Google’s algorithm-powered internal job marketplace

Google’s algorithm-powered internal job marketplace
Google’s frontline teams needed a new staffing model to better support strategy shifts and employee development. It needed to be scalable and dynamic, so we created a job market, giving employees and managers choices.

gTech is a support and operations organization and within gTech the Users and Products division works to ensure users and partners get the most out of Google. For a long time, the team was organized around individual products but whenever Products strategies shifted (a frequent occurrence in tech) we would need to reassign resources quickly. The old model wasn’t agile enough to quickly assemble teams with the right mix of technical and operational skills. Meanwhile, we knew from our employee engagement survey that teams wanted more development opportunities, career mobility, and knowledge sharing.

We considered hiring buffer capacity (too expensive), rotation programs (too rigid), and trying to evangelize our existing internal job board (too slow) and ruled them all out. We needed a scalable and dynamic staffing model. Working with an in-house economist, we decided to test a marketplace approach where employees and managers “bid” for assignments.

But the devil was in the details. How would we actually make this work for an organization of hundreds of people? How would we prioritize projects? How would we get the right skill fit between employee and role? How would we prevent people from gaming the system? How would we know if the program was successful?

We want to offer real choice to employees. This meant we needed to create a market that was “thick” enough with a critical mass of varied opportunities. We settled on batching job postings in three rounds a year. The idea was to get employees and managers to be less possessive of projects or teams. We needed employees to move around more, adapting to their surroundings, so we called the effort “Project Chameleon.”

We also had to build a platform for this market. To help nudge employees and managers to consider skill fit, the tool displayed a color-coded match score based on the overlap between the Googler’s experiences and the job’s requirements. Googlers and managers ranked their preferences and we paired them using a Gale-Shapley Deferred Acceptance Algorithm. The algorithm would match employees to roles based on their preferences and those of the manager. A similar approach has been used for decades in the US to match tens of thousands of graduating doctors with hospitals.

We discussed running a small pilot, but the market required a critical mass of jobs, so we launched with rolling enrollment, posting about a third of all positions each round so that after the first year, all roles would be in Project Chameleon.

Building trust in the marketplace and matching was tough, even though Google is a place built on algorithms. We needed everyone to list their true preferences for Project Chameleon to work. It was a slow start, with only 11% of employees actually moving in the first round; of those that moved, more than 25% didn’t get their top choice. And while we never intervened with the algorithm’s output, we would talk with a Googler if they matched with a very low-ranked job to see if there were still other roles they’d rather pursue. By the second year, more Googlers switched roles in a single batch than had switched in all the prior rounds. At this point, our data showed that Googlers trusted the market with 74% agreeing in an anonymous survey that “I felt comfortable sharing my true preferences” (and only 5% disagreed).

The program has been running for a few years now and while there have been plenty of tweaks along the way (like getting rid of a round during the hectic fourth quarter) we’re seeing some of our intended outcomes:

  1. Prioritization - The new business prioritization process at the organization level, rather than by product, provided valuable insights across the leadership team and helped everyone think less defensively about their teams and more expansively about Google’s strategic opportunities.
  2. Transparency - Publishing all project roles and required skills provided transparency to Googlers of all available opportunities.
  3. Agility - Launching centralized, continuous hiring for the Chameleon-eligible roles in conjunction with the program added agility to the resourcing model. New hires, who didn’t know enough to make informed bids, were assigned after the marketplace had matched everyone else.
  4. Choice - After six rounds of the program, Googlers and managers got one of their top three choices 90% of the time. More people are moving around and 80% of employees and managers report they’re satisfied with the process.

There were also some unintended benefits. We found that Chameleon normalized movement in the organization, making it more comfortable for Googlers to talk with their managers about career development. The program also facilitated networking as Googlers reached out to managers to learn more about the opportunities available. We even created a whole culture around the program; “Chameleon” has actually become a verb in the organization. For example, you might hear: “The Product team needs someone on this new launch, can we Chameleon that?” As we think about sharing our lessons with teams across Google and beyond, here’s what we found is needed to make the Chameleon model work:

  • Commitment from leadership and execution teams
  • Employees with transferable skills seeking development opportunities
  • Substantial volume of jobs that require generalist skills
  • Inclusive design and implementation approach
  • Program management with a focus on people development, organizational design and change management

To learn more about the Chameleon Program through an academic lens, please see this Harvard Business School Case Study.