Bringing generative AI to hourly workers - how Legion's workforce management solutions use AI to serve a fluid workforce

Jon Reed Profile picture for user jreed January 10, 2024
Getting enterprise value out of generative AI is still early days. But can vendors with domain expertise advance us? Legion's approach to building an AI-driven workforce management solution is worth a closer look.

Sanish Mondkar, CEO and founder of Legion at Workday Rising
(Sanish Mondkar CEO of Legion at Workday Rising)

The stakes for enterprise AI in 2024 are already high. The shakeout at OpenAI and the EU AI Act have added new levels of complexity, raising questions about whether open source AI is viable - and how companies will approach AI amidst new IP lawsuits.

Not to mention concerning stories about unethical training datawho owns the model's content IP, and model drift.

Will enterprises pull back on their AI pursuits? I don't believe so - but this type of volatility will compel customers to seek out AI solutions from trusted software vendors, who can assume a major chunk of the liability risk - and source different Large Language Models (LLMs) as needed.

A parallel line of thinking is that these developments point to big tech in firm control of our AI futures. To some extent, this is true. As long as we are reliant on LLMs with voracious needs for training data, organizations will find it difficult to train their own models from scratch - it's too resource/cost intensive.

Whether that will remain an impediment remains to be seen - but for now, there are other ways to move ahead. I've already documented use cases where the vendor doesn't need to build their own LLMs to provide gen AI results for customers. In such cases, customer-specific data is infused into the architecture via RAG (Retrieval Augmented Generation) or other scenarios.

Gen AI startups with domain experts - a force for enterprise change in 2023?

This opens up the intriguing possibility that smaller startups of domain experts can build AI into their solutions - and provide impact that is much closer to "out of the box" than training your own LLM, and without the risk mitigation issues that do-AI-yourself currently raises.

With that in mind, I have a couple of standout examples I picked up on my travels, starting with Legion, an HR solution focused on the problem of the deskless/hourly worker. This is actually a difficult workforce to manage - and the solutions to do so must encompass the complexity of on-the-fly workforce scheduling.

No surprise then, that Legion has been knee deep in applying AI to labor challenges for years now. But when I met Sanish Mondkar, CEO and founder of Legion at Workday Rising, Legion had just announced their Copilot Generative AI Assistant. It was not a coincidence to find the Legion team at Workday Rising. Legion is a part of the Workday Ventures program.

Some vendors seem to mistakenly assume that the scale of big AI requires them to develop all their customers' AI solutions internally. But the savvier vendors will find ways of opening their platform to their partners' AI applications. That's why I thought Workday's announcement of their AI marketplace (and certification program) was one of the underrated AI announcements of the fall.

Legion: we need better workforce management in labor-intensive industries

During our sit down, I asked Mondkar how this all started:

Legion is workforce management, largely catered towards labor-intensive industries. So retail, hospitality and so forth. Our customers in those industries, we help them manage their labor force, from the perspective of scheduling them to work - and then capturing time and creating the timesheets. So payroll can be efficiently generated, and also increasingly, enabling employees to have more flexibility and ownership over their schedules - and work at different locations of the customer.

Mondkar says this kind of a workforce management can empower employees who typically don't feel very empowered:

This gives them more options to earn more wages, and also provides a lot of self-service tools for them if they want to manage this - if they're running late to a shift, like how to communicate to the manager or swap their shifts, or if they want to get paid instantly after the shift is over. So these are all the capabilities that add a lot of value to these labor intensive industries, because labor is a challenge for them.

I spent many years writing blog posts at a local brewery. I saw firsthand the complexity and urgency of shift-swapping and schedule management. Handling this manually means contacting every other worker every time a shift opens up - and individual workers may not even have all the contact numbers. Frantic Facebook Messenger threads aren't the way forward. As Mondkar explains, providing user-friendly apps for this purpose is crucial in today's tight labor market:

The turnover is very high; anything they can do to increase efficiency, but also simultaneously engage and improve the employee experience is a highly-valuable thing. So that's where Legion comes into the picture.

But many companies already rely on an HR system of record/cloud platform like Workday. So how does Legion fit in?

We integrate with Workday. We basically integrate Workday employee record into Legion, and then from Legion, we've downloaded the Legion app to managers in these locations. They use Legion to generate the schedule, and all that good stuff. So that's the fundamental value proposition.

Investing in workforce management - intelligent automation is a must

Given the popsicle headaches of hourly worker scheduling, it's no surprise that Legion invested in AI early. Mondkar says it comes down to their founding question: "How are we going to improve workforce management?"

We saw two ways to do that. One is to make it more employee-centric as we talked about - so, make workforce management actually about the workforce. And the second was automation. If you look at any labor-intensive job like restaurants or retail, the managers, they are not managers of that store because they like sitting in front of software and creating schedules. That's not why they are there. They like to be in front of customers, and like to run it as a business.

No one will argue against automation, but they just want to make sure it's done right, because a lot of important decisions happen on a daily basis: how much labor needs to be allocated, where should it be allocated? How many people should work, who should work at what time? How to deal with timesheets?

That took Legion's team into an "intelligent automation" journey. But in practice, what does "intelligent automation" really mean? The response:

Intelligent Automation means the right decisions are made not just for the sake of automation. The model we built when Legion was founded was machine-learning-based demand forecasting... Legion went very deep with AI and machine learning from day one - and that's very important. We built AI-based demand forecasting for automated scheduling. We do things like when managers edit the schedule, we learn from those edits and we correlate that with employee behavior, employee preferences and things like me automatically.

For the last five years, Legion's AI-based scheduling automation has integrated a range of third party data, from weather forecasts to religious holidays. Long before generative AI, Legion's team was working with predictive algorithms, optimization through expert systems, and AI learning through neural nets.  So when generative AI heated up, Legion was on the case with their new Copilot. Mondkar says gen AI is a very good fit with the needs of hourly workers, who need a fast/intuitive conversational interface on-the-go:

The thing we find so fascinating about generative AI is for workforce management, 99% of our workers are deskless workers... We think and generative AI provides this natural language interface for this workforce. Plus, they do not have time to get trained on software. The turnover is so high. When a new manager shows up, you're training how to be productive as soon as possible.

As Legion puts it:

Regardless of their location or task, managers and employees can now engage with Legion Copilot to gain insights into labor regulations, internal company policies, and even bespoke training content.

Legion's Copilot roadmap includes:

  • Knowledge-Based Self-Service
  • Labor and Hour Compliance Q&A
  • WFM Conversational Interface - with ability to pull up and manage schedule, with user-permissioned actions like adding shifts and changing shift assignments (natural language queries like "what should I know about the schedule?" brings up the relevant info - full details on Legion's blog)

My take - enterprise AI is about domain expertise in 2023

As the new year kicked off, I asked the Legion team for an update, before they head off to the NRF Retail "Big Show" in New York City (you can book an NRF meeting via their web site). Legion hit several key milestones including gross revenue retention (GRR) of 97%+ and 55% growth in revenue, and 88% weekly active users on the platform. Legion this growth to "an increasing need for AI-native solutions for workforce management."

Via email, I asked Mondkar about the potential for smaller vendors to play a key role in enterprise AI adoption. Can customers move ahead despite the various roadblocks with consumer LLMs I noted in the intro? Mondkar's response emphasized the importance of customer data - and the collaboration that entails.

It is true that the ownership of data is a crucial factor in training LLMs. This may imply that larger players who has access to bigger datasets have an advantage. However, the world for enterprise software vendors is more complex. SaaS vendors do not automatically have rights to customer data, even if it is stored in their databases. Instead, SaaS vendors must collaborate with their customers to establish guidelines and obtain permission to use the data effectively. This collaborative approach ensures that customer privacy and data security are safeguarded.

In the generative AI debate I took part in on CRM Konvos this week, I cautioned about the copyright and IP problems the large LLMs are facing, but I also encouraged enterprises not to sit on the sidelines. This is a good time for customers to look to industry-specialized vendors to help them move forward with focused AI projects, with focused data - and minimize LLM risk, while addressing something that moves the business forward.

On the customer use case side, I'd like to talk further with a Legion customer once the gen AI use cases mature. For Mondkar, the ability to serve these customers has been propelled by their time with Workday Ventures.

There were a lot of synergies. We already had common customers. They liked what they got from us. They liked what they got from Workday. They said, 'You guys should be working more closely together.'

I talked about domain expertise being crucial to the next phase in enterprise AI. But maybe I should have also said a strong organizational purpose - one that elevates human work even while it augments and automates. That certainly seems to be what's driving Legion. As Mondkar told me:

Our founding thesis for Legion is to turn hourly jobs into good jobs. That's our mission.

The pandemic put pressure on that mission, as hourly labor reached a breaking point, and that pressure hasn't really let up.

The reason why it was building up was that there wasn't much innovation, or even frankly, desire to improve the experiences and uplift these jobs, fundamentally, from the perspective of experience. So I feel pretty strongly about that.

If you can build a business on that, I'll be rooting for you.

A grey colored placeholder image