Is it possible to stitch together the mess of data lying around corporations and make of it something new and useful? One enterprise software startup hopes to show that machine learning can find new treasure in old records.
Six-year-old Clari of Sunnyvale, Calif., says that artificial intelligence can do something "really big" in enterprise software, something that goes beyond customer relationship management and human resource planning, even though it relates to both those domains. The company can model deals to see possible outcomes, and spread the insights to teams throughout a company's office, including not just sales but also marketing and customer support.
Clari was previously profiled in October an interview with co-founder Andy Byrne, by ZDNet's Colin Barker. The company, founded in 2013, has raised over $70 million from top-tier investors such as Sequoia Capital and Bain Capital. Byrne and co-founder and CTO Venkat Rangan previously founded enterprise software maker Clearwell Systems, built it to $120 million in revenue, and sold it to Symantec.
On Wednesday, Clari announced its software now has functions for not just sales teams, which is where it started out, but also for people in marketing and customer service. It also announced integration with a raft of programs that serve various parts of those functions.
Where previously the software plugged into Salesforce data, and email systems, it can now hook up to data from Chorus.ai, Dialpad, DiscoverOrg, Gong.io, Highspot, Outreach, PFL, RingCentral, Salesloft, Sendoso, Showpad, and Yesware.
Clari bills its integration of all these data sources as a revenue "platform," to bridge the functions that comprise the "go-to-market" work of a company, all the responsibilities that span sales to marketing to customer support. The premise is that machine learning can draw connections between this data to make data useful in a way that it's not when data isolated sitting in various repositories.
The AI in this case is a blend of the old and the new. The chief implement used is what's called a Hidden Markov Model, or "HMM," a statistical procedure first used in the 1960s to determine unseen states of affairs in a system by working backward from observations.
The HMM is used "as one of the models for determining the true state of an opportunity," meaning, an opportunity in the business sense, says CTO Rangan. That analysis happens only after a Support Vector Machine (SVM) classifier first identifies features to go looking for in the data, he explains. SVM is a machine learning approach that became prevalent in the 1990s.
In addition, "The other models that we also bring in are GBDT [gradient boosting] (a variant of a so-called decision tree), and LSTM (deep neural network) for forecast projections," says Rangan, referring to long short-term memory, or "LSTMs," a machine learning approach for dealing with sequential data, such as a time-series of information.
All of that is meant to clean up what has become a mess of data scattered all over organizations, in email, in apps, sometimes in Microsoft's Excel. "A lot of pain," was what CEO Byrne found when he talked with Fortune 500 companies at the outset of the company's journey, in 2013.
"We found that there were data quality issues, with [sales] reps not wanting to manually enter their sales calls into CRM, managers having no visibility on what's going on, what's really happening in the field."
"One of the most important processes in the company," he says, "was in Excel Hell," meaning, reps just dumping their notes about prospectives into a spreadsheet.
With the platform approach announced Wednesday, more actors across a company can see what's happening at different points in the funnel. That had already been taking shape, claims Byrne, as the product spread to other teams, in a kind of stealth, grass-roots way. "Marketing reps have started to use the AI to predict how much pipeline is needed to reach future revenue goals," says Byrne. "The other bucket is customer success and renewal teams; unbeknownst to us, they were using us to manage and forecast their churn."
It's all about bringing together a "myriad of signals, a total number of signals," says Byrne.
The process starts with an "auto-capture" feature where the software digs out all the data trapped in various apps, things such as missing phone numbers or appointments. In one Fortune 500 company, says Byrne, when the system was first deployed, "we found 52,000 people [records] within minutes, and over 19,000 opportunities that were being managed" that hadn't been logged in any formal way. "No one had visibility into the reps talking with all those people."
What makes possible the connections between CRM and Salesloft and RingCentral is that "all these systems -- Chorus and Outreach and Marketo and Eloqua and Hubspot -- they all have rich, open APIs," says VP of products Kurt Leafstrand. "So it's relatively straightforward," as opposed to trying to integrate with, and extract from, traditional ERP apps.
The HMM and attendant machine learning techniques examine "objects" in the data, such as deals, "to analyze all the paths of successful deals, and compare which deal is most like another deal, and which ones the company has won or lost," Leafstrand explains. The technology "looks at all previous trends of previous quarters, months, weeks," he says.
The HMM builds a model of deals that teases out hypotheticals. And then, "We get two years of sales data from Salesforce, it's a time machine that let's us go back and figure out how accurately our models would have predicted what would have happened, which allows us to verify the models we're building," says Leafstrand.
- 'AI is very, very stupid,' says Google's AI leader (CNET)
- How to get all of Google Assistant's new voices right now (CNET)
- Unified Google AI division a clear signal of AI's future (TechRepublic)
- Top 5: Things to know about AI (TechRepublic)
There are, of course, many enterprise software companies suddenly branding themselves with AI. ZDNet has written about startup People.ai, backed by Andreessen Horowitz, which aims to resolve the inertia to update Salesforce records. Another one is InsideSales, which analyzes CRM data to try and find patterns that lead to successful sales.
"There is an explosion of companies trying to solve the sales problem," concedes Byrne. "We are absolutely breaking ground, we haven't seen anyone out there who is doing this sort of thing at the scale and speed we can, with verified results."
"No one is connecting all these different signals across all these different systems," he says. "Last year, we analyzed over 50 billion data points, representing $200 billion worth of pipeline across our customers."
Some very big CRM vendors, observes Byrne, claim to do the same, "but it would take them eight months [to deploy the software], whereas our product is up and running and adding value within the same day."
Down the road, the company plans to offer more AI technology of the deep learning sort, says Leafstrand. The technology, currently in beta, addresses "the most complicated part of the revenue problem, what's called the new bucket," he says. "These are opportunities that don't exist yet but will exist; that's the first place that we have found a bonafide benefit from deep learning, where it provides some really novel insights, to predict the stuff that doesn't exist."
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.