What’s the difference between a delivery in 30 minutes and one in less than 10 minutes? Not much. Even as Zomato claimed recently that it’ll launch a 10-minute food delivery, the internet wasn’t convinced. The company fielded mass outrage online, with people questioning the ethics, practice and security of the delivery fleet and the restaurant workers who will be pushed further to make the 10-minute delivery happen. 

But, what didn’t get noticed enough was the relative opacity of the operations behind the proposed plan, and how it’s all supposed to come together. This is crucial because even Zomato founder Deepinder Goyal on Twitter said that it will be possible through “sophisticated dish-level demand prediction algorithms” and “in-stations robotics’”. 

In fact, phrases like these often hide more than they reveal. Consider how proclamations such as “demand predictability”, “deep analytics”, artificial intelligence and machine learning have become almost commonplace in the technology narrative. There is, however, no denying that the rise in computing prowess has vastly improved the real-world experience of even the simplest services for the common man. This is not just the doing of sheer computing power but also the ingenuity with which new-age firms have used technology and data to their advantage. 

To fully grasp how a 10-minute or even a 30-minute delivery is made possible, it’s a must to look under the hood and see how an algorithm can tell that we want biryani on Sundays at 1.30 pm even before we consciously want it. 

Demand prediction: It looks like any other online store, but the reason Amazon is the e-commerce giant that the world swears by is because it has developed sophisticated demand prediction capabilities. For instance, Amazon leveraged its experience in selling only books earlier on to predict which cohort of customers would want to buy adjacent multimedia goods like music CDs. It even nailed modelling to predict which authors, brands and artists are going to sell how much and in which geographies. This allowed Amazon not only to stock its centres to serve people faster but also reduce costs of carrying huge inventories of stuff that’s not likely to be demanded much. 

In the same vein, food delivery companies such as Swiggy and Zomato have built prediction engines that calculate the probability of a given user ordering a given dish at a given time. They do this by feeding the computers tons of data on past behaviour and merge it with the current demand trends along with feedback from restaurants in the locality. For instance, the sales of biryani during the month of Ramzan are sure to spike up across the country but pizzas will be demanded much more on big sporting days or national holidays. 

Now, armed with this data, the companies simply tap into their customer data to predict whether you are likely to order a certain dish from their app on a given day. This works as well for rare users of these apps, as it does for superusers who tend to order up to three meals a day from these services. The reason is that with every additional delivery, the company gains precious data about the customer — from their locality, food preference, average order value, coupon code hunting, to even finer details such as the probability of last-minute cancellation or complaints about the food quality. 

Restaurant matching: It’s not enough to just predict demand. It’s also important that the delivery speeds match the claims. This means that restaurants that are too far shouldn’t be shown to the customer, while restaurants that are closer, but don’t stock a certain preferred item get pushed lower down the order. The idea is to manage to give a consumer a wide range of options without the buyer feeling a lack of choices, or getting disappointed with longer delivery times than expected. 

This is achieved through parallel processing of a large amount of data on actual traffic on the roads, restaurant’s past servicing times and the kind of dish ordered. For instance, it’s faster to pack a scoop of ice-cream as opposed to cooking a complex order of five different varieties of kebabs and three curries. This helps zero in on the expected preparation time, packing time, time on the road, which together determine the time it’ll take to get one’s lunch from the kitchen to the customer’s door. 

Delivering fast without driving faster: Even with all the data and prediction models around a certain order’s expected fulfilment timeline, the companies cannot promise a fast delivery time without knowing the delivery conditions. The roads could be jammed, the apartment complex could have several security checks for entry, and the delivery personnel itself could be far from both the restaurant and the final delivery point. 

To deal with this, the delivery fulfilment teams in these companies start working to find the best positioned delivery partner as soon as the order is placed. While the restaurant starts preparing the order, the companies send the order to the delivery partner stationed nearest to the outlet. As a result, the delivery partner reaches the restaurant just-in-time when the order is likely to be packed and ready for pick-up. This reduces wait time at the restaurant and helps the same number of partners deliver more orders by cutting their time spent on the road travelling to far-off restaurants or waiting to be handed the order by the restaurant.  

Fulfilment calculus: Predicting demand for a certain dish, finding the best restaurants to show to each customer, assigning a delivery partner that’s nearby do not guarantee a timely delivery. The product teams at food-delivery companies also spend a lot of time optimising the overall fulfilment process. This means building real-time dashboards that provide details of traffic on each key road or intersection, building a database of every large complex and the time it takes for the delivery person to enter, and even intelligence on the best routes that can be taken. 

For instance, a lot of new-age grocery delivery start-ups such as Blinkit and Zepto choose to open dark stores (grocery equivalent of cloud kitchens) next to densely packed townships, and their delivery executives are often told to walk through the service lanes than ride a bike through the main roads where signals, u-turns and jams can hamper the delivery speeds. 

This sounds ironic that the fastest delivery promise in some cases can only be fulfilled through a neighbourhood shop and delivered on foot rather than a motor vehicle but it’s the kind of common sense that humans have had for years and computers are only now getting to exercise. 

While the deep discounting, ambitious delivery times and the almost round-the-clock convenience is thrilling for consumers, it’s yet to be seen how the existing grocery and small eateries cope up with the challenge posed by these delivery apps. 

It’s hard to tell just how much damage will the restaurants, delivery partners and employees endure due to this race to be the fastest. But it’s clear that any conversation around the overall wellbeing of the humans of this ecosystem must begin by a more coherent understanding of the computer systems and algorithms at play since that’s where a lot of these decisions are made. 

This post appeared first in Times of India. It can be accessed here.