An incomplete starter's guide to attrition metrics

Something for nothing

I started writing this post months ago and for some reason haven’t found the will to finish it. Rather than let it keep gathering virtual dust, I’ve decided to share what I do have in the hopes it will be helpful for somebody, even in its incomplete state.

Ben

---

Employee attrition and retention is an evergreen topic in people analytics and in business management more broadly. Because employee attrition is costly — due to hiring and onboarding costs, and lost productivity (among other costs) — business and HR leaders frequently have questions about which kinds of employees are leaving, whether they are critical losses, why people are leaving, how to retain them, and so forth.

Having spent over four years focused on attrition reporting and research at my current employer, I thought it would be helpful to collect some of what I’ve learned about the nuances of attrition data, research, and reporting, so others don’t have to learn these lessons the same way I did. That said, I’m not sharing anything confidential and this represents my own opinions only.

Note: I use turnover, retention, exit, terminations, attrition, etc. interchangeably. If I want different terms to mean different things, I add those differences to the beginning of the word as an adjective, such as “attrition that won’t be backfilled.” Of course, if your organization is used to assigning different definitions to these words, you’ll want to be consistent.

Before I jump in, I’d like to note that I will not spend much time on attrition forecasting or workforce planning specifically. Attrition forecasts play a key role in workforce planning, particularly when they enable recruiting teams to plan for backfill volume ahead of time. I’ve spent most of my time on the research and reporting side, so I’ll stick to what I feel comfortable talking about. 

Why does it matter?

(missing for now - lots has been said on this but ideally would be included here too)

• Business continuity

• Cost of recruiting

• Onboarding/ramping/productivity

Data sources

Where does attrition data come from?

(missing for now) - meant to include a better summary here

• HRIS

• Exit survey/interview

• Engagement survey

• External vendors

Metrics

There are a lot of useful ways to measure attrition, and I’m certainly not going to cover them all. Instead, I’ll cover the ones I’ve found useful, define them, and attempt to explain when and why each might be useful. More details on each metric follow below this summary table:

Trailing X-month attrition rate

The standard way I look at attrition trends is a Trailing 12 Month (T12) attrition rate. T12 attrition is defined as:

# of exits (over the past 12 months) average headcount (over the past 12 months)

The # of exits metric is straightforward once you’ve determined what your group of interest is, and have ensured you have a consistent way to determine somebody’s “exit date” (which could be the official termination date, or their last day worked, or something else appropriate, but be consistent!).

Average headcount is trickier, and there are three primary ways to calculate average headcount.

1. The first way is to cheat and use SOP or EOP headcount, or headcount at the start (or end) of the period in which you're counting attrition (e.g., 12 months ago). This isn’t actually “average” headcount, but it remains a commonly used denominator in headcount calculations.

   1. The primary issue with this approach is that your group size may change quite a bit over the period, so using SOP or EOP headcount as the denominator can produce misleading attrition rates.

2. The second way is to grab SOP and EOP headcount, add them together, and divide by two, giving you the average of the two points. This is straightforward and addresses the main issue with SOP or EOP headcount, such as if your group of interest is growing or shrinking a lot over the period.

3. Option 3 is the same idea as option 2, but even more careful: we’re going to take SOP and EOP headcount, but also the headcount at the end of each intervening month. For a 12 month period, that gives us 13 headcount numbers (12 month ends, plus SOP which is usually the end of the month prior), so we divide the sum by 13 to get our average.

I recommend either option 2 or option 3, with option 3 being better but potentially more difficult to implement. You can decide when and whether the tradeoff is worth it — though I wouldn’t recommend using different denominators across reporting surfaces, since you could end up with different numbers.

T12 attrition rates have several advantages for attrition reporting:

• Not affected by seasonality, such as higher attrition after bonus payouts or lower attrition during winter months

• Less likely to startle executives, since month over month (m/m) changes are relatively tame

There are also variations on this, such as Trailing 6 Month (T6) or Trailing 3 Month (T3) attrition rates; these give some of the benefits of T12 (such as reduced seasonality, especially if your main seasonal effects are quarterly or semiannual) while being a bit more responsive to changes in employee behavior.

Annualized/YTD/Monthly attrition rate

• (missing for now)

Cohort-style rates

Sometimes we’re specifically interested in the attrition rate for a specific group of people who became a cohesive group (or cohort) at the same time. This could be a class of new hires, a group who received a special bonus, or graduates of a training program. In these cases we may not want to mess around with average headcount — we just want to know what percentage of the group has left after a certain amount of time. The general formula is:

number of exits in the group during the period number of people in the group at the start of the period (SOP headcount)

Cohort-style rates can be useful when comparing a cohort against a comparison group (after 6 months, 2% of the training cohort had left the company, compared to 7% of similar employees who hadn’t gone through the training!) or against past versions of the cohort (typically we lose 10% of our high achiever bonus awardees in the year after the bonus is granted, but over the past two years when we implemented a networking group for these recipients, that dropped to 4%!).

Attrition counts

This one is pretty straightforward — it’s the numerator to most of the other metrics we’ve looked at. But counts can be a useful metric on their own, too:

• Very small groups (e.g., VPs) may have attrition rates that fluctuate wildly, so sticking with counts can actually be clearer (we lost 4 of our VPs this year!).

• How many positions in this team may need backfilling?

Exit Reasons

• (missing for now)

More detail on metrics

Attrition as a term can mean different things to different people. Some people define attrition and turnover differently based on whether a backfill is planned, for example. My preference is to use other adjectives to make clear what I’m talking about, so I don’t need to expect people to differentiate between attrition, turnover, exits, terminations, etc. In the previous example I’d rather use names such as “attrition in positions with planned backfills” or some shorter term that would resonate in my organization.

Here are some other common modifiers for attrition metrics:

• Voluntary/Involuntary: whether the employee’s exit was voluntary (e.g., they took a new job and gave notice) or involuntary (they were dismissed for some reason, including a layoff) 

• Performance-related: the employee’s exit was related to poor performance. This doesn’t usually include people who left who happened to be low performers, but rather people who were on some kind of performance management or coaching plan. These are usually involuntary, though you could choose to include people who left voluntarily while they were on the plan.

• Layoff or RIF (Reduction in Force): in the United States, employees can be terminated at any time. When enough are let go all at once, this is known as a layoff or RIF. This is a unique kind of attrition that is worth labeling.

• Divestiture: when a company sells off (divests) a subsidiary, those employee are no longer working for the original company, which is a form of attrition, but they are still employed by the subsidiary in most cases. This is also worth labeling.

• Employee types: there are likely too many to list here but as some examples:

  • Full-time/Part-time

  • Salaried/Hourly

  • Interns

  • Contractors

• Location: the country or office location (or Remote!) where employees work.

• “Key talent”: most companies have some variation of “key talent” which could represent a key job type (e.g., Traders at a brokerage) or performance band (“High Performers”). Knowing about attrition for this specific group is likely to be most impactful.

Analysis approaches

This is not intended to be an in-depth guide for using any of these approaches, nor for how to implement them. My goal is to inform which methods are especially appropriate for attrition; other resources will help with the implementations in various tools.

(missing for now) - intended to share more details/examples of these various approaches

• Rate comparisons

  • Vs company average

  • Vs history

  • Vs comparison group

  • Benchmarks and my opinions

• Cohort analysis

  • What happened to two groups starting at the same time, or similar groups starting at different times?

• Survival analysis

• Tree-based methods

• Driver analysis

  • Logistic regression

  • Decision trees

Notes

Additional notes that didn’t make their way into the main body:

• If you have tenure-related patterns in your attrition data, such as new hires being especially likely or unlikely to leave, your attrition rates will be affected by how many people you’re hiring at any given period.

• Forecasting and workforce planning is completely untouched here, but is valuable and worth doing

• Had planned to share a whole thing about how “average tenure” is almost always the wrong metric and why you should use “years at which x% of the cohort has left” as the metric in most cases

• Can be useful to look at why people stay as well as why they leave; they may be different or even unrelated in some cases!