Our mission is to democratise e-commerce, says Meesho founder & CTO

To understand the tech infrastructure, one also needs to understand the consumer category we are tapping into, says Barnwal

Softbank-backed Meesho wants to democratise e-commerce. Its biggest user base is in Bharat, or smaller cities that have yet to experience the power of the internet. SANJEEV BARNWAL, the company’s co-founder and chief technology officer, in an interview with Shivani Shinde, talks about how tech is powering Meesho’s growth and  what the founders are doing to reimagine e-commerce for Bharat. Edited excerpts:

Your website says you are reimagining e-commerce for Bharat. How have you reimagined your tech infrastructure?

To understand the tech infrastructure, one also needs to understand the consumer category we are tapping into. A majority of users are experiencing online buying for the first time. They may have used Facebook and WhatsApp and YouTube, but may not be comfortable buying online. The second aspect is suppliers, who are selling online for the first time, and they too have unique issues which do not exist for brands. The third aspect is that the devices used in these places are mostly low-end Android devices with unstable internet. The question then is, how do we cater to this consumer set?

We figured that the user experience (UX) has to be very simple, to the extent that they would want to go deeper into multiple levels, which also means we need to figure out what they need, and accordingly personalise it. We notice that this user base is moving from intent buying to discovery-led buying. Hence, understanding this intent becomes very important.

What are the challenges when building in this scenario? And, how do you convince people to download?

For instance, while we can keep the UX simple, it is useful only if the app gets downloaded. That was the first challenge, getting people to download the app, and I think the teams did a fabulous job of keeping the compressed app size low. It is the lowest across all e-commerce platforms.

The other reason to keep the app size small was our mission to democratise internet commerce, which also means that we should be able to cater to older versions of Chrome OS, which are still used among this group of buyers. Also, a simplified UX starts with a very simple interface that makes it easier to navigate the platform. But then, how do we mix visuals with text in a way that makes it easier for people who are experiencing e-commerce?

If you want to keep your APK size small, it has downsides, too -- for example, you cannot rely on cross-platform frameworks such as React Native, or even Flutter, promoted by Google, since they add to the APK size. This also means it adds performance overloads. The other focus area was caching, because network connectivity is low and unreliable. One needs to leverage caching to some extent, to mitigate these issues, and other optimisation tools and solutions. As for convincing people to use Meesho, initially it was about communicating the value proposition, but now we are a well-known brand.

How have you leveraged technologies like artificial intelligence and machine learning? Can you give some instances?

We are a data-driven company. We have invested heavily in capturing behavioural insights, keeping in mind that user privacy is intact. Our data journey started 1.5 years back, and it took us time to get where we are today because of the kind of skew that we have in our business, which has more unbranded products. You will get a lot of orders on very few products, unlike in the branded e-commerce space. It took us about three months to get to the version of personalisation that gave a significant impact. When we went in for personalisation, our order per user went up by 50 per cent for a large cohort. For this to happen, we tinkered with a lot of algorithms, the common one being collaborative filtering.

Another problem we faced in personalisation was that the data we received was not structured. We relied on deep learning techniques to get more information, which makes personalisation data more reliable.

On the supply-side front, RTO or return to origin was a big issue. Addresses provided by users are not always concise, but very generic in nature. Predicting RTO is important, so we use a data science combination of user history and address. We looked at the address and ran it through a data science system that tries to predict that a particular order has a higher probability of RTO.

On the supplier side, there is no right way to price an unbranded product. So, we have a price recommendation algorithm that looks at the product, image, and product details, and then comes up with a price range. Even finding the right product to sell online is something we suggest, so that suppliers can invest accordingly.

How big is your team, and are you scaling up?

Our engineering team is over 350 this year, and we want to more than double it. The data science team, which we are investing heavily in right now, is a 25-member team, which we want to take to 50 this year-end. It is a difficult market to hire in, but while getting the numbers is a focus, getting the right talent and a person’s ability to fit into our culture is also important.

What would you recommend to startups who want to build for Bharat?

First, the traditional mindset does not work, so do not carry it here. For a tier-1 user who is tech savvy there are a lot of options to discover features easily, but for our user base the UX needs to be extremely simple. The functionality needs to be upfront. Second, sensitise your team about the market you are going after. If a team sitting in a city is making an app for users in smaller cities, they may be testing it on a Rs 30,000 smartphone, but your user base may still be using phones in the Rs 3000-4000 range. Third, connectivity is always challenging -- you’ll have to figure out a way to work with it.

What are your three big concerns?

The percentage of users who are starting to use any app for the first time is increasing as we exhaust the pool that is already active on social media, or is big on the app economy. This newer demographic has smartphones but doesn’t use any app. Further simplifying e-commerce becomes a priority. Our teams will have to work even harder and be more innovative in helping them change this behaviour. This year, our target is to get to 100 million monthly transacting users. A lot of them will be first-time app users.

Second, we have a huge supplier base (400,000) and we have done a lot to make them succeed, but not everyone is successful. Figuring out an automated mechanism to help suppliers become successful platforms is the second-biggest challenge.

The third challenge is scale. We get 200,000 requests per second at the peak, and given the kind of growth target we have set for ourselves, that number will grow by 4-5x this year alone. To be able to cater to this we will need a robust tech infrastructure to absorb this scale, which is going to be interesting to manage.

Three tech focus areas for Meesho?

For this year, personalisation will be a big focus area, especially real-time personalisation. We will be doubling down on it. Second, leveraging data science to optimise logistics. Investing more in logistics optimisation, experience optimisation, and recommendation algorithms will be the focus. On the supplier side, we will help them on better price discovery.

How critical is it for companies, especially who are building for Bharat, to leverage data?

The access to computing power is very easy, and cheap. Leveraging data is extremely important, and as you go deeper into the country to smaller cities, expecting users to tell you what they want is difficult, but it is even harder for you to get access to know these problems at this level. This user also needs a behavioural change, and when we are trying to achieve this, you cannot just rely on what they say.

If you want to go deeper into how they are using your product, which screen are they getting locked, why are your NPS low or high, those answers lie in the ability to collect user data from day one, but also invest in team capabilities to easily get insights.

A lot of players that are targeting this segment are using voice search and translation tools. What is your experience of dealing with voice?

For now, we have limited voice tech to search use-cases, and we support six to eight languages. We are investing more in this direction. We see a good adoption for this and it’s effective for search. But when using voice for navigation, you have to understand the context to search, and that is difficult when it comes to the vernacular. We do not use voice for navigation, but for search.

Do you see Metaverse being used by e-commerce?

It’s an interesting concept. I definitely see a lot of use-cases. The current way we shop online is different from the way we shop offline. I think Metaverse can bridge that gap.  For instance, shopping for many is a fun experience. You may go with friends and have fun. If you can simulate that group buying experience together, maybe online shopping will change. Metaverse will definitely happen.