Your Ads Aren't the Problem, Your Store Is: Shray Arora on Why D2C Brands Are Leaking Revenue at Checkout

There's a conversation happening in every D2C war room right now. CAC is up, margins are tight, and the founder is asking the same question: where do we cut, or where do we double down? The answer most people arrive at is the same playbook. Get more traffic, improve your creatives, target smarter. Honestly, it just needs more money.

Shray Arora thinks that entire conversation is happening in the wrong room. The leak isn't in performance marketing. It's not in the ads. It's what happens after the click, on the store itself.

Shray is the co-founder and CEO of Helium, a Genai platform that personalizes D2C storefronts in real time, visitor by visitor, session by session. He spent time at McKinsey, then Indiefy, then was part of the early founding team at Appdrew before starting Helium a couple of years ago. He raised a pre-seed round, landed on the Forbes 30 Under 30 Asia list this year, and counts Noise, W for Woman, and Marks & Spencer among Helium's customers.

Today we're getting into how D2C brands actually optimize their stores, why "personalization" has become one of the most abused words in ecommerce, and how a sari brand ended up creating 2,000 variations of the same collection page.

Key Takeaways: Why the Store, Not the Ad, Is Where D2C Brands Actually Lose Money

The Real Leak in the Funnel:

• 60% of ecom platforms obsess over ROAS, 40% talk about conversion rate, and almost nobody looks past those two metrics
• Meta has effectively won distribution and targeting. Trying to out-algorithm Meta is a losing game. The unexplored territory is what happens after the click
• A brand spending 50 lakh rupees a month typically sits between 1.5 and 2.5 crore in monthly revenue, meaning 5-8 lakh rupees a day rides on whether the landing experience actually converts

Personalization Is Not Recommendations:

• Most brands think "recently viewed" widgets or two variants of a product page count as personalization. Shray calls this a shallow, if-else version of the real thing
• Helium tracks correlation between on-site interactions and purchase, not just add-to-cart. For one beauty brand, visitors who read reviews convert 30% higher than average
• A single sari brand collection page gets close to 2,000 live variations based on persona and geography, generated programmatically rather than manually

Building a Data Company Before an AI Company:

• Helium's own playbook shifted after a feature that worked brilliantly for one brand completely flopped for another with a nearly identical catalog
• The fix wasn't a better feature. It was realizing they were "throwing darts in a dark room" without full-funnel data, from impression to third purchase
• External signals matter as much as on-site ones. IPL match nights change purchase timing. Gas stove shortages triggered a 3x revenue spike for induction stove brands that Helium's anomaly engine caught before the brand even noticed

Q: What made you decide to leave the corporate track and build in ecommerce?

Shray Arora: It's interesting, it was never one moment. Before I even started my corporate journey I was very sure I love building products. While in college I was dabbling with a few ideas, in fact took a shot at building the Roku for India. This was back in 2016. We thought smart TV penetration wasn't high yet, but people buy TVs for 5 to 10 years and the world changes every 2 years, so there should be a portable device you hang on to instead.

Long story short, that itch of building something meaningful for people started back in college. I took those learnings, went to McKinsey, then Indiefy where I was chief of staff when the company was 400 people, then Appdrew where I was part of the early founding team, maybe 5 people. So I naturally got exposed to systems and process at McKinsey and then what a five-member team could achieve. I always knew I wanted to come back and build a product of my own, and ecom is where I got exposed to a lot of inefficiencies in terms of where tech is and where the industry is.

🔥 ChaiNet's Hot Take: Going from McKinsey's scale to a five-person startup isn't a step down, it's a masterclass in seeing the same problem at every possible headcount.

Q: Everyone talks about ROAS and CAC. How did you land on the store itself as the real problem?

Shray Arora: If you look at 100 platforms, a good split of 60% are talking about improving ROAS and 40% are talking about conversion rate. There's maybe 1-2% talking about anything else. Here's Meta, doing their best job, and I genuinely believe there's no programmatic ad platform that's done better in distribution and algorithms. Any SaaS tool has a certain longevity, but eventually Meta makes a lot of that irrelevant.

So the territory of understanding what you're targeting on Meta, what you're targeting on Google, what your brand stands for, and using that to enrich the landing experience felt relatively unexplored. We didn't consciously choose it either, honestly. We were developers and product people. We didn't know how to do Meta ads. We knew we could build better websites. So naturally, we picked our side.

🔥 ChaiNet's Hot Take: Meta already won the ad game. The next war isn't for attention, it's for what happens in the three seconds after someone lands.

Q: How did you land your first customer?

Shray Arora: People ask this a lot and there's no clean answer, just my anecdotal story. It was cold outreach, and interestingly our first customer was actually a US brand even though the US isn't a huge chunk of our business today. We thought, who's actually aware their website speed is bad? Can I find that out and pitch them a fix?

So we built a small page speed analysis tool. I'd cold email people saying here are three things wrong with your website, remove these two apps, improve this first fold image, and this is how you do it, here's a full report, do you want to talk? No payment ask, no signup friction, just value up front. This is actually how we landed Noise later too, not as a first customer, but the same outreach strategy worked.

🔥 ChaiNet's Hot Take: Asking for payment is friction. Asking for a signup is friction. A free, specific, embarrassingly useful report is not.

Q: What can actually go wrong on a Shopify store beyond page speed?

Shray Arora: There's no one-size-fits-all, but let me paint anecdotal pictures. If a brand is spending 50 lakh rupees a month, they're typically sitting between 1.5 and 2.5 crore in monthly revenue, so anywhere from 5 to 8 lakh rupees a day. I bucket brands into large catalog and small catalog, and each has a completely different failure mode.

Small catalog brands are usually catering to multiple personas through the same product. Take a leg massager, you're selling to someone tired after office work, but also an athlete, also an elderly person. The problem there is contextualization and storytelling. There's a brand operating at a 300-400 rupee average order value, doing seven-figure daily sales purely through Meta, which by any D2C playbook shouldn't work. It works because their hook and storytelling are on point.

Large catalog is a completely different game. Think of it like a stock market. Are you optimizing spend on the right products? Winning products change every season, every geography, every price point.

🔥 ChaiNet's Hot Take: Small catalog brands need a better story. Large catalog brands need a better stock market algorithm. Selling both the same playbook is why most generic CRO tools underperform.

Q: You've spoken about tracking "signals" most people never even think about. What do you mean?

Shray Arora: Think of it like a store analogy. Someone walks in, what does a good salesperson notice? You're asking about the price point, you're comparing two products, you're asking qualified questions. Online, all you have are interactions, no question-answering mechanism. So can you find correlation between every interaction and eventual success, meaning purchase?

For a beauty brand, if someone is going through reviews, their likelihood of conversion is 30% higher than every other visitor. Right now most of ecommerce operates on three events: add to cart, time on page, and purchase. What we look at is everything happening on the site and what correlation each interaction has to a purchase. Add to cart might be 50% correlated. Reviews, 30%. Search, 15%.

🔥 ChaiNet's Hot Take: Add to cart is the loudest signal on your site. It's not the only one, and it's definitely not the earliest one.

Q: Do you track people across websites to build these personas?

Shray Arora: People ask us that, but why would we want to do that? It's interesting, we internally ask our team, how do you buy sunscreen? Almost everybody is clueless about what parameters they use to buy sunscreen. But almost everybody knows exactly how they buy protein powder. Every category has its own nuance of how it's purchased and how different personas interact when buying it. So we build that differently, category by category, not by chasing someone's browsing history across the internet.

🔥 ChaiNet's Hot Take: You don't need someone's entire internet history. You need to know how people shop for your specific category, because sunscreen buyers and protein powder buyers are not the same species.

Q: "Personalization" is a word every SaaS tool claims. What's actually made it hard to sell?

Shray Arora: Everybody has a different notion of what personalization means. Most people think recommendations are personalization. If I show the right products on your product page, that's personalization to most people. It's a very, very abused word. I could bake in handpicked recommendations, or make two variations of a product page, and call it personalization.

The second issue is that people want control, but the moment you seek control you're stuck in a shallow world of personalization. If a user is a returning visitor, let's show them recently viewed, that's the typical if-else logic. I don't even want to do that at Helium because that's a software engineering problem, not a data science problem. If people in Delhi like floral patterns for an apparel brand and people in Chennai like darker shades, from tracking that insight to actually making it live takes us a couple of minutes. The brand doesn't know what happened, they just feel a certain product isn't supposed to be on top.

🔥 ChaiNet's Hot Take: True personalization is invisible by design. If you can point to the if-else logic behind it, it's not personalization, it's a spreadsheet with extra steps.

Q: How do you get brands comfortable with not understanding exactly how it works?

Shray Arora: Now we've realized we don't need to explain how it happens. It's like explaining how a transformer works, most agentic companies, even I don't know the entirety of it. All I know is I can use OpenAI's APIs and build on top. I just need to show them the numbers and the outcome, that's all that matters. It took a lot of time for us, coming from engineering backgrounds, to stop explaining the tech and start selling the outcome. We were fascinated by our own engineering and kept trying to sell the feature instead of the result.

🔥 ChaiNet's Hot Take: You don't need to understand a car engine to trust it'll get you from A to B. Brands don't need to understand your model, they need to see the revenue line move.

Q: Are you a business person who learned tech, or a techie who was forced into business?

Shray Arora: I still don't feel like a business guy. I think I'm a product guy who had to become both a business guy and a techie. There are a lot of archetypes within product people. Some are very sharp at articulating exactly what a user would want. I don't think I'm that. I'm the one who ventures into territory that doesn't exist yet, figures out what needs to be done, and then makes his mark. So at heart I'm a product guy, but if you want to be an entrepreneur, there's no shying away from also becoming both. I know enough tech to not be fooled by engineers, and I think that's enough.

🔥 ChaiNet's Hot Take: Marketing people overcommit, engineers undercommit. Founders don't need to master either craft, they just need enough fluency to catch both kinds of nonsense.

Q: Walk me through the Sudhati Sarees case study. What was breaking, and what changed?

Shray Arora: They were using an alternative of ours before, and honestly all D2C enablers overlap, you could pick a hundred tools and find overlap across all of them. Their core problem was 60% of traffic landing on collection pages, and those collection pages had 200-250 products across 20 collections. It was nearly impossible to figure out, daily or weekly, which products were actually selling well and prioritize them. Some days engagement was great, some days it tanked, and we didn't know why.

We treated it the way Amazon or Myntra treat search real estate. You penalize products with low clickthrough and conversion, and push up products lower in the listing that actually convert well once they get impressions, programmatically, not manually.

🔥 ChaiNet's Hot Take: Your collection page is prime real estate you're renting out for free to your worst-performing products. Start charging rent.

Q: You mentioned geography changes everything too. What did you learn building this for Indian shoppers specifically?

Shray Arora: Every region has its own nuance of how they purchase a sari. Delhi is going through peak summer while Bombay is going through humidity, so the fabric Delhi buys is completely different from what Bombay buys. What's being celebrated is different too, there's Buddha Purnima happening in this window, but also Mother's Day. Can we understand that geographical nuance and bake it into product ordering, subtly, without calling it out loud? At any given point we create close to 2,000 variations of the same collection based on persona and geography.

🔥 ChaiNet's Hot Take: A sari isn't a sari. It's a humidity forecast, a festival calendar, and a fabric preference all disguised as one product listing.

Q: You mentioned a feature that worked for one brand and completely failed for a near-identical one. What happened?

Shray Arora: We built the same collection page feature for another brand with a very similar catalog, close to 500 products, and it just didn't work. We were confused. Turns out only 10% of their traffic was landing on collection pages, because their ads were DPA, dynamic product ads, sending people straight to product pages instead. These are the anecdotal examples we learned from incrementally.

What we realized is if you're not grounded in data, if you don't have the full picture from impression to third or fourth purchase, you're always throwing darts in a dark room. Some stick, some don't, and I'm a decent dart thrower so I'd land more than most, but I'd still keep missing. That's when we transitioned from a feature company to a data company that happens to ship features.

🔥 ChaiNet's Hot Take: A great feature built on incomplete data is just a lucky guess wearing a nice UI.

Q: What's an example of an external event you've had to model into the system?

Shray Arora: We built anomaly engines that alert brands when something's steeply risen or fallen. With a couple of brands, we didn't initially realize their revenue shot up 3x because of a gas stove shortage pushing people to induction stoves. Or, IPL match nights, post 8pm purchases drop because people are watching, and for men's brands specifically, ads on Hotstar start performing better than ads on Meta because viewership shifted there. So CPMs go up too, because the same number of brands are competing for fewer available people. Can we correlate that IPL is happening with why CPMs spiked? That's where more and more data signals come in.

🔥 ChaiNet's Hot Take: Your CPM spike isn't a Meta problem. It might just be a cricket match.

Q: What's your advice to people building in the D2C AI space right now?

Shray Arora: Find an anchor problem and build with design partners. If you can start a brand of your own and figure out the problems you personally face, that would be the most beautiful journey, because you'd pinpointedly know what you're solving for. Or work inside a brand solving growth problems, then scale the repeatable systems out. With AI, there's going to be a lot of consolidation.

There's a clear choice you have to make: build shallowly but broad, or build deeply and vertical. Any good AI tool right now is essentially a wrapper on top of ChatGPT or Semrush-style data. Could you find really deep use cases and solve them, that's my bias, though plenty of entrepreneurs have made insane value going broad and shallow instead. Choose one, figure out where you want to go, and build design partners to find and solve the right problem. Empathize with the D2C person, the brand, and you'll potentially build something meaningful.

🔥 ChaiNet's Hot Take: Shallow and broad, or deep and narrow, just don't build medium and mediocre.

Final Thoughts: The Store Was the Leak All Along

Shray's closing perspective: "If you're not grounded on data, you'll always be throwing darts in a dark room. Some would stick, some wouldn't, and I'm a great dart thrower, so I'd land more. But I'd still keep missing a lot of those."

The bottom line: Every D2C founder has been trained to look at the same two dashboards, ROAS and conversion rate, and assume the fix always lives upstream in the ad account. Shray's core insight flips that. The ad platforms, Meta especially, have already gotten frighteningly good at their job. The unexplored, undervalued territory is everything that happens in the seconds after someone lands on your store.

What's striking about Helium's journey is how much of it was built by unlearning their own instincts. Engineers who wanted to explain the transformer instead of the outcome. A feature that worked brilliantly for one sari brand and flopped for a near-identical one, until they realized the real gap was data, not features. A personalization category so abused that half the job was just defining what it actually means.

For founders building in this space, the takeaway isn't "add more personalization." It's go find the anchor problem nobody else is grounding in real data, whether that's a collection page, a persona mismatch, or a CPM spike nobody bothered to explain. Build it deep, and let the outcome do the talking.

Q: How can people connect with you and learn more about Helium?

Shray Arora: If you want to see what Helium can do for your store, head to gethelium.com, that's the best place to reach out and see it in action.

Final words: The next time your ROAS dips and the instinct is to pour more into creative testing, pause and ask a different question: what's actually happening after the click? The leak might not be in your funnel's top half at all.

---