Buoy

What the research actually says about asking a colleague

· 11 min read · the Buoy humans

Buoy rests on a hunch: when you're stuck on an AI tool, the fastest way out is ten minutes with someone who figured it out last month, looking at the same window you are. Hunches are cheap, so we spent a day reading what researchers have measured about one-to-one help, informal learning at work, mentoring, and what happens when you make helping visible. Here is what held up, what didn't, and what we're changing because of it.

The two-sigma legend is smaller than you were told

Everyone in education has heard that one-to-one tutoring moves students two standard deviations, Benjamin Bloom's famous 1984 claim. It came from two small dissertations that mixed tutoring with mastery learning and extra feedback. The best modern synthesis of randomized trials, 96 studies of tutoring in schools, lands at about 0.37 standard deviations,1 and the effects grow with trained tutors and frequent sessions. Still one of the most effective interventions anyone has measured. Just not magic.

Peer tutoring sits in the same range, and it pays twice. Across 65 programs, the students being tutored gained about 0.40 standard deviations and the students doing the tutoring gained about 0.33.2 A 2019 meta-analysis of learning by teaching found that actually explaining something to another person, after preparing to, produced a gain of 0.56 compared with studying for a test, and the effect was larger when the teaching was interactive.3 The person who helps you learns nearly as much as you do. That is not a nice side effect. It is half the point.

One level ahead beats the expert, for getting unstuck

This is the finding we built the product on, and it comes from adults, not schoolchildren. Pamela Hinds and colleagues had experts, intermediates and beginners write instructions for a wiring task. Experts used more abstract language and fewer concrete steps. Novices taught by beginners did better on the task than novices taught by experts, and reported fewer problems with the instructions.4 Earlier work by Hinds showed experts consistently underestimate how long novices will take and how many errors they'll make, stay overconfident even when paid for accuracy, and can't be debiased.4

There is a catch, and we'll keep it in view: novices taught by experts did better on a different task in the same domain. Near-peers are best for getting unstuck right now and for language you can actually follow. When the question is conceptual, an expert's abstraction transfers further. Buoy should default to one level ahead and keep "an expert instead" one tap away.

What happens inside the session matters more than who is in it

Michelene Chi's team ran a study where tutors were forbidden to explain or give feedback. They could only prompt. Students learned just as much.5 The learning came from what the learner constructed, not what the tutor said. The workplace version of this is the helper grabbing the mouse and doing it for you. In intelligent-tutoring systems the equivalent, clicking straight through to the final hint without trying, turned out to be so common that the systems had to be redesigned to delay it.5 Your helper should show you once, then let you drive.

Seeing the screen is the whole difference

Robert Kraut, Susan Fussell and Darren Gergle spent years measuring what a shared view of the work does to remote help. Pairs with a live shared view were faster and more accurate than pairs talking over audio, and the advantage was largest "when tasks are visually complex or when participants have no simple vocabulary for describing their environments."6 Delay the shared view and the benefit erodes. A software interface full of dialogs and settings is exactly that kind of task, and a text description of your problem is exactly the condition that loses. Timing matters the same way: students learned more from hints they asked for than from hints pushed at them, and the top two reasons university students skip scheduled office hours are "no questions right now" (41%) and the schedule itself (41%). Help that arrives when you are stuck sidesteps both.

The barrier is the ask, and the asker is wrong about the cost

People don't ask. Fiona Lee's work on organizations shows help-seekers fear looking incompetent, dependent and inferior, most of all on tasks central to their job. The people who need help most ask least. And they are miscalibrated: across more than 14,000 requests in Vanessa Bohns' studies, askers underestimated how often people would say yes by roughly half.7 Alison Wood Brooks, Francesca Gino and Maurice Schweitzer found that people who ask for advice on hard tasks are rated more competent, not less.8

Visibility is the live wire. In an experiment with 2,521 white-collar professionals, advice-seeking fell sharply when a manager could see the decision to seek it, even though managers in the same study did not judge askers negatively.9 For AI tools specifically, Microsoft's survey of 31,000 workers found 52% of AI users are reluctant to admit using it for important tasks,10 and a separate experiment found engineers believed to have used AI received lower competence ratings for identical code.11 Interviews with knowledge workers describe the result: people hide their AI use, juniors assume seniors don't use it, and one participant solved a colleague's spreadsheet problem with AI without saying so ("She thought I was a genius").12 The practice everybody needs to see is the one everybody is hiding.

Most learning at work is informal, and remote work cut into it

You have probably heard that 70% of workplace learning is informal. That number traces to a 1996 self-report survey of about 200 executives and, in the one peer-reviewed assessment, has an evidential basis described as weak.13 The direction survives: Michael Eraut's studies of professionals found most workplace learning is informal, mixes learning from other people with learning from experience, and is usually invisible even to the learner.14

The channel is proximity. Microsoft's natural experiment on 61,000 employees found firm-wide remote work made collaboration networks more static and siloed and cut time collaborating across groups by about a quarter.15 Software engineers who sat in the same building as their teammates got roughly a fifth more feedback on their code, with the gains concentrated among juniors, and senior engineers wrote less code when near juniors because they were busy mentoring.16 What rebuilds the map digitally is not a skills database. Paul Leonardi's six-month field experiment found that simply watching colleagues' exchanges on an internal network improved people's accuracy about who knows what by 31% and who knows whom by 88%.17

Mentoring programs pay modestly, and the RCT has a twist

Four meta-analyses agree: formal mentoring helps, but the effects are small and mostly attitudinal, with correlations around 0.05 to 0.20 and the authors cautioning readers not to overestimate them.18 The one well-identified workplace randomized trial is more interesting. Six hundred new call-center hires were assigned four weeks of weekly meetings with an established peer, not a supervisor. Mentored agents produced 18% to 19% more revenue in their first two months, and most of the gain persisted. The twist: among workers who opted in, the gain was negligible. The people who volunteer for help are already the strong ones. The benefit was concentrated among people who would never have signed up.19 Frequency matters more than duration for this kind of task help; in the meta-analytic data, how often mentor and protégé interact correlates with relationship quality at 0.44, while how long the relationship has lasted correlates at 0.09.18 Many short exchanges with a known person beat a long formal relationship, as long as the reluctant are pulled in.

Thank people, don't rank them

Recognition has sharp edges. In a controlled experiment where helping cost the helper money, a leaderboard ranking people by recognition they had given increased helping, while leaderboards ranking recognition received reduced it, because coworkers began to doubt the motives behind the help.20 A peer-recognition system raised help-seeking when most of the team used it and lowered help-seeking among non-members when only a few did.21 Across 128 experiments, tangible rewards tied to performance undermined intrinsic motivation while verbal praise raised it.22 Points, levels and leaderboards reliably raise the quantity of contributions and not their quality, and in a semester-long course the section with a leaderboard ended up less motivated and with lower exam scores.23

And the archive problem is real. A study of thousands of corporate wikis found thriving ones were a minority, 66% of users at one company said theirs hadn't been touched in six months, and even the best team wiki went "to weed" after its gardener left.24 On Stack Overflow, most obsolete answers were already obsolete when posted and only a fifth are ever fixed. Knowledge captured as a chore dies. Knowledge captured as a by-product of a live conversation, with an owner and a shelf life, has a chance.

What this changes about Buoy

Most of it confirms the core: a colleague one level ahead, on voice, looking at the window you chose, within about a minute, is the configuration the evidence likes best. The rest of what we read argues for taking things away rather than adding them.

We'll keep updating this as we learn, and as the studies do. If you think we've misread one, the sources are right here.

Sources

  1. Nickow, A., Oreopoulos, P., & Quan, V. (2020). The Impressive Effects of Tutoring on PreK-12 Learning: A Systematic Review and Meta-Analysis of the Experimental Evidence. NBER Working Paper 27476. nber.org/papers/w27476
  2. Cohen, P. A., Kulik, J. A., & Kulik, C.-L. C. (1982). Educational outcomes of tutoring: A meta-analysis of findings. American Educational Research Journal, 19(2), 237–248. doi:10.3102/00028312019002237
  3. Kobayashi, K. (2019). Learning by preparing-to-teach and teaching: A meta-analysis. Frontiers in Psychology, 9, 2755. doi:10.3389/fpsyg.2018.02755
  4. Hinds, P. J., Patterson, M., & Pfeffer, J. (2001). Bothered by abstraction: The effect of expertise on knowledge transfer and subsequent novice performance. Journal of Applied Psychology, 86(6), 1232–1243. doi:10.1037/0021-9010.86.6.1232 · Hinds, P. J. (1999). The curse of expertise. Journal of Experimental Psychology: Applied, 5(2), 205–221. doi:10.1037/1076-898X.5.2.205
  5. Chi, M. T. H., Siler, S. A., Jeong, H., Yamauchi, T., & Hausmann, R. G. (2001). Learning from human tutoring. Cognitive Science, 25(4), 471–533. doi:10.1207/s15516709cog2504_1 · Aleven, V., & Koedinger, K. R. (2000). Limitations of student control: Do students know when they need help? ITS 2000. doi:10.1007/3-540-45108-0_33
  6. Gergle, D., Kraut, R. E., & Fussell, S. R. (2004). Language efficiency and visual technology: Minimizing collaborative effort with visual information. Journal of Language and Social Psychology, 23(4), 491–517. collablab.northwestern.edu (PDF) · Razzaq, L., & Heffernan, N. T. (2010). Hints: Is it better to give or wait to be asked? ITS 2010. wpi.edu (PDF) · Office-hours survey: PMC9727601
  7. Bohns, V. K. (2016). (Mis)Understanding our influence over others: A review of the underestimation-of-compliance effect. Current Directions in Psychological Science, 25(2), 119–123. doi:10.1177/0963721415628011 · Lee, F. (2002). The social costs of seeking help. Journal of Applied Behavioral Science, 38(1), 17–35. doi:10.1177/0021886302381002
  8. Brooks, A. W., Gino, F., & Schweitzer, M. E. (2015). Smart people ask for (my) advice: Seeking advice boosts perceptions of competence. Management Science, 61(6), 1421–1435. doi:10.1287/mnsc.2014.2054
  9. Heursen, L., Friess, S., & Chugunova, M. (2023). Reputational Concerns and Advice-Seeking at Work. Max Planck Institute for Innovation and Competition Research Paper. ip.mpg.de
  10. Microsoft & LinkedIn (2024). 2024 Work Trend Index Annual Report: AI at Work Is Here. Now Comes the Hard Part. microsoft.com/worklab
  11. Microsoft Research (2025). New Future of Work Report 2025. microsoft.com/research (PDF)
  12. Xia, H., et al. (2026). Hidden in plain sight: How knowledge workers conceal and share generative AI use. CHIWORK '26. arXiv:2602.01386. arxiv.org/abs/2602.01386
  13. Clardy, A. (2018). 70-20-10 and the dominance of informal learning: A fact in search of evidence. Human Resource Development Review, 17(2), 153–178. doi:10.1177/1534484318759399
  14. Eraut, M. (2004). Informal learning in the workplace. Studies in Continuing Education, 26(2), 247–273. doi:10.1080/158037042000225245
  15. Yang, L., Holtz, D., Jaffe, S., et al. (2022). The effects of remote work on collaboration among information workers. Nature Human Behaviour, 6, 43–54. doi:10.1038/s41562-021-01196-4
  16. Emanuel, N., Harrington, E., & Pallais, A. (2023). The Power of Proximity to Coworkers: Training for Tomorrow or Productivity Today? NBER Working Paper 31880. nber.org/papers/w31880
  17. Leonardi, P. M. (2015). Ambient awareness and knowledge acquisition: Using social media to learn "who knows what" and "who knows whom." MIS Quarterly, 39(4), 747–762. aisel.aisnet.org
  18. Eby, L. T., Allen, T. D., Evans, S. C., Ng, T., & DuBois, D. L. (2008). Does mentoring matter? A multidisciplinary meta-analysis comparing mentored and non-mentored individuals. Journal of Vocational Behavior, 72(2), 254–267. PMC2352144 · Eby, L. T., et al. (2013). An interdisciplinary meta-analysis of the potential antecedents, correlates, and consequences of protégé perceptions of mentoring. Psychological Bulletin, 139(2), 441–476. doi:10.1037/a0029279
  19. Sandvik, J., Saouma, R., Seegert, N., & Stanton, C. (2022). Should Workplace Programs Be Voluntary or Mandatory? Evidence from a Field Experiment on Mentorship. sioe.org (PDF)
  20. Evans, J. H., Presslee, A., & Vandenberg, J. (2026). Peer-to-peer recognition leaderboards: Givers vs. receivers. The Accounting Review. giesbusiness.illinois.edu
  21. Burke, J., Sommerfeldt, R., & Wang, Z. (2025). Peer recognition systems and help-seeking. Management Science. scholarsarchive.byu.edu
  22. Deci, E. L., Koestner, R., & Ryan, R. M. (1999). A meta-analytic review of experiments examining the effects of extrinsic rewards on intrinsic motivation. Psychological Bulletin, 125(6), 627–668. doi:10.1037/0033-2909.125.6.627
  23. Mekler, E. D., Brühlmann, F., Tuch, A. N., & Opwis, K. (2017). Towards understanding the effects of individual gamification elements on intrinsic motivation and performance. Computers in Human Behavior, 71, 525–534. doi:10.1016/j.chb.2015.08.048 · Hanus, M. D., & Fox, J. (2015). Assessing the effects of gamification in the classroom. Computers & Education, 80, 152–161. doi:10.1016/j.compedu.2014.08.019
  24. Grudin, J., & Poole, E. S. (2010). Wikis at work: Success factors and challenges for sustainability of enterprise wikis. WikiSym 2010. doi:10.1145/1832772.1832780 · Zhang, H., Wang, S., Chen, T.-H., & Hassan, A. E. (2019). An empirical study of obsolete answers on Stack Overflow. IEEE TSE. arxiv.org/abs/1903.12282