why new delivery drivers quit before their first order

the part of the funnel nobody owns

every delivery company counts drivers differently, and the way they count tells you where the pain is. Shadowfax, in its filing, only counts a partner once they have "completed at least one delivery". Eternal reports "delivery partner recruitment and onboarding cost" as its own line, across 5.52 lakh monthly active delivery partners in FY26. Swiggy's DRHP talks about "quicker onboarding of delivery partners" as a way to "reduce delivery partner attrition rates". all three are saying the same thing: install to first order is expensive, and a lot of people fall out on the way.

the cost keeps going up. vendors who sell onboarding services put it at ₹1,600 to 2,400 per rider by mid 2026, up from ₹800 to 1,400 in 2024 (vendor claim, Caller Digital and RichAutomate). the same vendors say half a new cohort stops logging in somewhere between day 88 and day 110. tenure across gig work has fallen to three or four months, according to Mint. you pay more for each driver and keep them for less time.

recruitment has a team. dispatch has a team. the hours between install and first order usually belong to nobody, and the only record of what went wrong is a one star review.

where it breaks, step by step

the launch sequence looks roughly the same in every partner app: download, signup, documents and KYC, training, first order. Caller Digital counts 12 to 17 steps in that stretch (vendor claim). the Play Store reviews people leave from inside it are more honest than any dashboard, and a few are worth quoting exactly as written, typos and all.

signup

on Snabbit's listing, a reviewer on 9 march 2026 wrote that the app kept "asking for onvording agent code". someone wanted to work and hit a field that assumed they already knew a person inside the company. no screen said where the code comes from.

language breaks at this step too. on 18 october 2026 a Zypp reviewer wrote that there was "no option to choose telugu as language". that driver may have been perfectly capable. the app just never spoke to them in a language they think in.

documents and KYC

a Borzo reviewer on 18 september 2026: "same approval message appears again and again". they've done what was asked and the app repeats itself. is it waiting on them, or on the company? did a photo fail? the message doesn't say, so they guess, and guessing on a phone usually ends in an uninstall.

training

a review of the Rapido captain app's training video: "Playing In Loop Again And Again". a MagicFleet reviewer on 20 september 2026 got further and still got stuck: "manual training order not reviewed yet". in both cases the driver showed up to learn and ended up waiting with no idea what came next.

first order

the first order is a new screen with new buttons, real money and a customer waiting, after days of a process that never felt like anyone was explaining it. it's when drivers most need help and when the app gives the least.

who the new driver actually is

the lazy story is that drivers struggle because they aren't good with phones. the data says otherwise. a government study by VVGNLI across four metros found 92% of gig workers had secondary to graduate level schooling. IDinsight found 95% owned a smartphone before they joined.

what they mostly don't have is the app in their language. in a 2023 Borzo survey reported by CNBC TV18, 84% of drivers said they use a local language while delivering and 82% said they want one inside the app. an English label on a KYC screen is a small wall. a dozen of those in a row is enough to make someone give up.

and there are a lot of these people coming. NITI Aayog projects gig workers in India going from 7.7 million in 2020-21 to 23.5 million by 2029-30. StatCounter puts Android at 94.84% of mobile web traffic in India as of october 2026. a partner app here is an Android app used by someone who'd like to be told what to do in Hindi or Telugu, on the screen in front of them.

why calls, bots, videos and tour tools fall short

companies know this, and they've thrown real money at it. one large platform ran 300 to 500 outsourced callers just for partners who "had not taken their first order" (vendor story, Bolna). another reportedly spent ₹120 to 130 crore a year on telecallers at 8 to 10% conversion (vendor claim, Hunar). calls help, but the caller can't see the driver's screen, and the driver has to juggle a call and the app on one phone.

phone voice bots are cheaper than people but still a call. training videos are recorded once, drift out of date the day a button moves, and, as the Rapido review shows, sometimes just loop.

the in-app tools built for SaaS don't fit either. Whatfix and Pendo have mobile SDKs, but they're quote-only, built for big software teams, and speak in text tooltips with no Indian voices. UserGuiding has no native Android SDK. Plotline starts at $999 a month and has no voice. Apxor does in-app nudges for Porter (vendor claim), which is closer, but nudges aren't someone talking you through a screen. none were made for a driver who reads Hindi more easily than English and is trying to upload a licence photo at a petrol pump.

what voice in the app can fix, and what it can't

the idea behind heytappr is simple. when a new driver is on a screen, a voice explains that screen in their language and a highlight shows which thing to tap. every step is also written on screen, so it works on mute at a busy signal. the guides are the ones your ops team wrote and approved, so it never makes up an answer about payouts or documents. if a release moves a button, the broken step gets flagged. and the team gets a funnel from signup to first order that shows the exact screen where people stopped. it ships as an Android SDK, one Gradle dependency and a line in Application.onCreate.

that would have helped the Snabbit driver, who needed one sentence about the agent code, and the Zypp driver, if Telugu was among the guide's languages. it helps most on the first order screen, where nobody is around to explain.

it would not have fixed the Borzo or MagicFleet reviews, at least not fully. if the approval is stuck in a queue or nobody has reviewed the training order, a voice can tell the driver what's happening and how long it usually takes, which is better than silence, but it can't approve anything. the same goes for verification delays, joining fees and deposits, slot or team lead approvals, and the hardest one of all: a driver who finishes everything and then gets no orders in their area. those are back-office and supply problems. a guide that papers over them would just make drivers angrier, a little later.

why now

for years, explaining a screen in Hindi or Telugu meant a person on a phone, because good synthetic speech was expensive. that changed quickly. a 5 minute guided script runs to about 4,000 characters, which at Sarvam's text to speech prices comes to roughly ₹4 to 12 per driver (estimate). Porter added regional language audio to its app. a former Zomato product lead added vernacular audio walkthroughs in six languages inside an offer card, tested it in three cities for two weeks against the old flow, and hit their targets before rolling it out nationally. that was about incentives, not first orders, and no lift was shared, but big teams clearly trust the idea.

when the cost of explaining a screen drops from a phone call to a few rupees, it stops making sense to save explanation for drivers who are already lost. you can give it to every new driver, on every screen, from the first tap. some will still quit. i'd just like fewer of them to quit because a screen said "agent code" and nobody told them what that meant.

if you want the longer version of how spoken guidance works, i wrote about voice guided in-app help and in-app guidance for Android apps.