The risk to people's jobs from artificial intelligence, the prospect that machines will displace workers, has a kind of positive flip side, according to some: The possibility that taking away the more mundane parts of work may make those still with a job more productive.
That's the premise of a startup in enterprise software that's been blessed with $42 million in the past two years in order to chase down those parts of information work that lie abandoned in dark corners.
"I hated logging stuff into Salesforce," reflects Oleg Rogynskyy of his many years in sales and marketing using the marquee CRM software.
Rogynskyy is founder and chief executive of San Francisco-based People.ai, a two-and-a-half year old cloud software venture that on Tuesday announced a $30 million Series B round of funding from venture capital firm Andreessen Horowitz. The new money follows seed investment from Y Combinator, Index Ventures, Shasta, and a group of angel investors, and an A round led by Lightspeed Venture Partners. Y Combinator stuck around for both the A and the B rounds. (A-H partner Peter Levine has written a blog item on the funding, and Rogynskyy has written his own essay on the matter.)
Rogynskyy has convinced all these backers that the task for employees of maintaining Salesforce and other systems is not only loathsome, it is a profound productivity sink for enterprise teams.
The mission of People.ai, he says, is to use computers to fill in the blanks, freeing up employees' time.
Rogynskyy, 32, said in an interview this week with ZDNet that his epiphany began when he was working at Nstein Technologies, a natural-language processing company founded with the participation of Yoshua Bengio, a star of Toronto's MILA machine learning group. He had come to Nstein after studying business administration and political science at Boston University, and graduating.
Working in sales and marketing at NStein, Rogynskyy noticed that "I was forcing my people to use Salesforce."
"I could never get them to get the data into the system without wasting their time."
He believes the way to resolve that is a cloud-based service that automatically extracts information about people and activities from any part of a company's system, be it email, calendar, phone logs, Slack, DocuSign, or WebEx, and to draw connections between what is found.
A sales executive who goes to a meeting with 15 individuals leaves a paper trail in email, say, of who was at the meeting. If the sales rep fails to fill out notes for the meeting in Salesforce -- a reasonable possibility given that "you're too busy with fifty other meetings," Rogynskyy points out, and "nobody actually pays you to do that data entry" -- then the information can be automatically scraped from the email by the People.ai system.
The cloud-based system aggregates information from across People.ai customers -- some 50 Fortune-500 names, currently -- into a "massive time-series-data-set pipeline" consisting of over 120 million "people objects," and over 10 million company objects.
These can then be matched against one another in various ways by machine learning. "If you go into any CRM system of any established company, you will see a lot of records for the same customer -- like, 50 Coca-Colas, say, of which some are duplicates, but some are by design [such as different divisions or franchises]," explains Rogynskyy. Machine learning is used to tease out which entries really are pertinent to an executive's ongoing sales activities -- stitching together the right objects.
To do it, People.ai maintains a fairly liberal data science department, he says, with the team "free to explore and throw compute at whatever they want until it takes it to the right product solution." That means using not just recurrent neural networks, RNNs, and convolutional neural networks, or CNNs, but also approaches such as "random forests" for classic capture of things such as job titles. There is also a "little bit of a playground" in People.ai's labs with "long-short-term memory," or LSTMs, a machine learning approach that is especially good at working with sequential kinds of information. LSTMs may be a resource, he offers, for filling in the blanks on records for people where there is not sufficient data up-front.
"One application is what we call behavior similarity for people," he explains. "We can take a structured data profile of someone, including job history, what things they like, who they engage with, what time they engage with that person, and we can turn that into an image bitmap of them as a profile; then we can use computer vision to compare millions of bit maps of people one to another to find similarities."
"If you aggregate knowledge across tens of thousands of people, you actually get to know people really well," he continues. "LSTM is really good at identifying temporal changes to this thing." Another area the company is exploring is graph analysis, as "graph analytics is a very young, nascent area," he muses. "When it comes to analyzing at scale, there's not much in the public domain" for graph analysis. "We're doing a lot of work there, and we plan to open-source it."
The result of all this ML is not just filling in databases, but carrying out tasks on behalf of an executive, Rogynskyy says. "We scan your calendar and see those people who were on the conference call, based on your email or by phone number; we identify whether they have bought from you before; we create them as contacts automatically; we use the voice recording of the call, if we have it, and put that into Salesforce, along with a sentiment analysis of the meeting; we track the communications; and we follow up before or after the meeting, and update all that in your Salesforce, or tell you when you forgot to follow up."
The productivity payoff is time not spent by a human on all that data entry.
"We think that about 40 percent of white collar workers' activity can be automated," Rogynskyy says, the time "spent on non-value-add things like entering data into systems like CRM."
"We are already seeing up to 20 percent of employees' time being freed up," he claims. That is based on performing A/B tests with customers -- basically, seeing how much executives get done when the People.ai system is put in place and they are freed to do other things. "You see more productive activity, like meetings with C-level executives."
"Twenty percent of work saved adds up to one extra day per week," as he sees it. "If a sales rep spends half of that extra day playing golf, they will still be productive."
Even more so, one supposes, if it's golfing with clients.
Previous and related coverage:
An executive guide to artificial intelligence, from machine learning and general AI to neural networks.
The lowdown on deep learning: from how it relates to the wider field of machine learning through to how to get started with it.
This guide explains what machine learning is, how it is related to artificial intelligence, how it works and why it matters.
An introduction to cloud computing right from the basics up to IaaS and PaaS, hybrid, public, and private cloud.
- There is no one role for AI or data science: this is a team effort
- Startup Kindred brings sliver of hope for AI in robotics
- AI: The view from the Chief Data Science Office
- Salesforce intros Einstein Voice, an AI voice assistant for enterprises
- It's not the jobs AI is destroying that bother me, it's the ones that are growing
- How Facebook scales AI