This AI bot knows all about women’s tights. Here’s why that’s vital.

Photo credit: Flickr user Geoffrey Fairchild

AI-powered chatbots became a craze in 2016, thanks to advances in natural language processing, access to open source algorithms, and platforms like Facebook Messenger opening up to them. But the hype cycle soon hit a trough as more than two-thirds of the bots on FB turned out to be duds when they ventured beyond demos into real life.

A conversational layer for brands to engage with consumers 24/7 with intelligent bots still holds great promise. But companies have become wary about doing it in a ham-handed way that drives away users instead of winning their loyalty. That’s where deeper tech and more focused bots come in.

“The ones which will make a difference are those which have the domain knowledge to address core problems in a certain vertical. To use the same bot to order a pizza and book a bus ticket – those days are gone,” says Sanjoy Roy, CEO and co-founder of AskSid that makes AI bots for fashion brands.

AskSid’s IQ has passed early exams in Europe. The year-old Bangalore-based startup’s first big client is Austrian women’s lingerie brand Wolford. Five months on, the AI bot is engaging Wolford’s customers in 12 countries. Two other fashion brands have signed on in Europe, but Roy did not want to disclose their names at this early stage.

The world of tights

Roy gave up a successful career in tech business development spanning the US, Europe, and Asia to plunge into entrepreneurship. He headed the digital business group for Indian IT giant Mindtree’s Asia-Pacific markets before founding AskSid.

But it was years earlier, when he was doing his MBA in the Netherlands, that he first encountered the problem AskSid aims to solve. His wife Dolly needed a pair of tights to keep her warm in the cold weather, and it wasn’t something a traditional Indian is used to buying. That she was pregnant complicated things.

They decided to go for a high-end brand that was recommended to them, but were flummoxed with a selection of 200 tights. Finally, the couple just went to an offline store of Wolford, where a knowledgeable sales associate figured out Dolly’s requirements in no time. This got Roy thinking: How closely could this be replicated online?

The thought resurfaced in 2016 when AI-powered chatbots began to make waves. Roy reached out to the top boss of Wolford via his European business network, and got an appointment with the brand’s ecommerce lead.

He found out that Wolford had tried using human assistants to help its online shoppers, but the experience was inconsistent. It was hard to standardize staff responses and scale it up in 20 countries. The brand was happy to try an automated model. Roy then started to build AskSid with seed funding from Mindtree founders.

AskSid, TrainSid, and MeasureSid

AskSid’s chatbot is the tip of the iceberg; the bulk of the work goes on below the surface in two other modules.

One of them – TrainSid – ingests the brand’s product data from catalogs and other sources, including customer calls to the help desk. This helps form Q&A pairs, build product attributes, and apply AI to power engagement via the chatbot, which is text-based currently but could be a voice assistant in the future.

Then there’s MeasureSid, which derives insights from the chats for brands that they can use to update their catalogs, spot trends in what shoppers want, and so on. For example, Wolford launched a marketing campaign around opaque tights after noticing women asking for them in one of its markets.

AskSid is thus integrated into the brand’s product, inventory, pricing, ecommerce, and promotions data. It engages with customers on multiple channels, including the brand’s website, app, Facebook page, and Skype – with WhatsApp waiting in the wings.

See: Where Singapore lags behind India in retail tech, and opportunities hide

For now, the bot does not access personally identifiable data that would enable it to track a customer across multiple devices or channels. That means personalization has to be limited to a specific channel, but Roy doesn’t mind that given the current pushback against violation of privacy.

Deploying the bot on ecommerce marketplaces is also distant on the roadmap. AskSid is content to build its B2B (business-to-business) model of helping brands engage directly with customers. This also ties in with a trend of brands trying to build up their own ecommerce channels instead of becoming overly dependent on crowded marketplaces, where it’s harder for a seller to own the customer data.

From national digital ID to fashion bot

AskSid co-founders Sanjoy Roy and Dinesh Sharma / Photo credit: AskSid

While Roy spends his time finding the next big client for AskSid and understanding their requirements, his co-founder Dinesh Sharma takes the tech lead. They worked together at Mindtree, where Sharma was the architecture head. One of the highlights of his role there was to give shape to the blueprint of India’s digital ID for citizens, Aadhaar.

He describes the “sense and serve” function of AskSid as a triangulation of “context, intent, and entity.” The ‘context’ could be a black top priced below US$100. The size ‘XS’ would then be applied in that context, and the shopper’s ‘intent’ there is to apply a filter. Entities are various product attributes like color, size, type, material, and so on.

The point is to train the bot to propose the next best action to the shopper based on an understanding of contexts, intents, and entities specific to fashion. For example, colors in women’s lingerie are more nuanced than the usual ones we encounter in everyday things.

The bot has done 30,000 conversations in 20 different contexts. In the process, it expanded Wolford’s repertoire of typical customer queries to which it can respond intelligently in real time from 1,500 to 4,500. AskSid can also “identify 100 unique intents compared to 5-10 intents for most chatbots,” adds Sharma. That can make the difference between a sale and a postponed decision from a potential customer.

“Tights and lingerie worth 25,000 euro have gone through our chatbot. In absolute terms, that’s small, because Wolford is a 200-million-euro brand with its online business being 15 million euro. But people are getting into the habit,” says Roy. Repeat users make up 20 percent of those engaging the chatbot.

This month, Chinese investment giant Fosun paid 55 million euros (US$68 million) to take a controlling stake in Wolford. That could open new doors for AskSid as Fosun builds a fashion portfolio in Europe.

The next frontier

Like many other bots, AskSid is built on the MS bot framework. But after dabbling with Microsoft’s natural language processing engine Luis and others like Facebook’s Wit.ai, the startup decided they were too generic. “They all have some elements of commerce, but they’re not immersed in the entities and intents of fashion,” says Sharma. “Instead of going and enriching those APIs, it’s more predictable to build the components ourselves, and that’s the journey we’re taking now.”

There are hundreds of chatbots dabbling in fashion, but Roy says he has stopped tracking them because he doesn’t think they go deep enough into the vertical. He does track a few Valley startups that have taken a similar vertical approach. One of these is Mode.ai, which has virtual shopping assistants for Levi’s and Louis Vuitton.

“We feel good that we have a few big brands ourselves. But we don’t want to feel that we’ve arrived. There’s a lot that we have to learn and do,” says Roy. For him, the next stage is to find a partner that will widen the sales pipeline, instead of going from one client to the next.

See: The former Microsoft director who tossed a coin to decide if he should quit his startup

As for Sharma, he’s figuring out how to add more verticals to prove that the core architecture behind AskSid can be applied to related domains. This month, for instance, AskSid went live for a paints brand, and the deployment had taken just six weeks.