Introduction

Food waste is a significant global problem. A study by the United Nations Environment Programme (UNEP) estimated that in 2022, 19% of all food available to consumers was wasted, amounting to a total of 1.05 billion tons wasted. This amount of waste has far-reaching environmental, economic, and social consequences: food waste alone accounts for 8–10% of annual greenhouse gas emissions, the cost of this waste is estimated at approximately 1 trillion USD, and the amount of food that is wasted each year would be enough to provide every person in the world who faces hunger with more than a meal per day1. Reducing the amount of food that is wasted would come with substantial societal benefits and is an important global priority.

Corporate dining settings have an important role to play in waste reduction2. As institutions who provide millions of meals annually, even waste reduction techniques that only reduce waste by small percentages can have a very large absolute impact given the scale at which corporate dining entities operate. Corporate dining faces some unique challenges when it comes to reducing waste. Corporate dining programs can control back-of-house waste using various operational changes (e.g., reducing overproduction by collecting data on how much food should be offered during each meal period3,4, but there is an important additional driver of waste associated with corporate dining that programs cannot directly control: waste produced by the consumers, or plate waste. Especially in free, all-you-care-to-eat corporate dining, users are faced with a situation where every day they eat one or two meals at the office, and there is no monetary penalty for taking too much food. Finding ways to reduce user plate waste can prevent significant carbon emissions and lower costs. A problem corporate dining entities face, however, is that hospitality is a priority, and reducing options or creating a negative experience for guests in an effort to reduce waste is often not an option.

This is where behavioral science has a role to play. Behavioral science, a body of knowledge and scientific approach that borrows from disciplines including psychology and behavioral economics, specializes in fostering desirable behaviors – such as food waste reduction – in ways that do not restrict choice or cause negative user experiences5. There is considerable evidence from past studies that behavioral science techniques can be especially useful for promoting pro-sustainability behaviors, including installation of rooftop solar panels6, reduction in meat consumption7, and reduction in food waste8.

One specific technique that has been shown to be particularly effective in past research is the use of social norms messaging. Human behavior is often shaped by norms - the behaviors we believe that others are engaging in9. By making pro-sustainability social norms apparent to people, behavioral scientists can help foster desirable outcomes. One landmark study, for instance, found that messaging that made it clear that the majority of hotel guests opt to reuse their towels (rather than requiring clean ones each day) resulted in a significant decrease in the amount of guests requesting fresh towels - saving significantly on carbon emissions associated with clean linens and reducing labor and energy costs for the business10. Another study found that making social norms about energy consumption apparent to home owners led to significant reductions in energy consumption11.

Despite evidence supporting the effectiveness of norms-focused interventions across sustainability domains, their application to food waste reduction in corporate dining remains under-explored. While previous studies have successfully reduced food waste through environmental modifications12,13, these approaches often involve restricting choices or altering service methods. Such modifications may be unsuitable for corporate dining environments, where employee satisfaction is important. However, corporate dining settings offer unique opportunities for social norm interventions to thrive. Strong social identities and frequent interactions can amplify the impact of social norm messages by tapping into employees’ collective identity and increased sense of observability.

Existing literature finds that prompts or signage can reduce plate waste, especially in universities and restaurants. But no work to our knowledge has tested social norms signage as a lever to reduce food waste. For instance, in a six-week U.S. university trial, a simple printed prompt cut edible food waste by ~ 15% across 19,046 trays, while adding individualized feedback did not further improve outcomes14. In a Swiss restaurant, table place-cards increased requests to take leftovers home versus no prompt15. That said, the National Academies’ review emphasizes that much of the evidence base relies on pre/post designs without proper controls, often tracks intentions or proxy outcomes rather than wasted food itself, and that there are more methodologically weak (Tier-2) than strong (Tier-1) studies available16.

Corporate, all-you-care-to-eat settings meaningfully differ from universities and hotels: price signals are muted (free pricing), the same diners return frequently, and operational constraints often limit plate/tray changes. In these environments, signage must carry more of the behavior-change load.

While corporate dining (on-site cafeterias or subsidized workplace meals) is reasonably widespread among medium and large enterprises (especially in sectors and regions with strong employee benefits culture) its prevalence globally is difficult to quantify. That said, industry reports show that business dining is a substantial market: globally, estimates are that spending on business dining is about $245 billion per year17. However, relative to other sectors like universities or restaurants, the academic literature on corporate cafeteria food waste interventions is limited. A recent review identified only 2 of 63 selected studies focused on workplace canteens, and existing workplace papers tend to describe levels or perceptions rather than test signage content experimentally18,19.

In the present research, we aimed to test the impact that a social norms-focused intervention could have on reducing plate waste in corporate dining at a large technology company (Google). Our approach was to first conduct global survey research in order to get an understanding of what existing social norms toward food waste were like at the company. We were interested in multiple types of social norms around food waste. We measured descriptive norms (what employees believe that other employees do with respect to food waste), injunctive norms (what employees believe that other employees should do with respect to food waste), and personal norms (what employees survey respondents, themselves, say that they do with respect to food waste20. We then planned to use statistics from this study for use in messaging as part of a novel social norms messaging intervention aimed to reduce user food waste in workplace cafes. We used social norms messaging as well as what we call “portion prompting” messages - point-of-decision “nudges”5 that prompt users to be mindful about only taking the amount of food they will eat when making their food decisions in workplace cafes. We planned to validate these interventions as part of two randomized controlled trials (RCTs), one pilot study and one larger research study, with sites across Europe and the United States using difference-in-differences designs.

To increase statistical power and to allow us to gain finer-grained insights into patterns of waste, we opted to measure actual plate waste on a person-by-person basis. This freed us from relying on self-reported waste estimates, which past research has shown are often underestimates21, and it allowed us to get deeper insights than methods that measure waste on a day-by-day basis. To accomplish our person-by-person waste measurement, we used Orbisk’s Orbi machines, machines that allow for the amount of food wasted by each cafe user to be anonymously quantified and for the type of food that was wasted to be identified using a combination of a scale and computer vision methods (orbisk.com).

Our contribution moves beyond single-site pre/post design by using a credible counterfactual and continuous plate-level measurement across multiple cafés. Aligned with the National Academies’ Tier-1 criteria, we (i) implement signage-only interventions (holding serviceware constant within analysis windows and adjusting when an exogenous dishware change occurred), (ii) use objective plate-level waste scanning rather than attitudes or waste proxies, and (iii) estimate difference-in-differences effects instead of simple before–after contrasts16. Our rollout spans multiple cafes in several cities across Europe and the United States, providing cross-site replication and reducing the risk that results are idiosyncratic to a single venue or week. The large number of plate-level scans enables precise estimates of effect sizes.

Together, the present research aimed to gain valuable insights into social norms toward food waste at a large technology company, test the impact of social norms-based interventions to reduce overall user plate waste, and gain unique user-by-user insights into patterns of plate waste. We believe that gaining these in-depth insights in one company can lay the groundwork for a deeper understanding of dynamics around food waste than can be used elsewhere to develop future techniques to continue to tackle the significant societal problem posed by post-consumer waste.

Study 1

Study 1 focused on survey research to understand the descriptive, injunctive, and personal norms around food waste at a large technology company.

Study 1 methods

Participants

We conducted a global survey using an internal online message board that employees can use to learn about events and training opportunities. The opportunity to participate in the survey was made available to all Google employees in more than 50 countries around the world. To find out about the opportunity to participate, employees needed to browse this internal message board and find the listing for the survey. To avoid biasing the respondents, food waste in particular was not mentioned in the marketing for the survey, which instead focused on providing thoughts about food at the company in general. Participants were not compensated for completing the survey. Survey data collection and ethical procedures were approved by Google Ethics & Compliance, all methods were performed in accordance with this approval, and all participants provided written informed consent.

A total of 336 participants completed the survey. Participants were from 20 different countries across 6 continents (each one excluding Antarctica), and the average age of participants was 34.01 (SD = 7.56). Data on gender was not collected.

Procedure

Survey data collection was accomplished using Qualtrics (Qualtrics LLC., Provo, UT, USA). The survey included a total of 9 questions that aimed to understand social norms around food waste at the company and several others meant to either provide stats that could be used in the social norm intervention signage or to gather information to support other workplace initiatives.

Materials

Food waste social norms

Three types of social norms pertaining to food waste were measured. Injunctive social norms – norms about what employees should do – were measured with the following items: “Googlers should care about not wasting food”, “Googlers should do what they can to avoid wasting food”, and “When deciding how much to serve themselves, Googlers should be careful not to take more food than they’ll end up eating”. Descriptive social norms – the perception of what it is normal for employees to actually do – were measured with the following items: “Googlers care about not wasting food”, “Googlers do what they can to avoid wasting food”, and “When deciding how much to serve themselves, Googlers are careful not to take more food than they’ll end up eating”. Finally, personal norms – what the survey participants said they do – were measured with the following items: “I care about not wasting food”, “I do what I can to avoid wasting food”, and “When deciding how much to serve myself, I am careful not to take more food than I’ll end up eating”. Response options for each item were Strongly Agree (5), Agree (4), Slightly Agree (3), Slightly Disagree (2), Disagree (1), and Strongly Disagree (0).

The wording of injunctive, descriptive, and personal norm items was intentionally parallel, allowing direct comparison of average responses across the three norm types. For each of reference, the three question sets are later abbreviated as “Care About Not Wasting Food”, “Take Action to Avoid Wasting Food”, and “Take Right Amount of Food”, each with injunctive, descriptive, and personal norm versions.

Study 1 Results

We first assessed the percentage of survey respondents who agreed with each of the social norms statements about food waste at the company. For this analysis, respondents counted as “agreeing” with a statement if they selected either the Strongly Agree, Agree, or Slightly Agree response option. Three different types of social norms were assessed: injunctive norms (what employees think employees should do about food waste), personal norms (what respondents say they personally do about food waste), and descriptive norms (what respondents think is normal for other employees to actually do about food waste).

The cross-tabulated results shown in Table 1 indicated that employees almost unanimously agree that food waste is an issue that employees should care about and take action on, demonstrating the presence of a clear injunctive norm (> 97% agreement with each injunctive norm statement). Similarly, respondents almost unanimously agree that they, themselves, care about and take action to reduce food waste, demonstrating that the vast majority of employees have a personal norm to do what they can to avoid wasting food (> 97% agreement with each personal norm statement). However, when it comes to perceptions of what other employees actually do (descriptive norms), the results look different. While a clear majority of employees believe that other employees care about and do their best to reduce food waste, the rate of agreement with these statements is much lower (from 69% to 81%) than with the other social norms statements.

Table 1 Cross-tabulation of the agreement percentage with each food waste social norm statement.
Full size table

We next went beyond assessing rates of agreement with each of these statements by comparing the strength with which employees agreed with them. This analysis allowed us to understand whether injunctive norms, personal norms, and descriptive norms around food waste at the company differed significantly from one another in strength.

Results, shown in Fig. 1, demonstrate that for each question, descriptive norms (what respondents think employees actually do with regard to food waste; labeled “Employees Do”) were significantly weaker than both injunctive norms (what respondents think employees should do with regard to food waste; labeled “Employees Should”) and personal norms (what respondents say they actually do with regard to food waste; labeled “I do”); all paired t-test ps < 2E-16. The effect size associated with these differences was quite large in all cases (Cohen’s d > 1.2). This demonstrates that the food waste behaviors that employees see in their workplace cafes do not align with the food waste behaviors they think should happen or that they personally engage in themselves.

Fig. 1
figure 1

Bar charts that represent average agreement with food waste social norm items. Bars show responses to descriptive norms (“Employees Do”), injunctive norms (“Employees Should”), and personal norms (“I Do”) across three domains: caring about not wasting food, taking action to avoid waste, and taking the right amount of food.

Full size image

Interestingly, there was a smaller but still significant difference between injunctive norms and personal norms with regard to food waste for two out of the three prompts (Take Action to Avoid Wasting Food, t(335) = 4.81, p = 2.3E-6; Cohen’s d = 0.29; Take Right Amount of Food, t(335) = 2.99, p = .003; Cohen’s d = 0.19). This suggests that there is a small but potentially meaningful gap between what individual employees feel they should do vs. what they report actually doing with regard to food waste – an intention-action gap.

Additionally, there were small but significant differences between personal norms for Care About Not Wasting Food and both Take Action to Avoid Wasting Food (t(335) = 4.67, p = 4.4E-6; Cohen’s d = 0.21) and Take Right Amount of Food (t(335) = 4.93, p = 1.3E-6; Cohen’s d = 0.26). The strength with which employees agree that they care about avoiding food waste is greater than the strength with which they report actually engaging in anti-food-waste behaviors.

Study 1 discussion

The survey research in Study 1 uncovered key insights into social norms about food waste at the technology company involved in this research. Employees showed near universal support that food waste should be avoided at the company. Employees also almost unanimously agree that they are doing what they can to reduce food waste. However, the rate of agreement that employees in general care about and take action to reduce food waste is much lower, suggesting that employees see a substantial amount of plate waste around them when they eat at workplace cafes.

In addition to treating agreement as binary, we also assessed the strength of agreement about statements related to food waste. We found that the descriptive norm to do what you can to avoid food waste is much weaker than the injunctive and personal norms – the food waste behaviors that employees observe when they eat in workplace cafes does not match up with the behaviors they think should happen or what they report doing. Interestingly, we also found evidence that personal norms regarding food waste are weaker than injunctive norms – in other words, there is an intention-action gap where employees’ self-reported food waste behavior does not always match their ideals.

One of the striking findings here is that employees almost unanimously say they care about food waste and do what they can to avoid it, but at the same time many employees don’t believe that their coworkers care about food waste or do what they can to avoid it.

An explanation for this apparent contradiction may come from a different finding - employees don’t always live up to their food waste ideals. The small but significant gap between what employees say they should do and what they actually do could indicate that they sometimes waste more than intended. If these lapses are common, employees are likely to notice instances of high-waste plates regularly. Since they care about food waste, as the data shows, this discrepancy between what they think should happen (little waste) and what they see in front of them (at least a few high-waste plates among their peers) might be especially salient. Over time, this may reinforce perceptions that the average employee does not care as much about food waste. This could explain the comparatively weak descriptive norm despite very high injunctive and personal norms to prevent food waste.

Importantly, this comparatively weak descriptive norm could have negative implications for plate waste. There is substantial evidence that descriptive norms can strongly and negatively shape behavior even in situations where the people involved care about an issue22,23. Even though employees care about preventing food waste, their perception that their peers don’t care as strongly might normalize waste (“If others don’t care, my waste isn’t a big deal”). If so, working to correct employees’ misconceptions that their peers do not care very strongly about food waste could result in an overall reduction of food waste. We empirically test this idea in Studies 2 and 3.

Study 2

In Study 2, we used the insights from Study 1 to design and test a social norms messaging intervention aimed to reduce plate waste in workplace cafes. We also used Orbisk technology to gain anonymous user-by-user insights on patterns of waste by employees. The RCT was designed to focus on lunch rather than breakfast since lunch has about 3 times as many daily users as breakfast and as a result there were concerns about statistical power with breakfast.

Study 2 methods

Procedure

The pilot study RCT was conducted in workplace cafes located in London, England and Dublin, Ireland in March and April of 2023. One cafe in each city was included in the research. Both cafes include all-you-care-to-eat dining - users enter the cafe and are free to take whatever food they like. All employees have access to the cafes, and users are not charged for their meals. Both the cafes in London and Dublin offered a diverse menu, categorized by multiple main dish options and a few side dish options on a rotating menu. The cafes were located in company office buildings.

Study design

The RCT used a difference-in-differences design to test the effectiveness of the social norms and portion prompting intervention in reducing user plate waste. Difference-in-differences is a widely used quasi-experimental approach that estimates causal effects by comparing pre-post changes in outcomes between treatment and comparison groups under the assumption that both groups would have followed parallel trends in the absence of an intervention24. In this framework, an intervention is considered effective if the treated group’s outcome changes more favorably relative to the comparison group, whether by producing a larger decrease or by preventing an increase that would otherwise have occurred25. Applied to the current study, if treatment led to an overall reduction of waste or a smaller increase in waste compared to control, both would provide evidence of waste prevention.

We varied whether intervention signage was active or not for the duration of the 6-week study, and anonymous user-by-user data on plate waste was collected using Orbisk’s Orbi machines during the entire study period.

Table 2 shows the design of the pilot study RCT. The design allows us to compare the change in waste from Period 1 to Period 2 across both cafes and the change in waste from Period 2 to Period 3 across both cafes.

Table 2 The research design of the pilot study RCT.
Full size table

Materials

Social norms intervention signage

The social norm signage was created using actual stats on employees’ feelings toward food waste and sustainability collected in Study 1. The primary goal of the signage was to correct employees’ misperceptions that many employees do not care about preventing food waste. The approach we used was to create four different signs that followed a consistent pattern. Four different signs were used for two reasons. Multiple different signs allowed us to convey more statistics to employees, increasing the extent to which employees could gain an updated understanding that employees really do care about preventing food waste. Additionally, having multiple signs also increases how salient each sign is and increases the chances of employees reading the different signs and engaging with the messaging.

In large text at the top, we included a key stat from our survey. For example, the signs shown in Fig. 2 announce that “97% of Googlers think Googlers should do what they can to avoid wasting food” and “96% of Googlers care about being environmentally friendly”. The large font of the text showing the percentages was meant to highlight the large levels of agreement about the statements that were shown. Additionally, placing “97% of Googlers”, for example, in large text at the top of the sign was meant to be attention-grabbing – we wanted employees to see that and think “97% of Googlers do what?” and then read the sign more closely. The bright colors of the signage were similarly meant to increase how salient the signage was. The image of the anthropomorphized earth and the text “Here’s the Deal: Less food waste = healthier planet” was included to make it clear that food waste has implications for the health of the planet. Finally, at the bottom of the signage, we include our call-to-action: “Join in to stop food waste! Take what you’ll eat.” The “join in” phrase, in particular, was included to elicit the idea of collective action that all employees could contribute to. “Take what you’ll eat” was the behavior we wished to highlight for what employees could do to reduce food waste. This was chosen over other phrases related to “finishing your plate” or “having a clean plate” because these phrases emphasize eating whatever it is you put on your plate. The target behavior here, however, is taking the right amount of food for you rather than making sure you eat what you take – the former would actually reduce the amount of food employees took, allowing the cafes to reduce their carbon footprint, while the latter would not change the carbon footprint of workplace cafes and could inadvertently prompt employees to eat more than they would like to eat.

Fig. 2
figure 2

Examples of social norms signage.

Full size image

The signs were placed as table cards on all tables in the research site cafes. Larger signs were placed around the offices that the cafes were located in with a focus on common areas and especially the workplace kitchens where employees can get snacks and beverages.

Portion prompting intervention signage

The social norms signage was designed to make it clear to employees that their peers care about preventing food waste and to motivate employees to avoid waste. The portion prompting signage was designed to be a point-of-decision call-to-action to make it clear to employees that if they want to prevent food waste, they need to not take more than they will eat.

There were two portion prompting signs, shown in Fig. 3. The first one was placed near the cafe entrances, and it read “Join in to stop food waste! Take what you want, take what you need, take what you’ll eat!” The second sign was placed right at the beginning of food stations where employees made their decision about what food they wanted to take. At the top, it read “Join in to stop food waste!”, at the bottom, it read “take what you’ll eat”, and in the middle it showed spoons with different amounts of food in them that read “Taste”, “Half”, and “Full”. Both signs included key elements in common with the social norms signage: both elements of the social norms signage call to action (“Join in to stop food waste!” and “Take what you’ll eat.”) and the anthropomorphized earth. This was done to tie the signage together and make it clear that it was a single messaging campaign. These callbacks were also designed to make employees remember the stats from the social norms signage and activate the motivation to reduce waste.

Fig. 3
figure 3

Portion prompting signage.

Full size image
Measures

We used Orbisk’s Orbi machines (orbisk.com) to measure plate waste on an anonymous user-by-user basis in each of the study cafes. The Orbi machines have two key components. At the top of the device is a tablet that has a camera, and at the bottom of the device is a scale. A trash bin is placed on the scale that plate waste is scraped into, and the weight of the trash bin is zeroed out.

To complete a scan, a trained team member simply holds a plate underneath the tablet’s camera. The camera automatically takes a picture that shows the plate and the plate’s waste. Then, the team member dumps the contents of the plate into the trash bin, and the scale below weighs the amount of waste. The weight of the waste is then associated with the picture the tablet took. Afterwards, Orbisk uses advanced machine learning and computer vision techniques to identify what type of food was on the plate and makes estimates of how much of the overall weight of the waste was attributable to which types of food (e.g., if a plate had 50 g of waste on it, Orbisk provides an estimate of how much of the waste was due to the eggs it identified and how much was due to the potatoes it identified). Waste identification was beyond the scope of the present work, and as such our key dependent variable was the amount of waste included in each scan. For an example of what an Orbi scan looked like, see Fig. 4.

Dish return worked the same at both cafes: when they were finished with their meal, users placed their plate, bowl, or tray on a conveyor belt that delivered their receptacles back-of-house. Importantly, staff was trained to make each Orbi scan correspond to one person’s waste. If a person returned a single plate, the associated scan should include a single plate. If a person returned a plate and a bowl, the scan should include both the plate and the bowl. This way, each Orbi scan represents one user’s waste. The staff was instructed to scan each person’s food receptacles regardless of whether it had food on it - this way, we could capture the amount of users that had zero waste. Users did not know Orbisk scanning was occurring, and each scan was anonymous - there was no way to tie a scan back to the user.

Fig. 4
figure 4

Examples of images of food waste from Orbi scans.

Full size image
Data cleaning

When processing the data from the Orbi scans, we recoded weight associated with food scraps for parts of food that were not meant to be eaten (e.g., chicken bones) to be equal to 0, since for reducing this kind of waste was not the focus of the current research. We additionally removed scans with total weights that were more than three standard deviations above the mean to remove extreme outliers from analyses. Throughout the pilot study, a total of 3,344 eligible Orbi scans were completed and included in analysis.

Study 2 results

Impact of the social norms and portion prompting intervention on plate waste reduction

We first assessed whether the social norms and portion prompting intervention was impactful at reducing food waste. Our research design was a difference-in-differences design with two stages. In the first stage, we planned to assess the difference in the change in waste between one cafe going from baseline to treatment (i.e., going from having no signage active to having signage active) and the other cafe going from baseline to control (i.e., going from no signage active to continuing to have no signage active). In the second stage of the difference-in-differences design, we planned to assess the difference in the change in waste between one cafe removing their signage and the other cafe activating it for the first time. Because these were considered to be two different research questions, two separate analyses were conducted.

This difference-in-difference design is more powerful than a simple pre-post design because it can account for seasonal impacts – patterns of plate waste might change over time for reasons unrelated to our intervention. By comparing the change in waste across cafes, however, we can account for this seasonal change to see if the intervention prevents waste.

For the first stage of the difference-in-difference design, we compared the change in plate waste of the Dublin cafe going from baseline (no signage) to control (no signage) to the change in waste of the London cafe going from baseline (no signage) to treatment (social norms and portion prompting signage active). The results of the first stage of our difference-in-difference analysis are shown in Fig. 5.

Fig. 5
figure 5

Difference-in-differences results from the first stage of the RCT are shown. Error bars reflect standard errors. The dotted line indicates moving from no signage to no signage (i.e., baseline to control), while the solid line indicates moving from no signage to signage (i.e., from baseline to treatment).

Full size image

Results showed that when moving from Period 1 to Period 2, the Dublin cafe saw a statistically significant 12 g increase in waste-per-person (Period 1 - Baseline: 27.55 g waste-per-person, SD = 46.32 g, N = 170 scans; Period 2 - Control: 39.55 g waste-per-person, SD = 52.90, N = 595 scans; Difference = 12.00 g increase; t(763) = − 2.68, p = .008). This reflected a 43.5% increase in waste-per-person during this period. The London cafe, which activated their social norms and portion prompting signage at the beginning of Period 1, experienced a small uptick (3.23 g per person) in waste during the move from Period 1 to Period 2, but this change was not statistically significant (Period 1 - Baseline: 28.67 g waste-per-person, SD = 49.99 g, N = 678 scans; Period 2 - Treatment: 31.90 g waste-per-person, SD = 47.67, N = 707 scans; Difference = 3.23 g increase; t(763) = -2.68, p = .008). This was equivalent to only an 11.1% increase in waste-per-person.

We calculate our estimate of the impact of the intervention by subtracting the change experienced by the Dublin site (12.00 g increase) and the change experienced by the London site (3.23 g increase) and arrive at the estimate that the social norms and portion prompting intervention prevented 8.77 g of waste per-person. The difference in the percentage changes between the sites (43.5% increase for Dublin and 11.1% increase for London) suggested that the social norms and portion prompting intervention helped prevent a 32.4% increase in waste-per-person. A 2 (Cafe) by 2 (Period) ANOVA indicates that this change is marginally statistically significant (Main Effect of Cafe: F(1, 2146) = 4.73, p = .030, η2 = 0.002; Main Effect of Period: F(1, 2146) = 10.02, p = .002, η2 = 0.005; Interaction: F(1, 2146) = 2.97, p = .085, η2 = 0.001).

We next assessed stage 2 of the difference-in-differences design (moving from Period 1 to Period 2), when London removed their signage and Dublin activated theirs for the first time. Results are shown in Fig. 6.

Fig. 6
figure 6

Difference-in-differences results from the second stage of the RCT are shown. Error bars reflect standard errors. The dotted line indicates moving from signage to no signage (i.e., treatment to control), while the solid line indicates moving from no signage to signage (i.e., from control to treatment).

Full size image

Results showed that when moving from Period 2 to Period 3, the Dublin cafe saw a non-statistically significant 1.63 g decrease in waste-per-person (Period 2 - Control: 39.55 g waste-per-person, SD = 52.90, N = 595 scans; Period 3 - Treatment: 37.92 g waste-per-person, SD = 51.07, N = 676 scans; Difference = 1.63 g increase; t(1269) = 0.56, p = .576). This reflected a 4.1% decrease in waste-per-person during this period. The London cafe, which removed their social norms and portion prompting signage at the beginning of Period 3, experienced a significant increase (6.93 g per person) in waste when moving from Period 2 to Period 3 (Period 2 - Treatment: 31.90 g waste-per-person, SD = 47.67, N = 707 scans; Period 3 - Control: 38.83 g waste-per-person, SD = 53.47, N = 518 scans; Difference = 6.93 g increase; t(1223) = -2.39, p = .017). This was equivalent to a 17.8% increase in waste-per-person.

By subtracting the change experienced by the Dublin site (1.63 g decrease) and the change experienced by the London site (6.93 g increase), we arrive at the estimate of the difference between adding signage and removing signage of 8.56 g of waste per-person. The difference in the percentage changes between the sites (4.1% decrease for Dublin and 17.8% increase for London) suggested that the difference between adding and removing social norms and portion prompting signage is a 21.9% decrease in the amount of waste-per-person that would otherwise be expected. A 2 (Cafe) by 2 (Period) ANOVA indicates that this change is statistically significant (Main Effect of Cafe: F(1, 2492) = 3.02, p = .083, η2 = 0.001; Main Effect of Period: F(1, 2492) = 2.03, p = .154, η2 = 0.001; Interaction: F(1, 2492) = 4.32, p = .038, η2 = 0.002).

Together, analyses of both stage 1 and stage 2 of the difference-in-differences design suggest that the social norms and portion prompting intervention signage was effective at reducing plate waste.

Distribution of plate waste at workplace cafes

One of the main insights that user-by-user data on plate waste allows for is an understanding of the distribution of waste at workplace cafes. One possibility is that employees all waste roughly the same amount of food as each other. Another possibility is that there is very wide variation, with some employees wasting a great deal more than others. If this were the case, it could be that a relatively large portion of the waste can be attributed to a small portion of users.

Histograms of the amount of waste produced in the Dublin and London study cafes are shown in Fig. 7.

Fig. 7
figure 7

The distribution of per-user plate waste during lunch for both cafes.

Full size image

Results across both the Dublin and London cafes were nearly identical. At both cafes, the vast majority of users produce very little waste. The median for plate waste at lunch was 16.1 g in Dublin and 5.5 g in London, equivalent to .57oz and .10oz, respectively. Additionally, 40.2% of Dublin users and 46.8% of London users wasted zero food at all.

However, the distribution has a very large tail, with a small proportion of users ending up with large amounts of waste on their plates. The 90th percentile for plate waste at lunch was 110.3 g in Dublin and 100.2 g in London, equivalent to 3.9oz and 3.5oz, respectively.

A very important question arises from these distributions: How much of the overall waste produced by workplace cafes is attributable to just the top 10% of high-waste plates? (Here “high-waste plate”, refers to an Orbisk scan that has a high amount of waste. Orbisk scans were done on an anonymous user-by-user basis, so the focus here is not on individual plates, per se, but on cases when a user had a high amount of waste during a given meal period. We use the phrase “high-waste plates” as opposed to “high-waste user” to avoid insinuating that there are specific users who chronically have high waste, as this research does not provide any evidence that is the case. As we discuss below, we believe it is more likely that different users have high-waste plates on different days rather than there being a very high rate of true “high-waste users” who have high waste during most meal periods at the company.) Results indicate that, for Dublin, the top 10% of plates by waste weight account for 42.5% of all waste. For London, the number was even higher - the top 10% of plates by waste weight accounted for 47.2% of all waste. The fact that so few plates account for so much of the overall user waste suggests that targeting these high-waste plates could be a high-leverage strategy for reducing overall waste in future work.

Study 2 discussion

In Study 2, we conducted a pilot study to test whether social norms and portion prompting intervention reduced lunchtime food waste and to examine waste patterns at the user level.

The difference-in-differences results showed that introducing the social norms and portion prompting intervention successfully prevented ~ 22% to ~ 32% increases in waste-per-user. This indicates that correcting employees’ perceptions about descriptive norms around food waste can have a measurable impact on waste reduction. This suggests that when employees are made explicitly aware that their coworkers care about reducing food waste and are themselves trying to do what they can to avoid waste, they are more motivated to follow through on their own intentions to avoid waste.

We also observed detailed patterns of waste through anonymous person-by-person waste scanning using Orbisk’s Orbi technology. We found that the vast majority of employees waste very little food - in both Dublin and London, the median for waste at lunch was significantly less than an ounce of waste (.57oz and .10oz, respectively), and over 40% of users at both cafes wasted zero food at all. However, a small proportion of employees ended up wasting a significant amount of food, and the top 10% of plates by waste weight accounted for ~ 45% of all waste at each of the sites.

Interestingly, these patterns of waste align very closely with our hypothesis about why employees may almost unanimously say that they care about food waste but do not believe as strongly that other employees care that we found in our survey research in Study 1. The vast majority of employees do waste very little food – Study 2 has provided empirical evidence of this. However, a small number of employees end up with a large amount of waste on their plates, and these high-waste plates are likely very salient to other employees, especially given that they care so much about preventing food waste. Frequent exposure to such plates may reinforce the perception that others do not care, and correcting this misperception is likely why the intervention was successful.

Study 2 provided meaningful findings. It demonstrated that the social norms and portion prompting intervention could be an effective way to reduce user food waste, and it uncovered fine-grained insights into the distribution of plate waste in corporate dining settings. In Study 3, we ran a larger replication study to assess whether these findings were replicable in sites beyond the UK.

Study 3

In Study 3, we tested whether the findings from Study 2 would replicate at different workplace cafes in different cities throughout the world.

Study 3 methods

Procedure

The research study RCT was conducted in workplace cafes located in the following cities: Chicago, USA; Mountain View, USA; Amsterdam, Netherlands; Stockholm, Sweden; and Madrid, Spain. The research took place in August and September of 2023. One cafe from each city was included in the research. All cafes include all-you-care-to-eat dining, users are not charged for their meals, and cafes are located within company office buildings. Orbisk’s Orbi machines were used in all sites to collect user-by-user data on plate waste. The Madrid site used an experimental version of the Orbi that relied solely on visual information and did not include a scale, but the readings from this version differed significantly from the readings from the standard scale-based Orbi and as a result the Madrid cafe was excluded from analysis. Like Study 2, the RCT focused on lunch.

Study design

The Study 3 RCT used a very similar difference-in-differences design as was used in Study 2. See Table 3 for details on conditions by site. The initial design was for Mountain View to collect baseline data in Period 1 like the rest of the sites and to then activate the intervention in Period 2. However, for practical reasons, the cafe needed to make a mid-study adjustment to their dishware during Period 1. Because the type of dishware has been shown to impact the amount of food waste users produce26, we excluded Mountain View from analysis for Period 1 and shifted their timeline. This shift means that they are not included in the first leg of the difference-in-differences analysis (i.e., analyzing change in waste from Period 1 to Period 2), but they are included in analysis for the second leg (i.e., analyzing changes in waste from Period 2 to Period 3). Note that removing Mountain View from the analysis entirely does not change any of the intervention findings detailed below.

Table 3 The research design of the research study RCT.
Full size table

A change from the design of Study 2 is that here, we opted for the cafe that activated the signage in Period 2 to leave the signage up for Period 3. In Study 2, the cafe that activated the signage in Period 2 removed it for Period 3. This allowed us to conduct an informal internal test of whether leaving the signage up for longer periods continues to reduce waste.

Materials

All signage and signage placement for Study 3 was identical to that of Study 2.

Measures

Orbisk scanning was carried out in an identical manner to that described in Study 2.

Data cleaning

As in Study 2, we recoded weight associated with inedible food scraps to be equal to 0. We also removed scans with total weights that were more than three standard deviations above the mean from analysis to remove extreme outliers from analyses. Throughout the research study, a total of 9,806 eligible Orbi scans were completed and included in analysis.

Study 3 results

Impact of the social norms and portion prompting intervention on plate waste reduction

We first tested whether the social norms and portion prompting intervention was again impactful at reducing food waste. Like in Study 2, we assessed both stages of the difference-in-differences design separately.

The first stage of the difference-in-difference design focused on comparing changes in user plate waste between Period 1 and Period 2. We compared the change in waste that Amsterdam experienced going from no signage to having signage active to the change in waste experienced by the Chicago and Stockholm cafes, which did not activate signage in either Period 1 or Period 2. The Mountain View cafe was excluded from this analysis due to a change in dishware in the middle of Period 1. Results are shown in Fig. 8.

Fig. 8
figure 8

Difference-in-differences results from the first stage of the RCT are shown. Error bars reflect standard errors. The dotted line indicates moving from no signage to no signage (i.e., baseline to control), while the solid line indicates moving from no signage to signage (i.e., from baseline to treatment).

Full size image

Results showed that when moving from Period 1 to Period 2, the Chicago cafe saw a statistically significant 16.33 g increase in waste-per-person (Period 1 - Baseline: 42.81 g waste-per-person, SD = 49.77 g, N = 1477 scans; Period 2 - Control: 59.14 g waste-per-person, SD = 54.79, N = 619 scans; Difference = 16.33 g increase; t(2094) = -6.65, p = 4E-11). This reflected a 38.1% increase in waste-per-person during this period. The Stockholm cafe, which also did not activate signage in either Period 1 or 2, experienced very little change in plate waste between the two periods (Period 1 - Baseline: 36.69 g waste-per-person, SD = 44.08 g, N = 1033 scans; Period 2 - Control: 37.75 g waste-per-person, SD = 46.83, N = 941 scans; Difference = 1.06 g increase; t(1972) = − 0.52, p = .603), amounting to a non-significant 2.9% increase. Finally, the Amsterdam cafe, which activated the intervention signage in Period 2, saw a small increase in waste-per-person as well (Period 1 - Baseline: 55.42 g waste-per-person, SD = 59.99 g, N = 525 scans; Period 2 - Treatment: 58.41 g waste-per-person, SD = 55.86, N = 458 scans; Difference = 2.99 g increase; t(981) = − 0.81, p = .420), equivalent to a non-significant 5.4% increase in waste.

A 2 (Condition) by 2 (Period) ANOVA, controlling for cafe, indicates that the change in waste in the cafes that did not adopt the signage increased significantly more than the cafe that did adopt the signage, as indicated by a significant Period by Condition interaction (Main Effect of Condition: F(1, 5060) = 83.68, p < 2E-16, η2 = 0.020; Main Effect of Period: F(1, 5060) = 13.65, p = 2E-4, η2 = 0.003; Period by Condition Interaction: F(1, 5060) = 5.46, p = .019, η2 = 0.001; Effect of Cafe: F(1, 5060) = 49.17, p = 3E-12, η2 = 0.010).

To arrive at an estimate of the overall waste saved by the intervention from the first stage of the Study 3 RCT, we ran a regression model that includes the same variables as those in the above ANOVA model. The Condition by Period interaction term estimate from this model was 8.70, indicating that the social norms and portion prompting intervention signage saved an estimated 8.70 g of waste in the site that activated the signage.

We next assessed stage 2 of the difference-in-differences design – moving from Period 2 to Period 3. During this time, Amsterdam kept its signage up while the other sites (Chicago, Stockholm, and Mountain View) activated the intervention signage for the first time. Results are shown in Fig. 9.

Fig. 9
figure 9

Difference-in-differences results from the second stage of the RCT are shown. Error bars reflect standard errors. The dotted line indicates moving from signage to signage (i.e., treatment to treatment), while the solid line indicates moving from no signage to signage (i.e., from control to treatment).

Full size image

Results showed that when moving from Period 2 to Period 3, the Amsterdam cafe saw a statistically significant 14.02 g increase in waste-per-person (Period 2 - Treatment: 58.41 g waste-per-person, SD = 55.86, N = 458 scans; Period 3 - Treatment: 72.43 g waste-per-person, SD = 53.75, N = 413 scans; Difference = 14.02 g increase; t(869) = -3.76, p = 2E-4). This reflected a 24% increase in waste-per-person during this period.

The rest of the cafes activated their social norms and portion prompting intervention signage for the first time. The Chicago cafe saw a significant 18.22 g decrease in plate waste once their intervention signage was activated (Period 2 - Control: 59.14 g waste-per-person, SD = 54.79, N = 619 scans; Period 3 - Treatment: 40.92 g waste-per-person, SD = 52.91, N = 1073 scans; Difference = 18.22 g decrease; t(1690) = 6.73, p = 2E-11), equivalent to a 30.1% reduction in waste-per-person. The Stockholm cafe saw a non-significant 3.24 g decrease in plate waste once their intervention signage was activated (Period 2 - Control: 37.75 g waste-per-person, SD = 46.83, N = 941 scans; Period 3 - Treatment: 34.51 g waste-per-person, SD = 46.87, N = 1379 scans; Difference = 3.24 g decrease; t(2318) = 1.64, p = .101). Finally, the Mountain View cafe saw a non-significant 0.56 g increase in plate waste once their intervention signage was activated (Period 2 - Control: 37.35 g waste-per-person, SD = 56.59, N = 1001 scans; Period 3 - Treatment: 37.91 g waste-per-person, SD = 57.59, N = 887 scans; Difference = 0.56 g increase; t(1886) = − 0.21, p = .832).

A 2 (Condition) by 2 (Period) ANOVA, controlling for cafe, indicates that the change in waste in the cafes that adopted the new signage was significantly different from the change in waste from the cafe that had already had the signage up for 2 weeks, as indicated by a significant Period by Condition interaction (Main Effect of Condition: F(1, 6771) = 192.56, p < 2E-16, η2 = 0.030; Main Effect of Period: F(1, 6771) = 8.16, p = .004, η2 = 0.001; Period by Condition Interaction: F(1, 6771) = 30.42, p = 4E-8, η2 = 0.004; Effect of Cafe: F(1, 6771) = 27.31, p = 2E-12, η2 = 0.008).

To arrive at an estimate of the overall waste saved by the intervention from the second stage of the Study 3 RCT, we ran a regression model that includes the same variables as those in the above ANOVA model. The Condition by Period interaction term estimate from this model was − 21.30, indicating that the social norms and portion prompting intervention signage saved an estimated 21.30 g of waste in the site that activated the signage.

Together, these results show that the social norms and portion prompting intervention is a replicable technique for reducing plate waste in workplace cafes.

Distribution of plate waste at workplace cafes

As in Study 2, we also assessed the distributions of plate waste in each study cafe. For histograms of per-user plate waste at the Study 3 cafes, see Fig. 10, and for additional distribution details, see Table 4.

Fig. 10
figure 10

The distribution of per-user plate waste during lunch for each cafe.

Full size image
Table 4 Distribution details of lunch per-user plate waste for study 3 cafes.
Full size table

While there is some fluctuation in the distributions of waste from cafe to cafe, there is a remarkably similar pattern of waste across all cafes included as part of this research: the vast majority of users waste very little food, but there is a long-tailed distribution, and the top 10% of plates by waste weight accounts for an outsized proportion of the overall waste.

Study 3 discussion

The goal of Study 3 was to test whether the findings we observed in Study 2 were replicable across new, geographically diverse workplace cafes. We found that this was consistently the case.

We again found consistent evidence that the social norms and portion prompting intervention successfully reduced plate waste in workplace cafes. We also found that the key plate waste insights from Study 2 replicated in the Study 3 cafes. At each cafe, we observed the same general pattern of waste: the majority of employees wasted very little food, but a small minority of high-waste plates accounted for a substantial portion of the overall amount of waste produced by each cafe. Study 3 provides substantial additional confidence that these findings are generalizable across a wide array of workplace cafes.

General discussion

This global study on food waste at a large technology company had three primary goals. One goal was to examine existing social norms around food waste. A global survey at the company revealed strong injunctive and personal norms for preventing plate waste: respondents almost unanimously agreed that food waste is a problem and that they should do what they can to avoid it, and most reported taking action to prevent waste themselves. However, the descriptive norm of avoiding food waste was much weaker - even though employees almost unanimously reported that they avoid waste, a large proportion of survey respondents did not agree that company employees, in general, take similar action to avoid waste. Another research goal was to test the impact that a social norms-focused intervention can have on plate waste prevention. Across two research studies with sites throughout Europe and the United States, we found consistent evidence that our social norms and portion prompting intervention successfully prevented user plate waste. Our final goal was to gain user-by-user insights on how much food is wasted in workplace cafes using Orbisk’s Orbi technology. We found that patterns of food waste were very consistent across every cafe and meal period included in the research: the vast majority of employees waste very little food, but the minority of high-waste plates (i.e., the top 10% of plates by waste) account for a large proportion of the overall waste (estimates ranged from 31% to 62% across all cafes and meal periods). These results provide a clear picture of how employees see food waste, the patterns of food waste behaviors behavior employees engage in, and ways to prevent food waste in workplace cafes and beyond. The implications of these findings, as well as the limitations of the research, are discussed below.

Food waste social norms at a large technology company

The pattern of social norms observed at the company presents an intriguing inconsistency. Employees overwhelmingly agreed that company employees should avoid wasting food (indicating a widespread injunctive norm), they overwhelmingly report that they themselves do what they can to avoid food waste (indicating a widespread personal norm), but they are much less likely to endorse the idea that other employees do what they can to avoid food waste (indicating a comparatively weaker descriptive norm). This discrepancy raises an important question: if nearly everyone reports caring about and acting to prevent food waste, why do so many believe their coworkers do not?

In the Study 1 Discussion, we proposed a hypothesis to explain this pattern. Although personal norms to avoid waste were very strong, they were slightly but statistically significantly weaker than injunctive norms: The strength of belief that employees should do what they can to avoid waste was at near-ceiling levels, but the strength of belief that respondents, personally, actually do what they can to avoid waste was slightly weaker. Although social desirability bias may contribute to these uniformly high ratings, the fact that self-ratings were measurably lower than ideal suggests that respondents were at least somewhat candid about occasional lapses. This discrepancy suggests an intention–action gap – employees may strongly intend to prevent waste but occasionally fall short. If so, a small proportion of otherwise well-intentioned employees may produce high-waste plates at each meal. As a result, employees dining in workplace cafes would routinely observe a few instances of substantial waste. Because employees generally view food waste as wrong, these high-waste plates are likely to stand out, consistent with research showing that morally disapproved behaviors are especially salient to observers27. Repeated exposure to such instances may lead employees to infer (incorrectly) that their coworkers do not care much about preventing waste. This perception, in turn, could weaken their own motivation to minimize waste, as individuals often take behavioral cues from the perceived norms of those around them28.

In Studies 2 and 3, user-level data collected with Orbisk technology supported our hypothesized pattern. The vast majority of employees wasted very little food, likely reflecting widespread motivation to prevent waste. However, a small subset of users produced disproportionately high-waste plates, consistent with the proposed intention-action gap in which individuals who value waste reduction occasionally fail to meet their own standards. These studies advance prior work by directly measuring actual food waste through precise weighing, rather than relying on self-reported estimates, which are known to substantially underestimate waste21.

Using social norms as a driver of plate waste prevention

Building on these findings, we developed an intervention designed to correct employees’ misperceptions of the descriptive norm around food waste. The goal was to make it clear that most employees do, in fact, care about and act to prevent waste, thereby reinforcing waste prevention as the social norm. To accomplish this, we displayed signage that mirrored back results from Study 1, using real statistics to highlight employees’ shared commitment to prevent food waste. In addition to the social norms messaging, the intervention included portion prompting signage that provided a concrete behavioral cue: encouraging employees to take only what they intended to eat. Together, these components were intended to strengthen motivation to reduce waste and offer a simple, actionable strategy for doing so.

Using our difference-in-differences design in two RCTs, we found that, across a geographically diverse set of six workplace cafes, the intervention successfully prevented food waste. The exact estimates varied, but the average of the amount of waste prevented across each model was 11.8 g of waste per-person. Using the average waste-per-person at lunch during baseline across each of the cafes (38.2 g per person), this translates to the intervention preventing roughly 31% compared to the amount of waste that would have been expected without the intervention. Note that while the exact estimates varied, the results were consistent. In each leg of our difference-in-difference design across both RCTs, we found evidence that the intervention prevented waste. In some cases we found evidence that the intervention led to a decrease in waste and in some cases we found evidence that it prevented an increase in waste – both of which indicate that the intervention played a causal role in preventing waste that would have otherwise occurred25.

These results show that correcting social norms around food waste and giving cafe-goers the specific behavior they can engage in to avoid waste can be an effective way to reduce plate waste in corporate dining environments. This aligns with research demonstrating that social norms messages can promote pro-environmental behaviors29, including increasing leftover take-home rates30 and reducing plate waste in buffet-style hotels31. A major benefit of this intervention is that it is highly cost-effective and highly scalable - the main costs associated with implementing this intervention are the costs of printing some basic signage.

We believe it is likely that providing employees with actual stats about employees’ attitudes toward food waste was a key reason why the intervention was successful. Future interventions could build on this idea by not only providing stats about how employees feel about food waste, but also stats on how they actually behave with respect to food waste. Messages like “A large majority of employees waste less than X amount of food at workplace cafes - join them to fight food waste!” could be an effective way to evolve this intervention.

Distribution of plate waste at a large technology company

In addition to playing a key role in allowing us to dramatically increase statistical power for testing our interventions, using Orbisk’s Orbi technology also allowed us to understand plate waste behavior on a user-by-user basis. During the entire study duration at each site, each time a user returned their dishware at the end of their meal, a back-of-house team member scanned their waste (or lack thereof) using the Orbi. One key insight this enabled us to glean was the pattern of waste at the company. We found a very consistent distribution of food waste across all cafes studied as part of this research: the vast majority of employees wasted very little food, but a small minority end up with high-waste plates that end up accounting for a large proportion of the overall food waste. This insight provides a key leverage point for developing future interventions to reduce food waste – if methods could be found that successfully reduce the occurrence of high-waste plates, it could have an outsized impact on overall food waste reduction.

A crucial question emerges from this insight: why do these high-waste plates happen? It is important to point out that there is no evidence to suggest that the high-waste plates we observed were consistently due to the same individuals (i.e., it is not a given that the same people end up with high amounts of waste each day). Indeed, employees reported very high levels of agreement that waste should be prevented and that they do what they can to prevent waste, so it is unlikely that there are a large number of individuals who consistently return plates that fall into the top 10% of highest-waste plates. Rather, it could be that there are idiosyncratic reasons why different people may end up with high-waste plates on different occasions. People may not like the taste of a particular food that they thought they would like and therefore end up wasting as a result. People may also grab their lunch and then realize they were running late for a meeting. Future research that better-understands the why behind these high-waste plates could unlock the ability to make significant additional inroads into waste reduction at large technology companies and beyond.

Limitations

While these studies provided several important advances into our understanding of food waste and methods that can be used to reduce it, it is important to note their limitations. One limitation of this work was that while we were able to gain user-by-user information on how much food was wasted, we were not able to gain user-by-user information on how much food was taken in the first place. This kind of information could be informative to help understand whether the intervention worked as intended – by having people take less food to begin with.

Another limitation concerned the difference-in-differences designs. In Study 3, logistical issues resulted in a design that was not perfectly balanced: in one phase, the change in waste at one cafe was compared to that of two others, and in the next phase, to three. The original plan called for a balanced 2:2 comparison, which would have provided greater symmetry. Additionally, the cafes included in the design were all in different locations. While this is a strength in terms of external validity (we got evidence that the intervention could be successful across a wide array of geographies and cultures), it also meant that factors unrelated to the introduction of the intervention could cause changes in the amount of waste in a way that differed between the cafes. To counteract this, we monitored each cafe closely throughout the duration of the study and made adjustments as-needed if factors changed that could have an impact on plate waste – for instance, when the Mountain View cafe changed their dishware midway through their baseline period, we altered the design of the study to account for this by disregarding the data from the period in which this change took place. Additionally, we planned our design so that there would be several opportunities for replication. Across all stages of the difference-in-difference designs across both RCTs, we found the same thing: introducing the intervention signage was associated with less waste than would be expected compared to other cafes that either did not have signage active, removed their signage, or had old signage up. This level of consistency gives us substantial confidence in the finding that the intervention reduces waste, but future work could replicate this work further by testing the impact of the intervention within the same location.

Another limitation of this work is that we are not able to distinguish between the effects that the social norm signage had from the effects that the portion prompting signage had on food waste. We believe it is the case that each of these elements of the intervention likely played a role in reducing waste, but it is unclear if one element was more impactful than the other, whether they depend on each other to work, or whether their effects are additive. Future work would be needed to disentangle the effects of the different elements of the intervention we deployed. Future work is also needed to directly measure the mechanism by which the intervention was successful – we believe it is likely that the intervention worked largely via changing employees’ perceptions of the prevailing social norms around food waste, but future studies that directly measure employee attitudes and understanding of norms as part of the RCT design would be needed to confirm this.

Additionally, the RCTs here were designed to test whether the signage was effective, but more research would need to be done to get at some other practicalities, including the optimal time that signage should be left up for, testing the efficacy of very specific messages, and similar details.

Finally, all of this work took place at one large technology company (Google), which, like all companies, has its own unique culture. We believe the general approach taken here – surveying the employees within the company and using those responses to create social norms-based messaging to reduce food waste – is likely to be successful across many organizational contexts, as there is significant research that social norms are powerful behavior-shaping forces across many cultures32. However, additional research would need to be done to empirically assess the extent to which the findings here would generalize beyond this company.

Conclusion

In conclusion, this global study on food waste at a large technology company showed that social norms and portion prompting signage can be an effective way to prevent food waste in corporate dining environments. It also provided several unique insights into existing social norms around food waste at large technology companies and into patterns of actual waste behavior. It is our hope that the intervention we designed can be deployed widely to reduce waste at large technology companies and beyond and that the insights we gained can pave the way for more research to continue to effectively address the significant societal problem that post-consumer food waste represents.