Where are CEOs like Amazon’s Andy Jassy getting the data to inform their return-to-office policies? Unfortunately, too many are getting their data from the same place: an echo chamber of like-minded CEOs who use their feelings and intuitions to make these pivotal decisions. By relying on word of mouth and following their gut, rather than the data, these CEOs could be leading their companies into catastrophe.

During a recent internal fireside chat that was first reported by Insider and then confirmed by Amazon, Jassy defended his top-down return-to-office mandate–a drastic shift from a flexible policy of teams deciding what to do on their own to an obligation to come to the office three days a week. The company said employees knew all along that office policies would evolve with the pandemic–but it didn’t avert a backlash.

When asked for data to support the move, Jassy lacked a good answer. He said that he spoke to “60 to 80 CEOs of other companies over the last 18 months,” and “virtually all of them” preferred in-office work. He admitted it was a “judgment call” that wasn’t widely supported by data and compared it to another major decision that wasn’t supported by data in the past: the launch of the Amazon Web Services cloud unit.

Data over feelings

Is that an apt comparison? There were doubts internally and externally about the launch of AWS because it was an unproven business model. In other words, Amazon was going against the grain and taking a large risk with a huge potential upside if it worked. By contrast, a top-down RTO mandate of three days a week is not at all an “unproven business model.” There’s plenty of evidence around that model, much stronger than the echo chamber of like-minded CEOs. However, Jassy refused to lay out the evidence and relied instead on how he felt–and the validating echo chamber of other CEOs.

Doubling down on the rigid policy, Jassy told staff that “it’s probably not going to work out for you” if you don’t come to the office. Amazon is tracking badge swipe data, and in early August sent a message to some employees, saying “We are reaching out as you are not currently meeting our expectation of joining your colleagues in the office at least three days a week, even though your assigned building is ready.” Amazon says it shares aggregated and anonymized badge-swipe data with managers to provide an overall view of how many members of a team are coming into the office, but doesn’t give managers data about badge swipes by individual employees.

Of course, Jassy is far from the only CEO to launch a rigid, top-down RTO mandate without relying on data and instead focusing on feelings. Consider Starbucks CEO Howard Schultz’s directive for corporate staff to come to the office three days a week this January. According to Schultz, this policy stemmed from his annoyance that corporate employees hadn’t been following a guideline encouraging them to come into the office one or two days a week, as tracked by badge swipes into the building. What a terrible way to make a major RTO policy decision–based on feelings of annoyance about employee badge swipes!

In my discussions with five to 10 corporate leaders each week about best practices for RTO policies, I hear many similar stories. Most CEOs tell me they decided on their RTO approach based primarily on discussing their intuitions about what works best with other corporate leaders. That’s a textbook example of confirmation bias–a cognitive bias where we look for information that confirms our beliefs, and ignore information that doesn’t.

Confirmation bias and RTO

Why didn’t Jassy–or these other leaders–talk to the CEO or CHRO of Atlassian about how well their famous remote model works? Because they would have provided data that contradicts these beliefs.

Yet CEOs keep getting surprised by the negative consequences of their echo-chamber-driven RTO decisions. In May, hundreds of Amazon employees walked out on their first RTO day from the company’s Seattle headquarters during the lunch hour, with signs like “Hell no, RTO!” Following the latest announcement, more than 30,000 employees joined a new Slack channel called “remote advocacy” shortly after the announcement and organized a petition. In the petition, Amazon employees argued, based on research, that remote work improved productivity, recruitment, work/life balance, inclusion efforts, and reduced corporate expenses.

Andy Jassy’s judgment call rejected this data-driven approach, preferring instead to rely on the CEO echo chamber. That’s despite Amazon’s stock performance, which doubled during the pandemic’s remote work era. The dissonance between stock performance and the leadership’s discontent shows the gap between perceptions and reality around RTO.

We know that the perils of echo-chamber-driven approaches to RTO are not simply evident from case studies like with Amazon. A whopping 80% of bosses reported that they regret their initial return-to-office decisions, according to new research from Envoy, which interviewed more than 1,000 U.S. company executives and workplace managers who work in person at least one day per week.

Leaders said they’d do things differently if they knew more about how often employees actually show up at the office and use the amenities. Some are scratching their heads trying to figure out if their in-office policies are even working. Others are finding it tough to lock in long-term property deals without a clue about how their teams will feel about coming into the office down the road. Larry Gadea, Envoy’s CEO and founder, said “many companies are realizing they could have been a lot more measured in their approach, rather than making big, bold, very controversial decisions based on executives’ opinions rather than employee data.”

Or consider another data point, Unispace’s Returning for Good report. Unispace discovered that a surprising 42% of firms with back-to-office rules saw more people quitting than they’d expected. Additionally, about 29% of these companies are having a hard time hiring new talent. So, while bosses knew pushing folks back into the office would shake things up, they didn’t see the big headaches coming.

Unispace adds a twist: it’s all about options. Their study shows that generally, employees say they’re happy (31%), motivated (30%), and excited (27%) to be back at the desk. But those good vibes dip when going back isn’t optional–dropping to 27%, 26%, and 22% respectively. So, people are more into the office comeback when it’s their call, not a mandate. That benefit of optionality is the same thing I find when gathering data from more than two dozen organizations that I helped figure out their approach to RTO.

Do all RTOs fail to use data?

Far from all CEOs rely on intuition and the echo chamber in the return to office. In fact, (surprise, surprise) some actually use data. For example, Marc Benioff, the CEO of Salesforce, said that “for our new employees who are coming in, we know empirically that they do better if they’re in the office, meeting people, being onboarded, being trained. If they are at home and not going through that process, we don’t think they’re as successful.”

Benioff’s comments are supported by the data. A Harvard University working paper studied software engineers at a Fortune 500 company with a main campus spread across two buildings separated by several blocks. The research focused on the impact of physical proximity on feedback, coding output, and retention.

The findings? Engineers who shared a building with all their teammates received 22% more online feedback compared to those who were further apart. However, once the pandemic closed offices, this edge mostly vanished, shrinking to just an 8% advantage in feedback. Interestingly, being in the same location seemed to cut programming output by 24%. This dip was even more pronounced for senior engineers, who saw a 39% drop, largely because they spent more time giving feedback to others. So, while close quarters can boost communication, they may also come with a cost to productivity–especially for the more experienced folks on the team. However, the study found that proximity helped junior folks succeed in the long run. And that’s a fair call for Salesforce to make.

More broadly, Salesforce is adopting a differentiated approach to hybrid work, with different amounts of in-office time not simply for staff with different tenure at the company, but also for different roles. Engineers are looking at just 10 days in the office per quarter, while admin folks need to show up three days a week. Sales and marketing teams need to be in the office four days a week unless they’re out hustling deals. Benioff, won’t even label their setup as a mandate. “I don’t want to force anybody,” he said, warning that pushing the issue too strongly would lead to an outflow of talent. “We don’t want to lose our stars.”

Instead of mandates, Benioff aims to make the office a place people want to be. He’s keeping the door open for employees to negotiate full remote status with their bosses. But he also made it clear that there will always be roles that require some face time.

While that’s not a perfect approach to RTO, it’s much better than most. It hits one of the three required elements of a truly effective RTO: customizing it to your organization’s needs. After all, as Benioff rightly points out, different roles have divergent needs for being in the office. Engineers might only need to come in 10 days a quarter, perhaps at the start and end of a sprint. Admin staff–whose job involves office management–need to cover office duties like helping people with copiers or receiving visitors, so you should make sure there’s always an admin to cover such needs. Accountants can come in a couple of days at the end of the month to close the books and for a week at the end of the quarter.

Just demanding a set number of days per week–as the large majority of major corporate giants have done–is a sign of laziness and conformism. The company’s leaders and HR failed to think through who needs to be in the office and for what purpose. Employees feel a lot of resentment when they come to the office without a good reason to be there, doing the same thing they would do at home.

What a data-driven RTO looks like

Salesforce unfortunately lacked two other key elements of a data-driven RTO: assessing employee opinions and getting their buy-in. Both of these problems are solved with similar tools, killing two birds with one stone.

First, you need to survey your staff on RTO–ideally before the RTO, or after it if you already launched the RTO. At all the companies I worked with to help guide their RTO, whether it’s from the start of the process or to refine it after RTO was already launched, we started with a thorough survey of staff opinions about remote work and the return to office. That survey involves questions around their preferences on RTO, intent to stay with various versions of RTO, whether they would recommend working here to their peers given these versions, their productivity on individual and collaborative tasks at home vs. in the office, and similar questions on well-being, happiness, morale, stress, and so on. The survey is available in the appendix of my best-selling book, Returning to the Office and Leading Hybrid and Remote Teams.

Next, you’ll want to run focus groups. Choose representative staff from a variety of departments, both rank-and-file and managers at all levels of the organization. Make sure to avoid putting managers together with rank-and-file staff in focus groups, as many employees might be reluctant to be fully transparent when those in managerial roles are in the meeting.

Focus groups offer invaluable qualitative data, diving deeper into the “why” behind survey responses. They let you explore nuanced issues like work-life balance, home office setups, and team dynamics that a survey might not fully capture. This is where you can dig into some of the nuanced issues that a survey might not fully capture. For example, why do some employees prefer hybrid work over fully remote? Is it the office amenities, the team camaraderie, or something else? These sessions should ideally be run by a neutral party to encourage candid conversation and, with participant consent, be recorded for later analysis.

Once you’ve gathered this rich data, the next step is synthesis. Combine the qualitative insights from the focus groups with the quantitative metrics from the surveys. This dual approach not only informs the RTO strategy with a well-rounded perspective but also makes it deeply empathetic to employees’ needs and preferences.

Closing the loop is crucial. Share with employees the actionable steps you’ll take based on these discussions. Even if they don’t fully agree with the final decision, this transparent process makes them feel heard, listened to, and consulted.

The result? A significantly higher level of buy-in and reduced resistance to the RTO plan, as employees will feel that their opinions were genuinely considered in shaping the company’s future. Thus, unlike at Amazon, you won’t have employees leaking recordings of internal company meetings to the media and otherwise getting media involved in your company’s business. You also won’t have employees walking out and publicly protesting or signing mass petitions. None of that happened at any of my clients–because the employees felt listened to, respected, and heard, and because RTO policies were customized to the needs of each team and role.

Amazon illustrates the shortcomings of echo-chamber-driven RTO, characterized by employee resistance and low morale. In contrast, a nuanced, data-driven RTO approach provides a balanced and empathetic strategy. Given that research on CEO plans for the future by Nick Bloom and other scholars indicates that there will be more remote work, not less, it seems that many leaders have learned that their initial RTO approach was wrong. Indeed, Envoy finds that over 80% of bosses regret their initial RTO plans, saying they wished they gathered more information and made data-driven decisions. By stepping out of the CEO echo chamber and engaging in meaningful dialogue with their workforce, companies can succeed in navigating the complexities of RTO.

Gleb Tsipursky, Ph.D. (a.k.a. “the office whisperer”), helps tech and finance industry executives drive collaboration, innovation, and retention in hybrid work. He serves as the CEO of the boutique future-of-work consultancy Disaster Avoidance Experts. He is the bestselling author of seven books, including Never Go With Your Gut and Leading Hybrid and Remote Teams. His expertise comes from over 20 years of consulting for Fortune 500 companies from Aflac to Xerox and over 15 years in academia as a behavioral scientist at UNC–Chapel Hill and Ohio State.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.