Upskilling ramps up as gen AI forces enterprises to transform

Thomson Reuters is in the information business, and has been for a long time. Thomson Corporation was founded in 1934 as a newspaper company, and Reuters was founded even earlier, in 1851, to transmit stock prices. The emergence of the Internet could have been a death blow, but the company survived — and thrived. Over the past 15 years, its stock price grew from a low of $20 in 2008 to $170 today as it diversified into new areas of business, research, and workflow products for legal, accounting, and tax professionals.

Most recently, in June, it spent $650 million to buy Casetext, a 104-employee company that offers an AI assistant for legal professionals powered by OpenAI’s GPT-4, the same large language model (LLM) behind ChatGPT. But that’s not the only big bet the company is making on generative AI.

It also plans to spend $100 million per year to add generative AI technology to its flagship products, and has already built its Thomson Reuters Generative AI Platform, and released its first generative AI assistant for its legal research tool, Westlaw Precision. A more powerful AI assistant is planned to become the interface across other Thomson Reuters products next year. The goal is to shift the company from simply providing information to enterprises, says Mary-Alice Vuicic, chief people officer.

“It’s about how you shift roles to become more of a strategic partner, with more creative problem solving and more influencing,” she says. “We see huge value unlock in that.”

Generative AI will be key, improving the speed and quality of work in order to free up employee time to focus on these higher-value activities. Training and upskilling are also integral to this transformation, she says. “We’re taking this very seriously.”

In April, the company launched its generative AI education initiative with a global learning day, attended by 14,000 people. It covered the basics of AI and machine learning, generative AI and LLMs. This was followed by a six-hour course on AI and ML foundations, taken by 9,000 employees — with 3,000 more in progress.

“And the marketing department has launched Generative AI University in the marketing team curriculum,” Vuicic says. For more technical employees, there’s another course, about eight hours long, for 6,000 developers and engineers.

The bulk of the learning material was developed in-house, she says, with involvement from the head of product engineering and other technology and lab team members. Next year, this will be extended with the creation of new job roles inside different functional units, specifically dedicated to reskilling and integration.

“These reskiller and integrator roles are going to be essential to large-scale adoption of generative AI,” she says. “You have to dedicate headcount and resources to this, otherwise, it’ll be lost in the day-to-day.”

She says that Thomson Reuters is already seeing a positive business impact from the generative AI training.

“In areas like product engineering and customer services, we’re automating lower-level tasks and administrative work, and improving quality,” she says. “We’re seeing people able to better solve customer problems in significantly less time. The next set in training is about how we capture that augmentation and elevation. What does it mean to be a business partner? How do you apply strategic problem-solving?”

This is a new era of business transformation, she says, and will require a comprehensive rethinking of how jobs are structured.

Thomson Reuters isn’t alone in adopting generative AI, and needing to respond to the transformative pressures and opportunities this creates. As enterprises race to integrate cutting-edge tools and platforms, a greater focus to reskill employees on responsible and effective deployment becomes mission-critical.

According to an Ernst & Young survey of 1,200 global CEOs released in late October, a staggering 99% are either planning or are already making significant investments in generative AI.

Meanwhile, according to a Nash Squared survey of over 2,100 tech leaders, released in November, 54% say skill shortages prevent them from keeping up with the pace of change. At the basic level, every employee who uses AI applications will need to be somewhat familiar with prompt engineering, says Gartner analyst Arun Chandrasekaran.

Prompt engineering, at the end-user level, is about knowing how to ask questions in a way that gets the best response from a generative AI tool. At the developer level, it has additional nuances, involving, for example, embedding supplemental information from a vector database. Engineers might also need to learn how to fine-tune models, he says, and some employees should learn how to safely, ethically, and responsibly deploy AI models.

“Generative AI systems carry a lot of risk for enterprises,” he says. “There are risks around hallucinations, and the fact they’re black boxes in nature. And then there are the legal risks.”

Integrating gen AI in product, employee workflows

PhotoRoom is a 54-employee company that has developed AI-photo editing software, downloaded by more than 100 million people, but it has two main challenges related to generative AI. The first, says Eliot Andres, co-founder and CTO, is to add the latest AI features to the app itself to keep it competitive. The second is to enable everyone at the company to use generative AI tools in their own workflows.

As soon as Stable Diffusion, the first open-source diffusion model, came out in mid-2023, the company immediately pivoted.

“We stopped everything,” says Andres. “We said, ‘Stop what you’re doing. This is going to be a game-changer for photography. We need to learn how to use those tools and integrate them into our product. This is a big deal.’”

To jump-start the process, the company held a three-day hackathon to understand what the new diffusion models were capable of.

“Since then, every time there’s a new technology, we tell people to use the latest tools,” he says. “There are new diffusion models coming out, and we’re always encouraging employees to play with them.”

The company also has an internal channel where the latest research papers are shared. One popular source of information on latest developments is @_akhaliq on X (formerly Twitter).

For internal workflows, the challenge is slightly different. Here, instead of learning about the underlying technical aspects of creating generative AI models, employees have to learn how to use generative AI tools.

PhotoRoom doesn’t have a formal training program, he says. Instead, the company uses a teach-by-example approach, with employees who find useful new tools educating their peers using Slack and other channels. Andres’ advice? Give employees the freedom to experiment.

“If you limit your employees to a single tool approved by the company, you might be missing huge opportunities or something that’s transformational for your company,” he says. “Let them find the best tools for their job. If someone uncovers a tool that boosts their productivity, encourage them and try to spread the word to other employees.”

Leveling up for generative AI

Healthcare technology vendor athenahealth has over 6,500 employees, more than 1,200 of whom are engineers, and generative AI is core to the company’s product development roadmap — and to internal productivity improvements. And training is how the company will get there.

“We’ve offered a number of rounds of training, including third-party prompt engineering training, and workshops,” says Heather Lane, the company’s senior architect of data science. “There’s an enormous amount of training material out there, so we don’t have to develop it all in-house. We’ve also brought in external speakers who’ve talked about both the high-level state of generative AI and specific materials about how it may be reflected in the healthcare space, where they see the technological development going, where they see the risk.”

So far, more than 700 people have attended generative AI knowledge sessions, more than 1,200 hours of generative AI training have been consumed, and 300 developers have completed a generative AI bootcamp.

The problem, she says, is there’s too much material out there, and most of it isn’t useful. Filtering down to the part you care about is the challenge, as well as the pace of change. “It’s not like drinking from a firehose,” she adds. “It’s like standing under Niagara Falls and trying to take a sip. It’s absolutely insane.”

Marketing takes the lead on gen AI

IT services firm Ensono has over 3,400 employees, just 27 of whom are in marketing. But it’s the marketing department that’s been spearheading generative AI.

“It’s helping us do more with less,” says Jonathan Bumba, the company’s chief marketing officer. “Things that used to take several hours now take a few minutes; things that used to take several days now get done in a few hours; and things that we couldn’t do before at all we can now do in a reasonable amount of time.”

The company is using a wide variety of generative AI tools to create content and connect with customers, including ChatGPT, Dall-E, Midjourney, Adobe Firefly, and Salesloft among others.

“We started playing with it right away,” he says.

When the technology first started coming out, the team was mesmerized. “Then we were disillusioned pretty quickly,” he says. “We’re just not prompt engineers. I wanted my team to lean in and embrace this, but I could feel the frustration of not getting the output we were looking for.” He tried Coursera and LinkedIn courses, but they just weren’t specific enough, Bumba says.

Even vendor-provided courses were too general. “Without knowing our workflows and what we’re trying to do, none of it was terribly helpful,” he says.

His team struggled to figure things out on their own until the middle of this year. Then in September, Bumba brought in a boutique consulting firm, AI Technology Partners, to identify specific generative AI use cases, figure out the right tooling, and create customized training workshops for employees based on his team’s actual workflows.

“After the training workshop was over, we still put them on retainer so we can reach out to them with real-world projects,” he says. “They walk us through it and help us build prompts so we can continue to do this after they’re gone.”

Looking back, he says he wishes he’d gotten help sooner.

“I waited too long to ask for help, wasting months thinking we could figure this out ourselves,” he says. “I’d like to have those months back. In this new world, a three-month delay means someone else had a three-month head start on me. In the world of AI, this matters. He or she who learns first, wins.”

Large-scale upskilling

“PricewaterhouseCoopers has the commitment to give its people the most in-demand skills for today and tomorrow,” says PwC partner Robin Stein. “We realized very quickly we had to upskill our 75,000 people on the foundations of generative AI, how to apply gen AI responsibly, and how to become a prompt engineer.”

Because the company has a multi-generational workforce, it’s using a wide array of modalities, training approaches, and learning pathways, she says — everything from gamification to in-person seminars.

“Some people want to read an article and some want to listen to a podcast,” she says. “And there’s some gamification, including a live trivia game where people can earn rewards, which helps drive excitement about some of these programs.”

The final element is to give people tools they can use to apply these new skills.

“The adoption and engagement has been incredible, and people are highly interested and highly motivated to figure out how this will affect them,” she says.

Completing gen AI 101

Wipro Technologies is a technology and consulting firm with 245,000 employees. Of those, 200,000 have already completed basic generative AI training, says COO Amit Choudhary.

“We want all our people to be trained on generative AI,” he says.

The basic level of training includes a definition of what generative AI is, its history, and what responsible AI is all about.

“We want to educate our team members that there are risks to doing this,” he says. “And responsible AI means different things for legal people, for financial team members, or for cybersecurity experts.”

This program was developed in-house, he says, because the company started training earlier this year and there wasn’t enough external material available at the time.

Then employees move on to external training from external partners. That includes Udemy, Coursera, and LinkedIn, he says, as well as vendor-provided materials from companies like Microsoft, AWS, and Google for employees working on specific platforms.

Once the foundations are in place, Wipro offers industry- and function-specific training, such as what gen AI means for manufacturing, finance, HR, and supply chain management. As this is all created in-house, Wipro created an AI council earlier this year to handle it, tasked with building advanced courses customized to the individual requirements of the company.

There are hundreds of people involved in the training development process, Choudhary says, with a core team of over two dozen that creates the learning content and tracks its progress through the organization. Then there are all the people involved to execute the training plan, such as practice managers and service managers.

“We have consultants who are industry and domain-specific,” he says. “We have technologists who are high-end engineers, and a team of designers. We put all these people together, plus our partners and vendors. Our AI council sees the content coming in from outside, and the content we develop inside.”

And it’s not enough just to get the theoretical knowledge, he says. “Just telling them what it means is the first step. We’re also doing hackathons, and are now working on gamification of the training we already have. We also have our own platform for coders, where we give them live projects to work on as competitions. Once they’ve done some advanced courses, this is a problem they can solve.”