GitHub’s major metric for AI projects: Time saved

Like a cool boss wrapping up a company-wide Zoom ten minutes early, GitHub wants to give some time back to employees—and is looking at AI-powered ways to do it. The crucial factor: measuring time saved, a compounding gift that keeps giving, according to the chief operating officer of the company, which has already brought an AI-inspired approach to coding.

“If we get time back, we can reinvest that wherever,” Kyle Daigle, GitHub’s COO, said. “We have to be able to measure time gains. And luckily, most partners have some way of measuring that with us.”

Ask the OctoBot. In March of 2023, months after many tech-industry layoffs and the arrival of ChatGPT, Daigle made the company rounds to figure out how communications, corporate marketing, IT, and other “Hubbers” could do more with the technology of the moment: AI.

“I talked to everyone in my group, and I said, ‘If we’re going to help every Hubber, what should we be doing from an AI perspective?’” Daigle told IT Brew.

IT—a team that receives about 5,000 employee inquiries a quarter, Daigle said—raised its hand the fastest about how to use automation tools.

Just a few years ago, a three-year laptop replacement may have arrived a few months late, and Hubbers may not have even figured out that they were due for a device.

“We weren’t very proactive about letting folks know,” Philip Luedtke, VP of IT at the almost 3,000-employee-strong company admitted.

Enter the OctoBot—a Slack-integrated tool from the company Moveworks.The OctoBot has access to documentation, policies, internal systems, and process information, including three-year anniversaries.

Instead of an employee reaching out to IT and beginning a back and forth of tickets, the OctoBot arrives on schedule and asks if the employee wants, say, a Macbook Pro or Macbook Air. A selection on Slack triggers an order from the vendor to the Hubber’s door.

“It’s a touchless interaction from my team. It makes the Hubber much happier, because they’re just getting what they need [in] a very quick interaction,” Luedtke said.

And employees can reach out to OctoBot too, sending IT tech questions that used to involve a ticket: Need to re-enable MFA after getting a new phone? Need access to something? Ask the OctoBot, increasingly trained on prompt data and not requiring extra employee training beyond sending a Slack message.

Failure to communicate or solve the problem sends the conversation to the IT team for support.

“It would be like any normal DM, but instead of DMing a colleague, you’ll DM OctoBot,” Daigle said.

More time for…AI! Freeing up IT from doing repetitive work like break/fixes enabled more strategic tasks, like what to automate next, said Luedtke, who knows IT pros can come and go.

“The attrition and the turnover [in IT] tends to be pretty high. Because these generally entry-level or mid-career help desk folks, they want to do something more strategic or more engineering focused. And so this gives us the opportunity to let them continue to deliver that great customer service and fix people’s problems, but also now have more time to look for great solutions that help make life easier for all of us,” Luedtke said.

IT has been freed up to deploy initiatives like quick calendaring, an automated rearranging of meetings when, say, the unexpected sick day occurs. Each pilot—part of GitHub’s “AI for Everyone initiative,” which aims to offer strategic AI deployments for every employee—calls for trial periods, user-satisfaction surveys, and time-saved measurements, often available through the products themselves.

Daigle said OctoBot’s deployment cut out 30% of IT’s workload, or in GitHub terms: 3 hours, measured through ticket quantity and ticket-resolve time.

Surveys, deployed after every interaction, help determine a deployment that’s catching on with the company vs. one that’s not. The smart calendaring, for example, seems to be a hit, offering 30 minutes back per day for trial participants, according to Daigle. The time back is calculated through interactions registered and measured by the analytics platform.

And the COO’s ready for more AI-powered deployments.

“Let’s start and you get 60 days. And usually trials are free, at least in AI tooling these days. So: start, measure, try. And then, rinse and repeat,” Daigle said.

A 2,800-tech-pro survey from the technology and business-training platform O’Reilly Media, conducted between Sept. 14 and Sept. 27 of last year, found 67% of respondents reporting that their company used generative AI.

You just have to start by starting, Daigle said.

“The only people that are not able to get these benefits are the ones that are coming up with a foolproof plan on how to roll this out.”