Every wellbeing platform now says it uses AI. Very few say what they will not let it do, which is the more useful half of the sentence.
Here is ours.
What we think AI is genuinely good at here
The hard problem in personal wellbeing is not collecting data. Phones and watches solved that. The hard problem is that the data arrives in separate columns and a person has to be their own analyst.
You slept badly. Your calendar is full on Thursday. You have trained hard three days running and your last reflection said you felt flat. Each of those is legible on its own. What to actually do on Thursday morning is a synthesis question, and synthesis across noisy, partial, personal signals is something models are genuinely good at.
So in Whole, AI reads signals against each other and adapts what the plan asks of you next. Lighter movement after a bad night. Recovery moved earlier when the week is heavy. The harder session shifted to a day with room in it. How that works across Body, Mind and Soul is the longer version.
That is the job. It is narrower than "AI-powered wellbeing" implies and we would rather describe it accurately.
What AI should not decide
These are commitments about the product, not observations about the field:
The distinction we try to hold
There is a difference between a principle we hold and a feature we ship, and blurring the two is the most common dishonesty in this category.
The list above describes what is implemented today: the consent setting exists, the aggregation floor exists, the absence of a productivity model is an absence you can verify by looking for it.
Where we have a view but not yet a mechanism, we try to say so in those words. "We believe" and "Whole does" are different sentences and we are careful about which one we are writing.
Why the boundary is the product
It would be straightforward to build the version of this that employers occasionally ask for. Individual risk scores, a manager view, a ranked list of who needs attention. The data model would support it. The models would produce something plausible-looking.
It would also destroy the thing that makes the data worth having, because people do not tell the truth to a system they believe is reporting on them. You would end up with a confident interface describing a workforce that had learned to perform wellness.
So the constraint is not a limitation we are apologising for. In a product whose input depends on trust, the boundary is the feature. Everything else is a rendering choice.
Burnout, and a word we are careful with
We do not claim Whole detects burnout in an individual, and we would treat any vendor who claims that with suspicion.
The World Health Organization classifies burn-out in ICD-11 as an occupational phenomenon rather than a medical condition, defining it as a syndrome resulting from chronic workplace stress that has not been successfully managed, characterised by exhaustion, mental distance from the job, and reduced professional efficacy. It sits in the chapter covering factors influencing health status, not among the illnesses.
That definition is about a person's experience of their work over time. It is not a threshold that a model crosses on someone's behalf, and treating it as one is how a wellbeing tool turns into an accusation.
Source: World Health Organization, Burn-out an occupational phenomenon: International Classification of Diseases, 28 May 2019.
If you are working out where a tool like this fits alongside what you already run, what an employee wellbeing platform is is the better starting point.