Boards are tired of arguing about office attendance, so many executives reach for a softer rationale: we need people back so younger staff can learn from veterans. It sounds humane. It is also a weak case. The strongest syntheses of mentoring research show benefits that are real but generally small, and they say nothing about where the conversations happen. If you want mentoring to help across generations, build it on outcomes and accountability, not on a zip code.
What the evidence actually says
A seminal meta‑analysis in the Journal of Applied Psychology found that mentored employees do better on several career outcomes than non‑mentored peers, but the effects on hard metrics are modest (Allen, Eby, Poteet, Lentz and Lima, 2004). The authors report small associations with pay, roughly a tenth‑point correlation, and only moderate links with promotions, about a third of a correlation point, while subjective outcomes such as career satisfaction are more strongly associated. In plain terms, mentoring points in the right direction, yet it is no magic escalator to higher salaries.
A broader meta‑analysis across 112 studies reached a similar conclusion. Mentoring is associated with favorable outcomes across behavioral, attitudinal, health, relational, motivational, and career domains, but effect sizes are generally small. Career attitudes and situational satisfaction, for example, cluster around correlations of .14 and .16 respectively (Eby, Allen, Evans, Ng and DuBois, 2008). The pattern also varies by context. Academic mentoring shows the strongest associations, while workplace mentoring sits in the middle with effects typically between .03 and .19 (Eby, Allen, Evans, Ng and DuBois, 2008). That should temper what any single program promises inside a firm.
Notice what neither analysis tests. They compare mentored and non‑mentored people, not office attendance and remote work. Using mentoring as a blanket rationale for return‑to‑office is therefore a category error. The question is not where people sit. It is whether the mentoring you run is designed to move the specific outcomes that are most malleable and most valuable in your context.
The location fallacy
If the goal is to raise compensation quickly through mentoring, the evidence gives little support for big, near‑term jumps in the typical corporate setting. If the goal is to improve career satisfaction, belonging, or clarity about advancement, the association is stronger, although still modest on average (Allen, Eby, Poteet, Lentz and Lima, 2004; Eby, Allen, Evans, Ng and DuBois, 2008). None of this depends on a building. Mentoring can be synchronous or asynchronous, one‑to‑one or group based, in person or virtual. The mechanism is guidance and access, not a desk.
Treating mentoring as a place risks two failures at once. It can justify a costly mandate without improving outcomes, and it can underserve the people who most need support. Older workers, who carry much of a firm’s tacit knowledge, also face the steepest barriers to reskilling. Only about a third of adults aged 60 to 65 participated in training in 2023, compared with more than half of those aged 25 to 44 across the OECD (OECD, 2025). At the same time, demographic pressure is relentless. The old‑age dependency ratio across the OECD averaged 19 percent in 1980, 31 percent in 2023, and is projected to reach 52 percent by 2060 (OECD, 2025). A multigenerational workforce is not a slogan, it is the long run.
The lesson is simple. Design mentoring for the frictions you actually face. If early‑career staff lack context on how decisions are made, define that as an outcome and build a format that can deliver it at scale. If mid‑career or older staff need confidence or fluency with new tools, specify that and create a two‑way exchange. Office attendance may help some pairs, but it is not the variable the evidence highlights.
Build for outcomes you can measure
Executives do not need perfect evidence to act, but they do need proportion. Three design moves align with what the meta‑analyses support and with the talent math of an aging workforce:
- Set outcome targets you can actually shift. The literature shows stronger associations with subjective outcomes than with pay. Make those the first‑order goals for your program. Examples: clarity on promotion criteria, sense of belonging, and career satisfaction (Allen, Eby, Poteet, Lentz and Lima, 2004; Eby, Allen, Evans, Ng and DuBois, 2008).
- Match on skills, not birthdays. Use intake forms to map concrete gaps and assets, then pair people accordingly. Where tool adoption is the pain, consider structured reverse mentoring. Where political navigation is the pain, allocate mentors with cross‑functional experience. This is a design choice, not a claim of proven superiority.
- Choose the lightest medium that works. For knowledge transfer and feedback, alternating short video calls with written debriefs can be enough. For network access, group sessions with senior leaders can supplement one‑to‑ones. The meta‑analyses do not privilege a channel. Your measurement should.
Then run the program like a product. Start small, publish the intended outcomes, and measure pre‑post changes in the specific variables you target, such as self‑reported career clarity or satisfaction. Track retention or promotion decisions only if roles and performance management are stable enough to make comparisons meaningful. Given that workplace effects are typically in the .03 to .19 range across outcomes (Eby, Allen, Evans, Ng and DuBois, 2008), calibrate expectations and sample sizes accordingly. Sunsetting weak formats is a strength, not a failure.
The practical read for leaders
Two truths can coexist. Mentoring matters, and it is not a silver bullet. The meta‑analyses support offering mentoring, while urging realism about magnitude and timelines, especially for hard outcomes like pay (Allen, Eby, Poteet, Lentz and Lima, 2004; Eby, Allen, Evans, Ng and DuBois, 2008). Demographics make cross‑generational learning a strategic need, not a nice‑to‑have, as older workers’ share rises and their access to training lags (OECD, 2025).
So stop using mentoring as a proxy argument for where people work. Use it as a design test for how your organization learns. Define the outcomes you will hold yourself to. Pair people on skills. Mix formats. Measure what moves. If the numbers do not shift, change the design. If they do, you will not need an office mandate to claim success. You will have the data.
Sources
- Allen, Eby, Poteet, Lentz and Lima (2004). Career benefits associated with mentoring for protégés: a meta-analysis. Finds positive but generally small associations with objective career outcomes, with stronger links to subjective outcomes, grounding realistic expectations for mentoring. https://www.semanticscholar.org/paper/Career-benefits-associated-with-mentoring-for-a-Allen-Eby/a13c80a7f263f4378c07b474a0628105b91c6d28
- Eby, Allen, Evans, Ng and DuBois (2008). Does Mentoring Matter? A Multidisciplinary Meta-Analysis. Synthesizes 112 studies, showing mentoring has broadly favorable but small effects, with workplace effect sizes typically between .03 and .19 and stronger effects in academic contexts. https://pmc.ncbi.nlm.nih.gov/articles/PMC2352144/
- OECD (2025). OECD Employment Outlook 2025. Documents the rise in the old-age dependency ratio and the lower training participation among 60–65 year-olds, making cross-generational learning a structural priority. https://www.oecd.org/en/publications/oecd-employment-outlook-2025_194a947b-en.html