Inside the Everis Engine: biostatistics in plain language

The Everis Engine is our proprietary, in-house analytics layer. It's built and run by our own team. Here is what it does, in plain language.

Why biostatistics

Biostatistics is the branch of statistics built for biological and health data. That kind of data is noisy, repeated and individual: the same person measured every day, with natural ups and downs. Standard business analytics isn't designed for it. Biostatistics is.

Step 1: Clean the data

Wearable data isn't perfect. Devices come off, batteries die, sensors slip. Before anything else, the Engine applies data quality controls: minimum wear-time thresholds, filtering of obvious artefacts, and transparent handling of missing data. Bad data in means bad decisions out, so this step matters.

Step 2: Build a personal baseline

Every signal is expressed relative to the person's own rolling baseline, for example a standardised score against their last 28 days. People are compared with themselves, not with a population.

Step 3: Separate real change from noise

Every measurement has a typical error. The Engine uses that, plus a smallest meaningful change threshold, to tell genuine shifts from day-to-day variation. This is the same approach used in athlete monitoring.

Step 4: Follow the trend

Rolling 7, 14 and 28-day averages, and exponentially weighted averages that give more weight to recent days, smooth out single bad nights and highlight sustained drift.

Step 5: Combine signals into indices

Physiological signals and validated survey scores are combined into composite indices, such as the Stress Risk Index and Sleep Quality Index. One signal alone can mislead. Several moving together are more reliable.

Step 6: Separate people from teams

Mixed-effects, or multilevel, models handle the structure of the data: repeated measures within each employee, and employees within teams. That lets us separate individual change from team-level patterns, which is what leaders need to see.

Step 7: Stay stable with small numbers

Small teams and new clients produce less data. Bayesian hierarchical estimation borrows strength across groups to produce stable estimates, and always reports uncertainty alongside the number.

Step 8: Forecast

As data matures, the Engine projects absence days and cost to company for the coming period, with confidence bands. Time-to-event analysis estimates how risk of an absence episode changes after a sustained signal shift. Risk scores are calibrated against what actually happens, and evaluated on discrimination, calibration and lead time.

Step 9: Explain every flag

Every flag shows which signals drove it. Physiologists and HR see the reason, not just a score.

What this means for your organisation

The Engine turns noisy individual data into clear, de-identified team insight, with honest uncertainty. Leaders get forecasts they can plan around. Employees get support triggered by real change, not noise. And because it's built in-house, we control how it works and can explain every output.