Why did Google lead a $12-million, or about Rs 78-crore, investment round in Dunzo, a hyperlocal, concierge services startup from Bengaluru? Three months after the deal, that’s a question that still has many in India’s startup and venture investment circles scratching their heads for an answer.

You are unlikely to know of Dunzo unless you live in the Karnataka capital or have read this story from about 18 months ago. Started over a small Whatsapp group to fulfill random tasks for a small fee, Dunzo has been quick to get a fan following with customers ‘dunzoing’ cigarettes, sanitary pads, groceries, pet supplies, and even fruits and beer. Or, asking for a forgotten phone charger delivered to office from home, send homemade gulab jamuns to a friend, get a document clicked and have the image sent over WhatsApp, and even get notebooks covered in brown paper ahead of a school year.

Hyperlocal fulfilment, as Dunzo’s business is called, is not a new customer pain point that startups have attempted to address. Several did since 2014, but mostly unsuccessfully. But Dunzo’s doggedness and following among a relatively small, yet deeply loyal, set of customers attracted investment by Google, making it the first Indian company to be funded directly from the search giant’s balance sheet. (Blume Ventures, one of FactorDaily‘s investors, is an investor in Dunzo.) Google’s earlier investments in India in realty platform Commonfloor from Bengaluru and Chennai CRM company Freshworks, were made by its investment arm Google Capital (now Capital G).

But, still. Google is all about scale. About 65,000 internet searches are made on Google every second. Why would it be interested in a company that does 3,500 to 4,000 transactions a day and until the December funding was struggling to rein in costs? “We have reached a good number of daily orders,” says Kabeer Biswas, Dunzo co-founder and CEO, adding 70% of tasks are purchases and 25% pick-ups. His ambition: by end of 2020, rack up one million transactions a day across 14 cities with 6.5 million monthly active customers.

Context is everything in a fast-growing business like internet services. Just to add that dose of perspective to Dunzo’s ambitions, ecommerce giant Amazon records about 350,000 transactions a day in India, food delivery company Swiggy records some 140,000 deliveries daily here, and e-grocer BigBasket has crossed 50,000 orders-a-day mark.

Dunzo’s will need will to grow manifold – 285 times, to be precise – between now and 2020 if it’s to keep its deliver by its sales spiel.

Biswas insists that there’s as much in it for Google as there’s for Dunzo in the deal. “It’s a signal on both sides,” he says. “We use a lot of their products and drive growth and signal for ourselves and we can build a lot of product data at our end which makes them better.”

The possibility and scope of that data is limitless. The fact that Dunzo does not limit itself to verticals opens it to almost everything under the sun.

Biswas coaxes you to imagine a world of search plus delivery. “If you were to build a search function in the fulfilment world where it’s possible to get to you anything in 30 to 45 minutes, we are what local search would look like.”

Complexity delivers simplicity

Dunzo’s popularity comes from two things: its simple promise of running any task and an user experience while fulfilling the task that has a customer repeating transactions.

Getting this user experience right through a mobile app that automates the request is a humongous task before the Bengaluru startup.

“Our platform invokes a series of actions from the moment the user comes on the platform and clicks the plus button to initiate the task,” says Biswas.

Mukund Jha, the man behind the technology layer at Dunzo, walks us through these actions.

When a user comes on the platform and enters a request, the tech engine that Jha’s team has been building whirs into action to get data from three different search indexes that try to understand the user intent and request fulfilment.

Like Google’s search algorithm, Dunzo’s local search is based on crawling and indexing. Crawling is a software-invoked process of discovering content on publicly available web pages. Indexing is organising that content and keywords like an index at the back of a book for easy retrieval and reference. The Google Search index contains hundreds of billions of web pages and is well over 100,000 terabytes in size. Dunzo has over hundreds of gigabytes of similar data for localised search from its operations in Bengaluru alone. Each new transaction for Dunzo adds about a few tens of megabytes to these indexes.

Here’s how the index-based search works for Dunzo: let’s say a user needs a tube of burns cream Burnol. The keyword ‘Burnol’ is run through three indexes: products, stores, and tasks. The product index that has data of products at partner stores and products that have been delivered in the past helps categorise the new request into one of the buckets – Burnol matches the medicine bucket.

Once the keyword has been classified into the appropriate bucket, the stores index throws up partner stores from the medicine bucket that could potentially stock the Burnol. The third index based on transactions data helps identify previous such transactions.

Transactions data helps the platform pick appropriate riders for the task: someone who has done a similar task before is better equipped to fulfil the task. Once a rider is assigned, the task goes offline and becomes tricky.

Juggling rider schedules

One of the main reasons hyperlocal startups have gone belly up is their inability to keep their rider fleet busy enough to be able to sustain their salaries and remain profitable.

“There’s a simple math to why so many who attempted to crack the hyperlocal logistics model failed,” says the founder of a B2B hyperlocal logistics startup in Bengaluru, who asked to remain anonymous.

A delivery boy or a partner rider in hyperlocal lingo is paid around Rs 600 a day on an average. In the case of the most optimized hyperlocal use case of food delivery, every order earns the company around Rs 50 to Rs 60 from the merchant’s side. So, ideally, every rider needs to fulfil at least 10 orders in a day to justify his cost to company. “This model is still easier to implement in small cohorts or pockets of the city where the order density is more and distances covered to fulfil orders are less,” says the founder. “But it gets more and more complex and difficult to sustain as you scale business in more cities.”

Dunzo that is on track to expand to more cities, besides Bengaluru and Pune (where it introduced services in December, last year) is aware of this problem. Its approach involves using dynamic pricing to determine the value of each task to be able to justify the effort and cost of the delivery executive.

Some tasks are high value tasks like buying an iPad from a Croma store. Some others like buying a list of groceries is complex, while fetching a box of cigarettes is relatively easier. The platform determines whether to allot a task to a novice, an expert or an advanced delivery executive to the task per its requirement, Jha explains.

So essentially, the end user is charged not just on the basis of distance, but the entire effort of the task. If a skilled (and thus highly paid) delivery executive is assigned to the task, the user is charged more and vice versa. If a user places a request during low traffic hours of 2 pm to 6 pm, for instance, a lower band of delivery pricing kicks in.

Jha explains that to scale this model in an automated and efficient fashion, it is important to predict supply and demand. Demand side involves knowing its customers and their preferences better and supply side requires to know its merchant partners, their stocking patterns and inventory details better.

Load balancing

Every task at Dunzo collects data on both of the users and the merchants.

Each request from a user creates data relating to the user’s preference and consumption patterns. The platform uses this data to create a personalisation layer and bucket users into cohorts, based on their order history. Biswas says that the app will soon be able to recommend tasks in the form of a tap menu with 80% accuracy.

For example, if you are a user who asks for laundry-related help on Fridays, you will receive timely reminders and recommendations on the app.

Dunzo also plans to automate its user interface to solve a major problem of reducing interaction with its chat interface. Essentially, the app will have options or buttons to modify an order, change location, check estimated time of delivery, chat directly with the rider, and make payments, among several other options. This, the company believes, will increase task efficiency and transparency for the user.

Its real test will be on the merchants side. The company has an ambitious plan of tracking real-time data of all its merchants and items at merchant stores. That is a very large ground to cover. Real-time tracking of inventory is not going to be easy given the operational challenges of tracking inventory at every grocery store, every medical shop, every food joint, to name a few of Dunzo’s use cases. India is estimated to have some 50 million informal stores (mostly, kirana or mom-and-pop stores) and 220 million people dependent on them.

“It’s a very difficult problem to solve. The market is highly fragmented and currently there’s no good integration to point of sale devices. Indian product data on the internet is very thin. Google has ad data but that won’t help us. At the local level we have to go down and do it ourselves,” says Jha.

Dunzo is currently working on “standardisation of product data” starting with its user and rider generated data to build large SKU catalogues. SKU is short for stock keeping unit, a unit of inventory. Every item in a store is identified by its unique SKU to enable tracking of inventory.

God is in the detail

Jha says that Dunzo’s historical transactions have generated a lot of data to be able to map what kind of items are available at what store, pricing of these items, store timings, the items more in demand, what items get sold out, among others. Its team is now trying to build a machine learning model on top of this data to recognise patterns such as what and when does a store stack new inventory or what days a store remains closed.

There’s also a level of physical verification where, Dunzo riders collect local information offline like store timings, item availability, suggested alternatives for a product, approximate cost of items etc. while fulfilling orders.

Jha says the second and more exciting part of the problem will be real-time tracking of merchant inventory through integrations with stores. Direct integration of merchant inventory on the Dunzo backend will enable a real-time view of available products for riders and consumers. “In turn, we will be able to provide them interesting insights on high moving products, products doing well in a certain neighborhood that can held predict demand,” says Jha.

“Dunzo is generally the first one to pick trends,” boasts Jha, adding that would be an incentive for merchants to allow integrations. Like Dunzo was able to quickly predict from its data that iPhone 8 was not going to be as big as iPhone 7, says Biswas. Or, the Japanese store Muji showing up as suddenly popular in the Karnataka capital on Dunzo’s heatmap.

Deeper integration into the merchant stack is the next big thing on Dunzo’s plate, he adds.

This will be Dunzo’s secret sauce, says a retail expert. “Demand pattern predictability is the most critical variable in determining whether this model can scale in a meaningful way,” says Debashish Mukherjee, Partner & Head of Consumer & Retail Industries, India at consultancy A T Kearney.

He argues that even the large ecommerce players in India are competing in the home-delivery concierge space where deliveries can happen within 24 hours at least in large Indian cities.

For a model like Dunzo to scale, it needs a sharper proposition of what a buyer cannot get in 24 hours from large ecommerce companies in India. “That can be a potential white space,” says Mukherjee.

The Google connection

An immediate focus at Dunzo is to keep its 3,000 runners busy. This translates into more revenue for runners and business for the company. To achieve this, Dunzo uses multiple Google APIs, the most critical of which is the maps API. (APIs are short for application programming interfaces or code that allows apps or services interface and work with an operating system or service or another app.)

The Google Maps API helps Dunzo in efficient routing and, in turn, faster fulfillment of tasks. The Distance Matrix API helps determine the distance between the start and end points and offers directions to help riders plan efficient routes while fulfilling a series of orders at different locations. The Google Places API helps a rider with contextual information creating location awareness. For example: a store he is heading to will close in 30 minutes.

If a runner is at a moderately busy area like BTM Layout in Bengaluru at, say, 3.30 pm and is idle, the platform finds a location where the demand of requests is higher – and he is directed to the busier neighboring locality of Koramangala where the chances of getting an order are higher.

“This is something Uber had to build on its own. But Google has an insane understanding of distances and paths – both features make us very efficient,” says Biswas, the Dunzo CEO.

The long-term goal for the company is to have an in-built integration with Google Maps.

Today, if you look up Hamleys toy store on Google, its maps will tell you the closest one is about 40 minutes away. It also tells you that there is an Uber cab that can take you there in 45 minutes. Dunzo wants to take that experience up: Wait a second, it wants to tell you. “Why are you going? We can get you the toy you want in 40 minutes.”

“That’s the next step and a major growth feature for us,” says Biswas. “We are talking to have an integration in the map.”

With Google’s investment Dunzo, plans to build a stronger technology platform that can reduce the task time to 30 minutes. “This model can make or break if we can get the task time to below 30 minutes,” says Biswas.

Currently the task time is around 44 minutes per task with an average of 14 minutes being spent at merchant store.

Dunzo bike-rides?

Dunzo is planning to hire across fields including, data science, mobile development and the payment space and increase its technology team from 16 to 25 people in the next six months.

In the next quarter or so, the company also expects an additional revenue stream coming from sampling for consumers goods and motorbike-taxis that it plans to launch soon. Details are not available.

Google also recently announced its investment in Indonesian ride-hailing service Go-Jek. A play on the word ‘ojek’ in Bahasa for motorbike taxis, Go-Jek began as a ride-hailing app but today also delivers everything from meals and groceries to home services including, cleaners and hairdressers across Jakarta through a mobile app.

“By investing in local companies, building locally relevant products and training local talent, we hope to see more amazing local champions like GO-JEK emerge in Indonesia,” Google writes in a blog about its Go-Jek investment. Google India executives declined to be interviewed for this story.

At least one Indian internet expert believes the Dunzo investment signals a shift in Google’s strategy in growth markets – towards commerce. “Dunzo has a unique business model for India; it is solving a local problem. It address two things: India’s huge infrastructure problem and younger people who give more of a premium on their time,” says K Vaitheeswaran, founder of Indiaplaza, one of India’s early ecommerce ventures and recently author of Failing to Succeed: The Story of India’s First Ecommerce Company.

It is instructive to point out here again that the Dunzo investment is directly from the Google balance sheet and not from an investment arm. In other words, Dunzo’s future is aligned closer to the search giant’s strategy than some bean counter’s return on investments calculator. And, that India is already the world’s second largest internet market by users.

Google’s products – the world’s largest search engine, a maps platform, a user generated video platform, a voice-based phone assistant, a high-speed browser, a market monopolizing mobile operating system, and a video-enabled communication app – are proof enough of the company’s vision to aggregate and organise the world’s information for easy use.

Its interest in a tiny company like Dunzo indicates a larger hunger of aggregating hyperlocal, transactional information in markets that hold promise for its Next Billion Initiative flywheel. Can Google ‘dunzo’ itself on that ambition?

---