HomeProgramme Intelligence › Data & Confidentiality
Data & Confidentiality

The Industry Is Missing the Sensor, Not the Analysis.

Existing mobility technology reports on transactions. Nothing instruments the assignment itself. Bearings is the consented primary data source — and the confidentiality model is precisely what makes the data trustworthy enough to act on.

Book a Demo →Programme Intelligence

The Thesis

The Industry Is Not Short of Analysis. It Is Short of a Sensor.

Every mobility technology on the market reports on transactional data: what was booked, what was shipped, what was invoiced, what a ticket was closed against. That is a record of the process, not of the assignment. Nothing currently instruments the thing that actually determines whether a relocation works — what happens to the person in the field, week by week, once the truck has gone.

Bearings is that missing layer. Because the coaching is genuinely useful, assignees use it — and because they use it, the platform becomes the first consented primary data source in an industry that has been analysing secondary exhaust for twenty years.

1
Coaching earns the signal
Assignees engage because the help is real. Nobody fills in a wellbeing survey for twelve months.
2
Friction gets described
A debrief is an employee voluntarily describing a real problem, in the field, at the moment it happens.
3
Patterns aggregate
De-identified across the population, those descriptions become root causes by lane, location and phase.
4
The programme acts earlier
You change the briefing, the timing, or the destination support — before the next cohort hits the same wall.

Two-Tier Data Model

Confidentiality Is Not a Constraint on the Data. It Is the Reason the Data Is Any Good.

The obvious way to build this product is to give the employer everything: transcripts, sentiment scores, individual risk ratings. It is also the fastest way to destroy the asset. An assignee who believes their coaching is visible to their manager stops describing real problems within a fortnight, and what you are left with is a very expensive way to collect nothing.

So the model has exactly two tiers, and the boundary between them is not negotiable at any price point.

  • Tier one — always on: de-identified aggregate programme reporting. Failure points, root causes, patterns by location, lane and assignment phase.
  • Tier two — assignee initiated: the individual chooses to escalate a specific issue to their programme, with named visibility they control.
  • There is no tier three. No employer view of individual coaching content exists to be bought, requested, or subpoenaed into being.
  • Small-cohort suppression — aggregate views do not report where the population is small enough to identify someone.
See What Tier One Produces →
Programme View — What Is Visible
Cohort friction by lane and phaseAlways
Root cause themes by destinationAlways
Engagement and outcome trendsAlways
A named individual issueOpt-in only
Coaching transcriptsNever
Individual sentiment scoresNever
Partner or family detailNever
Early warning: the programme sees that a pattern has been detected and what to do about it. It does not see the underlying evidence.

Where It Comes From

Built by Someone Who Has Been on the Supply Side of These Failures

Bearings did not start as a software idea. It started from managing international corporate relocations at an established FIDI and IAM member moving company — the position where you see assignment failures happening in real time and watch every one of them go unrecorded by anyone.

The supply side knows which lanes go wrong, which destinations break people, and which assignments were in trouble months before anyone in HR found out. None of it is captured, because nobody in the chain is instrumented to capture it. That gap is the whole product.

  • Operator-built rather than vendor-built — the failure modes are observed, not researched
  • Designed around what a real assignee will actually engage with in month seven
  • Data model shaped by what a mobility director can defend to works councils and legal
Governance & Controls
Assignee data ownershipIndividual
Aggregate reporting thresholdSuppressed at low n
Escalation visibilityAssignee-controlled
Data residencyRegional options
SSO / HRISSupported
Export & deletionOn request
Certifications, DPAs, sub-processor list and works council documentation are provided during procurement.

Model Output Controls

Cultural Advice Is a Domain Where a Confident Invention Is Hard to Catch

If a model invents a statistic, someone eventually checks it. If it invents a custom, a hierarchy convention, or the meaning of a gesture, the assignee acts on it in a live meeting and nobody ever finds out why the relationship cooled. That asymmetry is why output discipline is treated as a safety property here rather than a quality metric.

ControlWhat it does
Grounded generationSpecific claims about customs, conventions and practices are drawn from a maintained evidence base and the assignee’s own situation data rather than generated from scratch.
Stated confidenceThin evidence produces an explicit “I do not have a reliable read on this, here is what to check” rather than a fluent guess.
Competing readsAmbiguous situations return more than one plausible explanation with the evidence that would distinguish them, rather than a single confident answer.
Hard refusalsMedical, legal, immigration and formal HR questions are routed to the appropriate human channel. No opinion is offered at any confidence level.
Coach, do not actNothing is sent, posted or scheduled on the assignee’s behalf. Every output is a read to test, which keeps a wrong answer visible and correctable.
Field correctionReported outcomes feed back into the evidence base, so guidance that fails in practice is corrected rather than repeated across the population.

Bring Your Data Protection Lead to the Demo.

We would rather have the hard questions in the first meeting than the fourth. Book 30 minutes and we will walk the model end to end.

Book a Team Demo → See Pricing

FAQ

Common Questions

Can we buy access to individual coaching data?
No. There is no employer view of individual coaching content in the product, and that is deliberate — it is not a feature held back for a higher tier. The moment assignees believe it exists, the aggregate data stops being worth anything.
Then what does the early warning engine actually tell us?
That a pattern consistent with historical early returns has been detected, and what the recommended action is. Programmes act on the prompt, not on the underlying evidence. In practice that is enough, because the action is almost always a conversation someone needs to have.
What does assignee-initiated escalation mean in practice?
The assignee chooses to raise a specific issue — a housing problem, a manager conflict, a support gap — and controls who sees it. It arrives named, with their consent, because they wanted something done about it.
How do you de-identify without destroying the signal?
Aggregation happens at cohort level by lane, location and assignment phase, with suppression below a minimum population size. The useful finding is almost never “this person is struggling” — it is “every assignee into this location hits the same wall at week nine.”
Where is data held and under what terms?
Regional residency options are available, with DPAs, sub-processor lists and works council documentation provided during procurement. SSO and HRIS integration are supported.
How do you prevent the AI making things up?
Grounded generation against a maintained evidence base, explicit confidence statements, competing explanations instead of single confident answers, hard refusals on medical, legal, immigration and HR questions, no autonomous action on the assignee’s behalf, and field outcomes fed back to correct guidance that does not survive contact with reality.