
| Handles questions and objections | No | No | Yes | Yes |
| Speed after drop-off | Instant | Instant | Hours to days | Minutes |
| Scales with hiring/signup spikes | Yes | Yes | No | Yes |
| Cost per follow-up | Lowest | Low | Highest | Low |
| Direct upload link | Yes | Yes | No | Yes (sent on WhatsApp after the call) |
WhatsApp is a useful channel in India, where Meta reports 500 million+ users. However, on its own, a message is easy to leave unread, and it can’t answer questions or explain why an upload failed. In contrast, a voice AI agent creates the moment of attention, guides the user through exactly what’s pending, and only then uses WhatsApp to deliver the upload link. As a result, agentic onboarding with voice AI drives completion that messaging alone can’t.
Where else partial KYC recovery applies
- NBFCs and lending apps: applicants who drop between eKYC and bank statement upload
- Insurance: policyholders with pending KYC before issuance
- Broking and investment apps: accounts stuck before activation
- Gig and logistics platforms: riders, drivers and field agents
- Real estate: buyers with pending documents for booking or home loan processing
In short, anywhere a user must submit documents before they become valuable, partial KYC is leaking revenue.
Implementation checklist: launch in under 2 days
- First, map your stall points using the five-point taxonomy above. Then pull counts for each.
- Expose a KYC status API that returns per-document status and rejection reasons.
- Set the trigger window. For example, start with 30–60 minutes after the last KYC action.
- Script the graph: greeting → live status check → specific ask → objection branches → WhatsApp link → close.
- Connect your WhatsApp Business provider with a deep-link template to the upload screen.
- Exclude stall point 5 (pending verification) from calls.
- Next, A/B test against your current reminders for two weeks. Measure completion rate and time-to-complete.
- Finally, add retries for users who don’t complete within 24 hours.
Frequently asked questions
Partial KYC is when a user starts an identity verification flow but doesn't finish it, for example by uploading Aadhaar but not PAN, or by never re-uploading a rejected document. The user is registered but can't transact or be activated.
A voice AI agent calls users whose KYC has stalled, checks their live status through your KYC API during the call, tells them exactly which document is pending or was rejected, answers their questions and sends a WhatsApp link to the exact upload screen.
No. The agent only tells users what to upload. Documents are submitted through a secure link, so call recordings never contain Aadhaar or PAN numbers.
SMS and push reminders are generic, easy to ignore and can't answer questions. Users often don't know which document is pending or why an upload failed, so a reminder alone doesn't move them.
Start with a window of 30–60 minutes after their last KYC action. Intent drops quickly, so calling within the hour gives you the best chance of recovery.
Any business that onboards users with documents: NBFCs and lending apps, insurance, broking, wallets, gig and logistics platforms, and real estate.