Chatbots have had a patchy track record world of tech, where early efforts not only failed to deliver on the magical idea of a computer producing the exact answers you were looking for in a chat-based-Q&A, they even produced surprising (and not in a good way) results instead. Things have moved along, though, and today a startup that’s built a platform to help improve chatbots’ responses is announcing a round of funding from a key strategic investor, a sign of demand and evidence that its solution is working.
Directly — which has built a platform to help train companies’ chatbots by crowdsourcing experts and analysing chatbot usage to better “teach” the AI systems underpinning them — has raised $20 million in funding led by Samsung NEXT (Samsung’s VC arm), with participation also from Industry Ventures, AvidBank, and existing investors M12 (Microsoft’s VC), Costanoa Ventures, True Ventures and Northgate.
Along with this, the company is announcing some executive news. Mike de la Cruz, who had been the company’s chief business officer and has held executive roles overseeing customer service products at SAP, HP Enterprise and elsewhere, is stepping up to be the new CEO. Antony Brydon, who co-founded the company with Jeff Patterson (currently head of product), is moving over from the CEO role to become head of platform, where he will focus on how and where to take Directly’s technology beyond its current market focus on chatbots built for customer service.
The idea is to give over company growth to an expert in that area, while giving a company founder the chance to help figure out how best to focus the company to forge into newer areas. “We are thrilled to have Mike at the helm of Directly,” Mark Selcow, partner at Costanoa Ventures, said in a statement. “He has driven record-setting growth for the company in 2019, and we look forward to the impact he’s going to have into 2020 and beyond.” Brydon and de la Cruz have known each other since college, which speaks to a long friendship and trust in each other too — a good sign, in my opinion.
The valuation is not being disclosed except Brydon and de la Cruz, in a joint interview this week, confirmed to TechCrunch that it is north of $100 million. When this round was being raised, PitchBook data noted that the valuation was $110 million, which roughly lines up with that.
They added that the idea is that this funding was opportunistic — Samsung is a customer, along with biggies like Microsoft (also a strategic investor), and Airbnb (not an investor!). The bigger plan is to raise a much larger Series C round in 2021 — which implies it has enough runway for at least the next year.
Directly has emerged at a key time in the world of customer service, and in AI.
Although call centres are still a fundamental cornerstone of how businesses interface with users, the rise of social media and messaging services has created an opportunity to complement and in some cases replace how those more traditional channels work.
While in many cases human agents are still at the other end of those messaging, typed conversations, sometimes you only might think they are, or when they really are, they are still using a lot of AI tools to help them be as informative as they can be.
That’s where Directly comes into the picture. While there may be some true superstars in the world of customer service who know the product they are representing backwards and forwards, chances are that the vast majority of people helping customers are not omniscient gurus, and are just human like the people complaining or asking questions.
Directly has therefore built a platform that helps businesses identify and reach out to ‘experts’ in the business or product in question, collect knowledge from them, and then fold that into a company’s AI to help train it and answer questions more accurately. It also looks at data input and output into those AI systems to figure out what is working, and what is not, and how to fix that, too.
The information is typically collected by way of question-and-answer sessions, and Directly has even built a way to compensate these experts both for submitting information, as well as to pay out royalties when their knowledge has been put to use, “just as you would in traditional copyright licensing in music,” Brydon noted. This system is used for technical support for games — Microsoft uses it to power its Xbox assistant for example — or devices, or for those who are engaging in building businesses on platforms, such as in the case of Airbnb providing an assistant to hosts to help answer questions about how to list on the platform.
At a time when we are talking a lot about bias in AI and other pitfalls of how these systems are trained, a company that is working on ways of giving the best shot possible, by limiting the training data just to what is most likely to be correct and verifiable, is an interesting prospect that lines up with the approaches that companies like Samasource and others focused on ethical AI are taking.
It seems ironic that tech giants like Microsoft and Samsung, which have put a significant amount of investment into acquiring businesses and organically building their own customer-facing AI systems — Samsung’s Bixby is built in part through the acquisition of Viv and a multitude of other related deals; ditto Microsoft’s Cortana — would rely on another company to help these along, but the role that Directly occupies somewhat sits outside the core technology, which needs computer vision, natural language processing, a larger machine learning engine to process all of the inbound data and more.
Training those AI systems — the area that Directly is focusing on — is likely to remain a key area for how these are used and develop, since it could be many years before we see what Brydon refers to as the “holy grail,” a general AI that can do the training for itself, and work across a diverse range of fields and specific interests.
In the meantime, it can take as little as 100 experts, but potentially 1,000 to train a system, depending on how much the information needs to be updated over time. The Xbox implementation, for example, includes 1,000 experts, but has to date answered around 2 million questions (and will likely answer many more as games and consoles get updated).
The longer term picture is that Directly is likely to work with a growing number of businesses as the use of chatbots continues to expand among organisations. With that bigger trend, it’s also likely to run into some of the biggest players in consumer information like Google and Amazon. Neither are customers of Directly yet — and truth be told they might just as easily end up competitors — but it makes for an interesting prospect, as AI finally starts to get more intelligent.