The rise of the prediction machines and a skills ontology - Workday Rising 2018

Den Howlett Profile picture for user gonzodaddy October 2, 2018
A series of product announcements and a roadmap for an AI/ML driven future featured heavily in Workday Rising 2018's keynote.

Without Aneel Bhusri and his ability to wow the crowd, the Workday Rising 2018 keynote was little more than a series of product announcements with an awful lot of talk about machine learning and artificial intelligence, much of which will come in the future, perhaps a year from now. Even so, the company was able to showcase it's 98% customer satisfaction rating among polled Workday customers. That has to be satisfying.

Co-founder Dave Duffield, who prefers to eschew the big stage spotlight said from the get-go that Bhusri could not make Rising due to a serious family illness. Staying away in those circumstances is the right call but it meant that Jim Bozzini, COO was left with the awkward task of acting as MC. Nevertheless, Workday served up a polished and tightly scripted set of performances from familiar faces like Leighanne Levensaler, Betsy Bland, Pete Schlampp and David Clarke. This year, Workday also fielded Tom Bogan, CEO of the recently acquired Adaptive Insights.

Skills Cloud

We have already discussed Workday People Analytics, which took center stage on this occasion. Alongside, I was pre-briefed about Workday Skills Cloud, another item that got major billing. With Skills Cloud, Workday is attempting to unravel the ontological mess that surrounds skills descriptions.

According to Cristina Goldt, VP of HCM Products, individuals use similar but different terms to describe the same skill. What's more, as skills emerge and others become dated, the skills ontology is constantly changing.

In Workday's understanding, this means that comparing people with acquired skills and the identification of skills gaps are difficult to understand in a meaningful way. The company hopes that Skills Cloud is the first step towards streamlining the skills ontology.

To that end, Workday has applied machine learning matching to its proprietary data, customer contributed data, and seed data from public sources to reduce one million user-entered skills down to a much more manageable 55,000 verified skills.

When you think about the 31 million workers on Workday, it's really crowd sourced skills. So really moving beyond a taxonomy of skills to a machine learning powered ontology of skills. Understanding the relationship between skills and the distance between those relationships. Doing that creates that common language, which, becomes a common currency for evaluating your workforce. From a skills administration and maintenance perspective, it's basically self administering and constantly learning and updating with current information.

This is an interesting problem with many implications. During our conversation, I mentioned to Ms Goldt that I've been researching the topic of 'ethical programming' in the context of AI and yet find it almost impossible to discover where those skills are in demand or in operation. Applying a common ontology therefore makes sense. Similarly, I have noted in the past that some events bring together people umbrellaed under the CDO moniker but then discovering people with that title is a fool's errand.

Going beyond the ontology creation, Workday is hoping that this will encourage people to standardize on the descriptions that express skill sets and so make the business of job opportunity management easier. It is then one short step to assuming that as people consume learning, that their skills profile will be automatically updated to reflect current skill sets.

Right now, Skills Cloud is freely available to all Workday customers on an opt-in basis. Over time, the ontology will be applied to different industry segments. During the keynote, Workday talked about building out a talent marketplace for release in 2020 (Workday 34.) This is a logical next step and will open up a fresh business model for the company, potentially putting it in competition with the likes of Glassdoor and Upwork.

Doing it right with prediction machines

During the keynote, Leighanne Levensaler answered a few questions that have been preying on my mind for a while. First up she acknowledged that building reliable machine learning based predictive models that work in the context of HR is much harder than anyone thought. More to the point though, Workday has figured out that:

We needed to pair data scientists with business people. The business people needed to be there to ensure the data scientists are solving the right problems. We are going through our own transformation to become the Prediction Machine.

In addition, Ms Levensaler addressed one of my big questions for which see the screenshot below

Workday Rising

I have discussed the ethics question before but I am sure this will be tested time and again as development proceeds.

Future finance

Elsewhere, Tom Bogan talked about getting the people who are closest to the business working on the financial and HR plans with the expectation that planning should be continuous and agile. Adn as a demonstration of how the company is eating its own dogfood, Barbara Larson, Vice President, Corporate Finance at Workday discussed how her team implemented Adaptive inside 10 weeks using scrum techniques and how that is allowing her to concentrate on revenue and headcount as her north star metrics rather than worrying about spreadsheet error.

Looking forward, Betsy Bland discussed how Workday is using machine learning to build out anticipatory recommendations for the treatment of anomalies with recommended resolutions so that the period close becomes automatic with an end state of 'continuous accounting.' That's some ways off but it is not difficult to see how the application of AI based methods will change the nature of finance operations across a swathe of business processes.

Finance is uniquely positioned to be the stewards of enterprise data but first we have to answer the question: is this a chihuahua or a muffin?

The dog and cake reference is to the difficulty for machines to determine the difference between two photos as illustrated below:

Workday Rising


My take

I had to duck out of the livestream a little before it finished so didn't get a chance to tap into the PaaS content. For the TL;DR, you're best checking out Holger Mueller's tweet:

My sense from the cheap seats of the live stream was that Workday is at something of an inflection point. There is plenty for Workday's most forward-thinking customers to anticipate but many of the announcements are at least a year away. That's a very long time in the current world of AI/ML development. Directionally, it felt kinda right but then Workday still has to win over many more financials customers before it can claim leadership. It is conceivable that Adaptive, tied into promises of an automated finance future will be enough to win the day but competitors are not hanging around.

I personally missed seeing Bhusri and while the event followed a predictable pattern, the lack of customer reference talk was a disappointment. I'm sure Brian Sommer will track some of those down over the next day or so.

I'll catch up on some 0f these themes when Rising rolls into Vienna next month. I hope by then that Bhusri's family have sagely got past the current health crisis.

A grey colored placeholder image