How you compare,
when you can trust the comparison
Benchmarks are easy to publish and hard to earn. Your first reading is not your real reading, and we would rather tell you that up front than hand you a number you cannot defend in a board meeting. Here is what SPARKS shows you at ninety days, at six months, at a year, and every day after that.
Every one of them arrives after the decision has been made
Most organizations already hold four or five readings on how their workplace is doing. They are not wrong, and they are not useless. They are late, and most of them come from a population that selected itself.
| Signal | Who produces it | When it arrives | What it actually tells you |
|---|---|---|---|
| Public review sites Glassdoor, Indeed, Comparably |
A small self-selected minority. Posting is driven by strong feeling, so the sample skews toward people who have already left unhappy or have decided to. | After the decision, and permanently | Reputation, not condition. Useful for recruiting optics, close to useless as a measure of how the current workforce is doing. |
| Annual engagement survey | Employees who choose to respond, once | Once a year | Sentiment during the week it was fielded, with no way to separate a bad quarter from a bad culture. |
| Year-in-review and performance cycle | Managers, writing about individuals | Annually | Performance and output. It measures what someone delivered, not what it cost them to deliver it. |
| Exit interviews | People who have already gone | After the loss | A post-mortem. Gallup finds that 52% of people who leave voluntarily say it was preventable, which is a finding you can only act on before the interview. |
| Best-workplace lists | Employers who apply, plus a surveyed sample | Annually, on application | A marketing asset. Valuable as one, not diagnostic. |
| SPARKS Strain Index | A validated instrument, taken repeatedly, reported only in cohorts of thirty or more | Monthly from month three, continuously from month twelve | Current load, by cohort, against your industry and against your own history. |
None of these are replaced. The Strain Index sits earlier in the same sequence.
Your engagement survey still tells you what people think. Your exit interviews still tell you why someone left. The Strain Index tells you which teams are heading that way while there is still something to do about it.
What you get, and when
We release each level of comparison only when it is defensible. Adoption has to climb, the early sample has to settle, and a year of your business has to run through the data before a benchmark means what it appears to mean.
-
Baseline, not benchmark
First 90 days
Adoption is still climbing, and the early sample is reactionary by nature: the people who sign up first are usually the people with something going on right now. A cohort comparison drawn from that sample would be a comparison of who is struggling loudest, not of how teams differ.
You see: adoption and participation by cohort, a first organisation-level baseline, and the industry benchmark for context. We hold back: cohort-level comparison, because it would not survive scrutiny and you would have to defend it anyway.
-
Your first defensible cohort read
3 to 6 months
Participation has stabilised and there is a second measurement point, so the early reactionary spike washes out of the average. Cohorts above the thirty-person floor now carry enough signal to stand comparison, both against each other and against the benchmark.
You see: cohort Strain Index against the industry benchmark, first quarter-over-quarter movement, and which cohorts sit above the action threshold.
-
Cycle-adjusted
6 to 12 months
A full year of your business has now run through the data: quarter ends, peak season, performance review cycles, budget season, the holidays. Until you have seen those once, you cannot tell a strained team from October. From here the comparison is adjusted for your cycle, not only for your industry.
You see: seasonally-aware baselines, before-and-after analysis on any intervention you ran, and correlation between strain and the outcomes you already track.
-
Continuous
12 months onward
Real time from here. Every new reading lands against two references at once: your own established baseline and the current industry benchmark, refreshed quarterly. This is the point at which the index stops being a report you read and becomes a signal that reaches you.
You see: live cohort movement, notification when a cohort crosses a threshold, and a benchmark that keeps moving with the industry rather than freezing at the date you signed.
Your first reading is not your real reading. We would rather show you less and have it hold.
Meaningful comparison means comparing like with like
The index is always scored out of 100. What changes is who you are scored beside. A forty-person firm measured against a forty-thousand-person enterprise learns nothing, so peer groups are built inside size tranches first, then narrowed by whichever lens answers your question.
First, size
Every organization sits in a tranche, and the peer set is drawn from that tranche before anything else is applied. Strain behaves differently at forty people than at four thousand, and comparing across that gap produces confident nonsense.
Then, the lens
Inside your tranche you choose what to compare against. Each lens answers a different question, and the useful findings usually come from two of them disagreeing.
Your organization against your peer set
Illustrative figures. Read across, not down. This organization is unremarkable overall at the 55th percentile, and its Sales team, the highest raw score after Finance, is entirely normal for sales work. Finance is the finding: fourteen points above both references and at the 88th percentile of finance teams in its peer set. A single-reference report would have flagged Sales and missed it.
The reading that matters is usually where two references disagree.
High and normal for the work is context. High and abnormal for the work is a finding. Low against your industry but high against your region says something about where you are, not who you hired.
When a peer set is too thin
A peer set renders only above a floor of contributing organizations, and no single organization may make up more than a stated share of it, so no median can be worked backwards into one company’s number. Where a set does not qualify, we widen it one step at a time and label the result, so you always know which comparison produced the figure you are reading.
- Size tranche and industry and region, the narrowest and most useful set
- Size tranche and industry, when the region is thin
- Size tranche only, when the industry is thin
- Global, always available, always labelled as such
You never see which organizations are in your peer set. That is what allows the comparison to exist at all: every participant contributes privately and receives an aggregate, in the same shape as a compensation survey.
Three things have to settle before a benchmark means anything
The benchmark itself already exists. What has to mature is your data inside it.
The denominator has to stop moving
A cohort at fifteen percent participation and the same cohort at sixty percent are not the same measurement. Early figures move because more people are answering, not because anything changed. We report participation alongside every number so you can see which is which.
Early joiners are not a random sample
People reach for a wellbeing tool when something is happening. That makes the first weeks of any deployment run hot, and it is the single most common way a workplace measure gets misread. Two measurement points are enough to see the effect and correct for it.
Your calendar is doing some of the work
Close periods, launch cycles, review season and school holidays all move strain, and they move it in different directions for different functions. One year of observation separates the pattern from the problem.
A private instrument, not a public score
Two boundaries are worth stating plainly, because the comparison above invites both questions.
- It is not published, and it is not a badge. Your Strain Index belongs to you. SPARKS does not publish it, does not rank organizations against each other by name, and does not licence it as a workplace certification. A number used for recruitment marketing is a number people manage, and a managed number is worthless to the person trying to run the place.
- It is not a score for any individual. No cohort below thirty people renders in any view, no individual index exists at any layer, and nothing here may be used in a decision about a person. Confidentiality is what makes participation high enough for the comparison to mean anything in the first place.