GoKwik Says AI Voice Calls Doubled Abandoned Cart Recovery: Reading the 30-Brand, Six-Month WhatsApp Data
India's GoKwik released data from 30 D2C brands that added AI voice calls to WhatsApp cart recovery. What sits behind the 2x recoveries and 60-65% answer rates, what remains undisclosed, how consent rules apply, and what it means for e-commerce teams in Japan.
Key Takeaways
- Indian e-commerce technology company GoKwik released six months of data from 30 D2C brands that use AI voice calls for abandoned cart recovery. Adding AI calls to WhatsApp reminders doubled cart recoveries, delivered more than 10x incremental return on spend versus WhatsApp alone, and produced an average answer rate of 60-65%
- The AI does more than remind shoppers about a pending cart. It answers last-minute questions about products, delivery dates, payment options, discounts and return policies, and helps shoppers complete the order. It is a seller-side conversational AI agent stepping into the transaction flow just before payment, with reported results to back it
- The figures are GoKwik's own data, and third-party verification, call costs and opt-out rates are undisclosed. India is moving toward rules that require businesses to declare automated calling, so for e-commerce teams the prerequisite for adoption is deciding how to measure impact and how to design consent before the first call
GoKwik publishes six months of AI calling data for cart recovery
AI-powered WhatsApp voice calls boost D2C recoveries and ROI, with 60-65% call answers driving 2.5x purchases vs single channels.
www.thehindubusinessline.comGoKwik, which provides checkout and customer engagement tools to Indian D2C (direct-to-consumer) brands, has released data on the results of its AI voice calls. According to BusinessLine's September 26, 2026 article, the data covers 30 brands using the AI Calling feature on its customer engagement platform, KwikEngage, over a six-month period.
Three numbers made the headlines. Brands that added AI calls to WhatsApp abandoned cart reminders recovered twice as many abandoned carts as with WhatsApp alone. The incremental return on spend was more than 10 times that of WhatsApp-only recovery, and the average call answer rate was 60-65%. Shoppers who engaged through both AI calls and WhatsApp were reportedly 2.5 times more likely to complete a purchase than those who interacted through only one channel.
KwikEngage currently handles more than one million AI calls a month. GoKwik, founded in 2020, says it serves more than 20,000 brands and 220 million online shoppers, so the 30 brands are a small subset.
In July 2026, GoKwik also launched a way to shop across multiple D2C brands inside ChatGPT with PayU. If that initiative put a storefront in front of buyer-side AI assistants, this one has AI reaching out to shoppers from the seller's side.
Carts are abandoned not for lack of intent but for lack of answers
The most important thing to read in this announcement is what the AI calls actually do. GoKwik co-founder and CEO Chirag Taneja offered a comment that reframes cart abandonment itself.
Abandoned carts are not always the result of weak purchase intent or pricing. In many cases, a shopper has a specific question that remains unanswered at the point of purchase.
In the announcement reported by Social Samosa, Taneja also described a division of roles. WhatsApp delivers the reminder, and a conversation gives the shopper a chance to resolve whatever is blocking the transaction. The types of questions GoKwik lists are product details, delivery timelines, payment options, discounts and return policies.
The AI call flow is described as follows. It contacts eligible shoppers, tells them they have a pending cart, explains applicable offers and helps them complete the order. Brands can set the tone of the conversation, the offer details and the calling schedule, and the feature plugs into existing WhatsApp recovery journeys. GoKwik says onboarding, configuration and sample-call testing can be completed within two working days.
What these conversations look like in practice becomes clearer from a customer story with GoKwik published by voice AI company Bolna. In one conversation, a shopper had chosen cash on delivery (COD), but the form showed an extra 50 rupee charge and the code number never arrived, so the order stalled. The AI offered to send a new checkout link over WhatsApp that would allow COD without any code. For a shopper whose discount code worked yesterday but failed today, the AI pointed to a GoKwik checkout link where the offer would be applied automatically. The conversations run in a mix of Hindi and English, and the AI switches languages mid-call.
This is not simply a voice version of a reminder. The AI listens to the shopper's question, picks a fix on the spot and resends a path to the checkout page. The purchase nudging that human call centers used to handle is being taken on by a conversational AI agent just before payment. The case study also includes a cross-selling scene in which the AI explains a skincare routine to a shopper with sensitive skin while recommending related products.
The scale shown in the case study is 250,000 minutes of calls and 400,000 unique customers. The page is undated, however, and it has not been disclosed whether the current AI Calling feature runs on Bolna's technology.
COD explains part of why this approach works in India. GoKwik built its business around tackling RTO (return to origin), where goods come back to the sender, for example when COD orders are refused at the door. The company now says it will extend AI calls to COD order confirmation, explaining that verifying orders before dispatch can reduce RTO risk. Non-delivery report (NDR) resolution, customer support, assisted selling and reactivation of inactive customers are also on the expansion list. Cart recovery looks like the entry point for AI calling, with the real target being the entire conversation before and after an order.
How to read the 2x and 10x figures
Strong numbers call for a check on what has not been disclosed. Start with the definition of incremental return. GoKwik says it calculated the figure from additional revenue from orders recovered after AI Calling was added to existing WhatsApp recovery flows. It has not said how call charges or AI fees are included in the spend side, or whether a randomly selected control group (holdout) was used. Pricing for AI Calling is also undisclosed.
Next is the 2.5x purchase completion comparison. It compares shoppers who engaged through both channels with those who engaged through only one, and shoppers who picked up the phone and stayed in the conversation may skew toward those who already had higher purchase intent. Indian retail intelligence site retailintel makes this counter-case, noting that much of the reported uplift may be an attribution artifact. Without disclosed call costs, discounting, escalation expenses, opt-out rates and a randomized holdout, it argues, a 10x figure is difficult to assess.
GoKwik itself says the improvement was recorded across all participating brands, but that the gains varied by category, customer base and existing recovery performance. The breakdown of the 30 brands, their names, absolute recovery rates and net revenue after returns and cancellations are all undisclosed. In other words, these figures are the company's own data showing that the approach can work, not a guarantee that another business will see the same multiple.
Even so, a 60-65% answer rate stands out. A call about a product the shopper just put in their cart has an obvious purpose. That close context is what makes AI calling viable.
Consent design comes before the call
Once AI is placing phone calls, regulation and consumer acceptance cannot be avoided. GoKwik says calls operate within consent, frequency and opt-out controls, are placed only between 10 am and 9 pm, and that shoppers can choose not to receive further calls.
In India, the rules in this area are still moving. According to Business Standard's September 15, 2026 explainer, the Telecom Regulatory Authority of India (TRAI) is finalizing amendments to its commercial communications regulations (TCCCPR) that would require businesses using automated calling systems to declare that traffic to their telecom providers. Under the proposal put out for consultation in March, undeclared automated calls could be treated as unsolicited commercial communication. Operators' AI-based systems flagged 22.99 billion incoming calls as suspected spam in the April-June 2026 quarter alone, and telling legitimate automated calls apart from spam has become a major challenge.
Elsewhere, the US Federal Communications Commission (FCC) ruled in 2024 that AI-generated voices fall within the Telephone Consumer Protection Act's (TCPA) definition of artificial or prerecorded voice. In the EU, Article 50 of the AI Act requires providers of AI systems that interact directly with people to ensure those people know they are dealing with an AI, an obligation that has applied since August 2, 2026.
The calling route is also an open question. BusinessLine's headline says "WhatsApp calls," but the body describes adding AI voice calls to WhatsApp campaigns, and the announcement does not make clear whether the calls go over regular phone lines or use WhatsApp's calling feature. Under the WhatsApp Business Calling API that Meta announced in July 2025, a business must obtain the customer's permission before placing a call. According to the developer documentation, businesses can send at most one permission request in 24 hours and two within seven days, and temporary permission lasts seven days. Four consecutive unanswered calls automatically revoke the permission. Whichever route is used, the shopper's ability to ignore or decline has to be built in.
What this means for e-commerce businesses in Japan
In Japan, LINE rather than WhatsApp is the dominant messaging app, and the conditions are not the same. LINE Official Account's "LINE Call" lets users call an official account, but businesses cannot place calls from their side. The WhatsApp flow of asking permission by message and then calling from the business cannot currently be reproduced as is on LINE.
The core of GoKwik's case, however, is less about placing calls than about the observation that some orders can be recovered by resolving last-minute questions through conversation. That can be tested on any channel. If a business offers a path for shoppers to start a call or chat themselves from the cart page or a LINE message, and lets AI answer questions about shipping fees, delivery dates, payment methods and return policies, it can test the same hypothesis while avoiding the regulatory and acceptance issues of business-initiated calls. Businesses in Japan that want to encourage purchases by phone should check beforehand how rules such as the telephone solicitation provisions of the Act on Specified Commercial Transactions apply.
Measurement should also be decided up front. The weaknesses pointed out in GoKwik's numbers double as a checklist for an internal test design. Randomly exclude some cart abandoners from outreach and compare, evaluate net gains after call and AI costs and discounts, and record opt-out rates and the content of questions. With that design in place, a business can judge its own results without relying on outside figures such as "2x."
The log of questions the AI answered is an asset in itself. Where shoppers hesitated and what they asked before nearly abandoning the purchase is data that shows gaps in product pages. As more external AI assistants read product information and answer shoppers directly, keeping delivery dates and return conditions in a machine-readable form becomes more valuable.
Conclusion
GoKwik's AI Calling reframes cart abandonment as a problem of unanswered questions rather than weak intent, and hands last-minute sales assistance to conversational AI. The 2x recovery and 10x-plus incremental return figures are the company's own data, with calculation assumptions and costs still undisclosed, but an operating scale of more than one million calls a month shows the approach has moved beyond the experimental stage.
The next things to watch are the results as the use cases expand to COD confirmation and non-delivery resolution, and how operations change once TRAI's automated calling rules are finalized. The moments when seller-side AI agents talk to shoppers are spreading from abandoned carts to after the order.
