Counselling intensity can be allocated from updated risk evidence instead of manual synthesis alone.
Executive brief
Separate realized OTT1 evidence from OTT2 pilot expectations.
AI value · Prediction accuracy
Company-reportedOTT1 achieved over 95% prediction accuracy; OTT2 above 98% remains a 2026 pilot target
OTT2 98% is prospective and is not counted as achieved.
Before
Provides registration information to the Estonian Unemployment Insurance Fund. → Interprets the one-time score with the client record and allocates support.
After
Combines registry events and client changes to refresh employment-prospect assessments regularly. → Reviews the current assessment and focuses intensive support on higher-risk clients.
Human boundary
Counsellors own interventions; the model is periodically retrained and does not change autonomously.
Why it matters
Counselling intensity can be allocated from updated risk evidence instead of manual synthesis alone.
This case is company-reported. Use it for the operating-model shift; do not treat the numbers as independently measured.
Before
How the work ran before the change.
Newly registered jobseeker
Provides registration information to the Estonian Unemployment Insurance Fund.
ControlOTT1 created one employment-prospect assessment at registration.
Employment counsellor
Interprets the one-time score with the client record and allocates support.
ControlCounsellor owns service choice; published sources do not report a pre-OTT manual cycle time.
What changed
Counselling intensity can be allocated from updated risk evidence instead of manual synthesis alone.
Decision rightAI acts within a human backstop
After
How the same work runs now.
OTT2
Combines registry events and client changes to refresh employment-prospect assessments regularly.
ControlModel is periodically retrained, does not self-modify, and explains influential factors.
Employment counsellor
Reviews the current assessment and focuses intensive support on higher-risk clients.
ControlHuman service decision; the >98% OTT2 target remains under 2026 pilot evaluation.
Exception path
Contradictory or changed circumstances receive counsellor review; OTT2 refreshes assessments.
Decision authority
Counsellors own interventions; the model is periodically retrained and does not change autonomously.
| # | Actor | Action | Control |
|---|---|---|---|
| 01 | Newly registered jobseeker | Provides registration information to the Estonian Unemployment Insurance Fund. | OTT1 created one employment-prospect assessment at registration. |
| 02 | Employment counsellor | Interprets the one-time score with the client record and allocates support. | Counsellor owns service choice; published sources do not report a pre-OTT manual cycle time. |
| # | Actor | Action | Control |
|---|---|---|---|
| 01 | OTT2 | Combines registry events and client changes to refresh employment-prospect assessments regularly. | Model is periodically retrained, does not self-modify, and explains influential factors. |
| 02 | Employment counsellor | Reviews the current assessment and focuses intensive support on higher-risk clients. | Human service decision; the >98% OTT2 target remains under 2026 pilot evaluation. |
Work that left the path
Human role before
Employment counsellors combined registration information, the one-time OTT1 assessment, and the client record to allocate support; the counsellor retained the service decision.
Human role after
Counsellors interpret risk and own service intensity and intervention choices.
AI role
Combines registry data to score employment prospects and explain influential factors.
Manual, laborious risk assessmentOTT1 achieved over 95% prediction accuracy; OTT2 above 98% remains a 2026 pilot target
OTT1 operational since 2020 · More than 100,000 client records
OTT2 98% is prospective and is not counted as achieved.
Anti-pattern
Do not publish forecast OTT2 accuracy as achieved.
Questions
Portability conditions
Reputation risk
low