We ran an AI agent on 36 Indian D2C funnels. The 5 leaks it found most
By Madhur Tyagi · · 11 min read
Arjun checked his store every night, on his laptop, on office wifi.
It looked perfect. His buyers were on phones, on mobile data, in a hurry.
Arjun is a made up Pune snack founder. If your store looks fine to you and sales still feel slow, you know his week.
The short answer
On 23 October 2026 our AI agent read the phone pages of 36 Indian D2C brands. The five leaks it found most: no reply time (35 of 36), no delivery, COD or returns line on the opening phone screen (31 of 34), product pages over 3 MB (27 of 36), no WhatsApp link on the product page (18 of 36) and no star rating up top (14 of 34).
Every founder I know checks their own store the same way. On a laptop. Logged in. With fast wifi and a page that has loaded a hundred times before.
Your buyer sees a different store. A small phone. Mobile data. No idea where anything is. The store you check is not the store they shop in.
So we built an AI agent that shops like a stranger. Then we pointed it at 36 Indian D2C brands, from big household names to young brands, and wrote down what it saw. No brand is named here. The point is the pattern, not the shame.
Key takeawayThe store you check on your laptop is not the store your buyer shops in on a phone.
I have run ads and funnels for Indian brands since 2017. Most money I have seen lost was not lost in the ad. It was lost in the minute after the click. This study shows exactly where that minute breaks. One of the five leaks showed up on 35 of 36 sites, and almost nobody talks about it. We get to it at leak 1.
What did the agent actually check?
Think of a mystery shopper at a kirana store. She walks in, looks around, asks a question and notes what happens. She does not go behind the counter.
Our agent works the same way. It is a small team of four AI helpers, each with one job. The planner picks what to check. The collector opens the public pages on a phone screen and a laptop screen and saves what it sees. The analyst scores 10 areas from 0 to 100 and drafts each leak with its proof. The writer turns that into a report. Then a review step keeps the leaks it can see on the screenshots and drops the rest.
How the agent team works
Four helpers read your store like a stranger
Goal Find what stops a new buyer on a phone
- 1PlannerPicks what to check for a D2C store: offer, proof, buttons, delivery line, WhatsApp path.
- 2CollectorOpens the home page and a product page on a phone and a laptop. Saves screenshots and timings. Never logs in.
- 3AnalystScores 10 areas from 0 to 100. Drafts each leak with its proof: page, screen, measured value.
- 4WriterTurns it into a short report: what it saw, why it costs buyers, one fix each.
Review Keeps the leaks the screenshots show. Drops the rest.
- Public pages
- No login
- No forms sent
- Follows robots.txt
We ran it on 23 October 2026, on 37 brands across skincare, food, grooming, fashion, home and more. For each brand it read two pages: the home page and one product page. It read just what any visitor can see, never logged in, never filled a form and followed each site's robots.txt rules. One run did not finish, so 36 brands gave full phone data. For the checks on the opening phone screen, we reviewed 34 sets of screenshots. Two sites did not show their Indian store to our test machine.
The study in four numbers
The average funnel scored 72 out of 100. The highest was 83. The lowest was 57. Nobody was a disaster. Nobody was leak free either.
Up next You know how the agent reads a store. Next: why your own eyes miss what it catches, and a chart of everything it found.
Why do busy founders miss these leaks?
Arjun runs a millet snack brand. His ads were running. His Instagram looked lovely. His orders were flat.
He did what we all do. He opened his store on his laptop and scrolled. Everything was there. The price. The reviews. The delivery promise. He could not see a problem, so he blamed the ad.
Here is the root of it. You know your own store too well. You know the delivery line is three swipes down. You know the WhatsApp button hides in the footer. Your brain fills the gaps for you. A stranger has no map. If they cannot see it, it does not exist.
That is why an outside reader helps. It has no memory of your store. It sees what a new buyer sees, every single time.
What the agent found
How often each leak showed up
Leak 1No reply time anywhere on the two pages
97%35 of 36 brands
Leak 2No delivery, COD or returns line on the opening phone screen
91%31 of 34 brands
Leak 3Home page over 3 MB on a phone
83%30 of 36 brands
Leak 3Product page over 3 MB on a phone
75%27 of 36 brands
Leak 4No WhatsApp link on the product page
50%18 of 36 brands
Leak 5No star rating on the opening phone screen
41%14 of 34 brands
No buy button on the opening phone screen
35%12 of 34 brands
No WhatsApp, phone or email link on either page
28%10 of 36 brands
Pop up covering the opening phone screen
15%5 of 34 brands
These are all the checks the agent counted, sorted by how often they showed up. Here are the five that showed up most, one by one, with Arjun's story and a fix for each.
Up next Leak 1 showed up on 35 of 36 sites. It starts with one WhatsApp message on a Sunday night.
1. Does your site say when you will reply?
Arjun got a WhatsApp message on Sunday night. “Is this okay for diabetics?” He saw it on Monday at lunch. The buyer had already ordered from someone else.
What we found: 35 of 36 sites never said when a reply would come. Not on the home page. Not on the product page. Not near the chat button. And 10 of 36 showed no WhatsApp, phone or email link on either page at all.
Why it leaks: A buyer with a doubt is a buyer who is ready. They just need one answer. When they do not know if the answer comes in five minutes or two days, many leave to check another brand. A clear promise keeps them waiting for you.
Think of a shop with a sign that says “back in 10 minutes”. People wait. A shop with no sign and a closed shutter loses them.
The chat button
Before
Millet Masala Chips
150 g
Add to cartChat on WhatsAppNo reply time shownAfter
Millet Masala Chips
150 g
Add to cartChat on WhatsAppReplies in 1 hour, 10 AM to 8 PMDo this: Pick a reply time you can keep, like “we reply on WhatsApp within 1 hour, 10 AM to 8 PM”. Put it next to every chat button and on the product page. Then keep it. Our guide on how Indian D2C brands use AI agents, and what to keep human shows which replies a helper can draft and which a person should send.
Up next Buyers now know when Arjun replies. But the question they ask most is still three swipes down the page.
2. Can a buyer see delivery, COD and returns without scrolling?
Arjun's delivery promise was great. Two days in Pune, cash on delivery, easy returns. It sat in a tab called “Shipping info”, below the reviews.
What we found: On the opening phone screen of the product page, 31 of 34 brands showed no line about delivery time, cash on delivery or returns. The price was there. The photo was there. The answers to “when will it come” and “can I pay at the door” were not.
Why it leaks: In India, these are the questions that decide whether a stranger orders at all. A new buyer does not trust you yet. Cash on delivery and easy returns lower the risk. If they have to hunt for it, many will not hunt.
The opening phone screen
Before
Millet Masala Chips
150 g
Add to cartDelivery, COD and returns: 3 swipes down
After
Millet Masala Chips
4.6 · 1,200 reviews
150 g
Delivered in 2 to 4 days · COD available · 7 day easy return
Add to cartDo this: Add one short line right under the price. For example: “Delivered in 2 to 4 days. COD available. 7 day easy return.” Real text, not inside an image, so phones and search engines can read it.
Up next Delivery and COD sit under the price now. Then Arjun opens his own page in a cab, on mobile data, and counts to six.
3. How heavy is your product page on a phone?
Arjun opened his own product page on his phone, on mobile data, in a cab. He counted. He got to six before he could read the price.
What we found: 27 of 36 product pages weighed more than 3 MB on a phone. 30 of 36 home pages did too. The middle product page was about 3.9 MB and asked for nearly 400 files. One product page weighed 47 MB.
A quick note on how we measured. The weight is what the page downloaded on our test phone view. Load time on our test machine was about 5 seconds for the middle page, and that machine is faster than most Indian 4G, so real phones likely wait longer.
Why it leaks: Every extra file is another thing the phone must fetch before the buyer can act. The web.dev guide on page speed says a good Largest Contentful Paint, the moment the main content shows up, is 2.5 seconds or less. Big images, video banners, and many little add ons for chat, reviews and offers add up fast.
Every extra file is a small wait. Small waits add up to a lost buyer.
Do this: Open your product page on your phone on mobile data. Compress every image. Cut the add ons you do not use. Remove any auto playing video from the top of the page. Then test again.
Up next The page is lighter. A buyer has a quick question about spice. Where on the product page can they ask it?
4. Can a buyer ask on WhatsApp from the product page?
Arjun's WhatsApp button lived on the contact page. A buyer on the product page who wanted to ask “Is it spicy?” had to leave the product, find the contact page, then type the whole question.
What we found: 18 of 36 product pages had no WhatsApp link at all. Of the 19 brands that did use WhatsApp, 6 opened the chat with an empty box or just “Hi”, so the buyer had to explain which product they meant.
Why it leaks: Most Indian buyers are already on WhatsApp all day. One tap to ask is the shortest path from doubt to order. A link that also carries the product name saves your team a round of “which product, sir?”.
The same WhatsApp tap
Before
Hi
Hello! Which product is this about, sir? 2 hours later
After
Hi, I have a question about the Millet Masala Chips 150 g. Is it spicy?
Hi! Mildly spicy, kids enjoy it too. COD is available. Shall I send the link? 10 min later
Do this: Add a WhatsApp button on every product page. Make the link carry a ready message with the product name, like “Hi, I have a question about the Millet Masala Chips 150 g”. Your team sees the product before they say hello.
Up next One tap opens WhatsApp with the product name ready. One more thing: 1,200 happy reviews that almost nobody sees.
5. Can a buyer see your star rating on the opening screen?
Arjun had 1,200 happy reviews. They lived at the very bottom of the product page. Most buyers never scrolled that far.
What we found: 14 of 34 product pages showed no star rating or review count on the opening phone screen. The agent's proof score was the second lowest of all 10 areas, at 44 out of 100.
Why it leaks: A stranger looks for one quick sign that others bought and liked it. Stars and a count next to the product name do that in half a second. Reviews at the bottom are proof that nobody sees.
Do this: Show the star rating and the number of reviews right under the product name. Make it a tap that jumps to the reviews. If you are new and have few reviews, show what you do have honestly, like a short buyer quote.
Key takeawayEach of the five fixes is a few lines of text or a lighter image. None of them touches your ad budget.
Up next Five leaks found. But how many of them did the agent get right? This part matters as much as the leaks.
What did the agent get wrong?
This part matters as much as the leaks.
The agent flagged 27 of 36 sites for “no button on the opening phone screen”. The review step looked at every screenshot. The real number was 12 of 34. Many stores use styled links that look like buttons, and the agent counted real buttons and nothing else. It also marked some sticky bars as pop ups, and it missed delivery lines that sat inside a banner.
So we threw out every leak the screenshots did not show. That is the rule at The Human Algorithm. AI does the reading. A person does the judging. An agent report that nobody checks can send you fixing things that are not broken.
Agent vs review
“No button on the opening screen”
The agent counted real buttons and missed links styled to look like buttons. So every finding here was checked against what the screenshots show.
How did the 10 areas score on average?
Average score per area
Strong at showing the product. Weak when a buyer hesitates.
- Speed39
- Proof44
- Call to action59
- Follow up path62
- Opening screen67
- Tracking73
- Mobile85
- Trust93
- Offer95
- Capture98
Speed and proof were the weakest areas by far. Capture, offer and trust scored high, because almost every store has a cart, a clear price and policy pages. In plain words: Indian D2C stores are good at showing the product and bad at the moment a buyer hesitates.
What happened to Arjun?
Arjun spent one Saturday on the five fixes. He wrote “Replies in 1 hour, 10 AM to 8 PM” next to his WhatsApp button. He added one delivery and COD line under the price. His developer cut the product page from heavy to light by shrinking the photos and removing two unused add ons. He put a WhatsApp button on every product, with the product name ready in the message. He moved his stars up under the product name.
None of it touched his ads. The next week, the same ads sent people to a store that answered their doubts before they had to ask. His WhatsApp got more questions, and his brother answered them within the hour.
The biggest change was not in the ad. It was in the minute after the click.
Key takeawayThe biggest change was not in the ad. It was in the minute after the click.
Where should you start?
Start with leak 1 today. Decide who replies, and write the promise next to your chat button. It costs nothing.
Then do leaks 2 and 5 on your top selling product page. Both are a few lines of text. Leave leak 3, the heavy page, for a day with your developer. Add the WhatsApp button from leak 4 while they are in there.
The numbers in one place
| Leak | What | Brands | Share | How we know |
|---|---|---|---|---|
| 1 | No reply time anywhere on the two pages | 35 of 36 | 97% | Agent text scan |
| 2 | No delivery, COD or returns line on the opening phone screen | 31 of 34 | 91% | Screenshot review |
| 3 | Product page over 3 MB on a phone (home page: 30 of 36) | 27 of 36 | 75% | Measured download |
| 4 | No WhatsApp link on the product page | 18 of 36 | 50% | Agent link scan |
| 5 | No star rating on the opening phone screen | 14 of 34 | 41% | Screenshot review |
| Average funnel score | 72 / 100 | 57 to 83 | Agent score | |
| Middle product page weight | 3.9 MB | about 400 files | Measured download |
Our data, 23 October 2026. Anonymised data: download the CSV.
Score your own funnel on the 5 leaks
0/100
5 leak score
Tick what is true for your store today
Each tick updates your score and the funnel. Check on your phone, on mobile data. “On my laptop” is a no.
Useful? Send it to your team or keep the checklist.
Opens WhatsApp with the link ready. You choose who gets it. Save gives you a one page checklist to print or keep as a PDF.
What should you remember?
- Our agent read the phone pages of 36 Indian D2C brands on 23 October 2026. The average score was 72 out of 100.
- 35 of 36 never said when they would reply. Put a reply time next to every chat button.
- 31 of 34 hid delivery, COD and returns below the opening screen. Add one line under the price.
- 27 of 36 product pages weighed over 3 MB on a phone. Compress images and cut add ons.
- 18 of 36 had no WhatsApp link on the product page. Add one, with the product name in the message.
- 14 of 34 showed no star rating on the opening screen. Move it under the product name.
- The agent got some things wrong. Check every agent finding against what you can see.
How we ran this study
- Date: 23 October 2026.
- Who: 37 Indian D2C brands across skincare, food, grooming, fashion, home, wellness and more, a mix of large and young brands. One run did not finish, so 36 count. No brand is named.
- Pages: the home page and one product page that the home page links to.
- How: our Funnel Audit Agent opened each page on a 390 px phone view and a 1366 px laptop view, on a test machine outside India. It read public pages and nothing else, never logged in, never filled or sent a form, did no security testing and followed each site's robots.txt.
- Review: checks on the opening phone screen (rating, delivery line, buy button, pop up) were judged from the screenshots for 34 brands. Two were left out: one product page did not show its content to our phone view, and one site showed a store for another country.
- Limits: a single day and two pages per brand. Speed was measured on our machine, not on Indian 4G. Sites change often. Treat this as a snapshot.
- Data: download the anonymised data (CSV). Free to use with a link back to this page.
Sources
- Our data, 23 October 2026: Funnel Audit Agent run on 36 Indian D2C brands (anonymised CSV above).
- web.dev: Largest Contentful Paint (LCP) (VERIFIED, opened 23 October 2026: a good LCP is 2.5 seconds or less).
Arjun, his brand and his story are made up to show the method. The study numbers are real and come from our own run.
Questions people ask
What is an AI funnel audit?
It is a check of your website by an AI agent that reads your pages like a new buyer would. Ours opens the public pages on a phone and a laptop, scores 10 areas from 0 to 100 and lists the leaks it can see, with screenshots as proof.
What were the most common funnel leaks on Indian D2C sites?
In our run of 36 brands on 23 October 2026: no reply time (35 of 36), no delivery, COD or returns line on the opening phone screen (31 of 34), product pages over 3 MB (27 of 36), no WhatsApp link on the product page (18 of 36) and no star rating on the opening screen (14 of 34).
Is it safe to let an AI agent audit my website?
Ours reads public pages and nothing more, the same ones any visitor sees. It never logs in, never fills or sends a form and does no security testing. It also follows your robots.txt rules.
Can an AI agent replace a human funnel audit?
Not on its own. In our study the agent flagged 27 sites for a missing button, and the screenshots showed it was 12. The agent is fast at reading. A person should judge what it finds before you spend money fixing it.
How heavy should a product page be on a phone?
There is no single rule, but lighter is better. In our run, the middle product page was about 3.9 MB. The web.dev guide says the main content should appear within 2.5 seconds. Compress images and remove add ons you do not use.
How can I check my own store for these leaks?
Use the 5 leak self check on this page. It takes two minutes. For a full check, the free Site Scan runs the same agent on your public pages, and Madhur reviews every report before it reaches you.
Know a founder who checks their store on a laptop? Send them Arjun’s story.
Opens WhatsApp with the link ready. You choose who gets it.
Want this handled for you? THA Agent Desk: AI agents set up for your business, with a person in charge.
THA Circle
Use AI. Still sound human.
A free WhatsApp channel for founders, coaches and creators who want AI to save time without losing the human touch.
- On joining A welcome pack: 3 AI prompts and 5 chat scripts you can copy today.
- Monday A short script for your own chats. Ten minutes to try.
- Wednesday One real chat taken apart, names removed.
- Friday Human of the Week. A moment a customer felt seen.
Free. Takes ten seconds. Channel followers stay private. Leave any time.
The Sunday note
How people pay attention, and how to earn it. One email a week.
I study AI, attention, human psychology and marketing. Every Sunday I send one idea and one script for your chats. Five minutes to read. Reply any time, I read every answer.
Get the Sunday noteFree. Leave any time. Your email stays private.
Get your brand found in AI answers
How D2C brands use AI agents, and what to keep human
A sales funnel for coaches in India