HR AI agent · Internal mobility
Internal Mobility AI Agent
Match employees to internal openings against the written requirements, keep the considered set beside the slate, and hand the match to the named manager who decides who is shown the opening.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Named manager
- Considered set
- After launch
- Agent Care
What this agent does
Keeps the names it did not surface
In
- 01
An opening is read against its written requirements, and the fields that moved the match are kept.
- 02
The output is a slate, and who is shown the opening under that slate is decided by a person.
Reason
- 03
The considered set is recorded beside the slate, so who was weighed and not shown is knowable.
- 04
Promotion and transfer are terms and conditions under 42 U.S.C. 2000e-2(a)(1), not side matters.
- 05
A match is a measure under 29 CFR 1607.16(Q), and 1607.16(B) names promotion in its own text.
Decide
- 06
After Muldrow, No. 22-193, 17 April 2024, a transfer harm need not be significant to be contested.
- 07
The 42 U.S.C. 2000e-5(e)(1) charge clock runs from the practice, not from the day it is found.
Out
- 08
OFCCP removes its own copy at 41 CFR 60-3 on 26 October 2026, while 29 CFR 1607 stays in force.
- 09
Write to the mobility record only inside the approval boundaries agreed during implementation.
Product statement
Matching, recording and holding belong to the agent. The slate belongs to a named manager, who decides who is shown the opening and owns it there.
Example workflow
One opening, profiles to slate
- 1Profiles and openings receivedHRIS records, declared skills, project histories, internal postings and move history
- 2Requirements fixed and versionedThe written requirements, the weights, the profile version and the day that set was fixed
- 3Matched and ranked apartThe match, the rank, the profile fields that moved it and the requirements it was read against
- 4Controls appliedProxy checks, inference checks, surfacing-ratio checks and match confidence
No human action required
Stages 1 to 4 run unaided, and nobody is dropped from the considered set at any of them — the agent is matching, and the manager lane opens at the slate gate.
Goes to the named manager to surface.
Adds a mobility-lead read first.
Manager review
The slate is held with its requirements, its surfacing figures and the considered set behind it.
Surface · Hold in the pool · Send to mobility lead
Surfaced — by the named manager
- 6Mobility and posting records updatedOnly where write access and personnel records policy allow it
- 7Outcome evaluatedSurfacing ratios, profile accuracy, manager corrections and what review found
Corrections
Each manager correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Deciding who is shown an internal opening.
- Passing over an employee on a match score.
- Setting where the slate line falls.
- Signing off a published bias-audit summary.
Automation boundary
Agent acts unaided
- ✓Read each profile against the written requirements.
- ✓Keep the considered set alongside the slate and the requirements read.
- ✓Flag career-break and source-employer indicators as possible proxies.
- ✓Record the surfacing rate for each named category.
Nobody is shown an opening except by a named manager, inside the agreed boundaries.
- Judging whether a requirement is job related.
- Telling an employee why they were not surfaced.
- Choosing which openings the agent matches on.
- Changes to the profile, the weights or the requirements.
Example output
One matched opening, annotated
This serves a mobility team who may have to account for an opening an employee was never shown; below is one match exactly as the agent leaves it.
Match record · single opening
Illustrative example- Match
- Platform lead, Manchester site
- Recorded as
- Ranked inside the considered set
- Opening
- A match, not a shortlist
- Evidence of record
- Skills profile, 3 August 2026
- Confidence
- Held unsurfaced
- Held for
- The named manager, by name
As received
Built from the stored profile and the written requirements of the opening, and it asserts nothing further.
What the record holds
- Stored profile
- Written requirements
- Considered set
Why no surfacing here
Which employees are shown an opening is a judgement a manager makes.
What the score decides
Below the set threshold a match gets a mobility-lead read before the manager sees it.
Action
SurfaceHold in the poolSend to mobility leadValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Each match
From the profile that carries it
Where the evidence is used
The agent does not vouch for an employee, only for what the profile records, where each entry was read from, and how the match was built on it.
Nobody sees a door not shown
An employee never surfaced for an opening receives no rejection and no notice: from inside their own career, no event occurred.
What was checked, and not found
Both EEOC technical assistance documents on algorithmic selection returned HTTP 404 on 27 August 2026, while the 2007 fact sheet on employment tests and selection procedures still resolves, so a check that stops at the first live page finds guidance where the guidance has gone.
Build an evidence trail
The profile, the opening it was matched to and the manager who considered it stay together.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Core people records
- Workday
- SuccessFactors
- HiBob
- Grades, moves and job history
Internal openings
- Internal postings
- requirements
- What an opening asks for
Employee-declared skills
- Profiles
- project and rotation logs
- What the employee chose to note
Agent
Internal mobility and matching
- Reads the profiles
- Matches and ranks
- Holds for the manager
Skills inference services
- Taxonomy vendors
- inference APIs
- Bought-in skill and level scores
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six meshes between the model and the manager
Six meshes stacked in one frame, the last the closest. What holds is set out in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a slate may stand
- L3
- Leaves the surfacing to a named manager
- L4 – L5
- Keep the match and the profile behind it
- L6
- Shows the opening to the whole population when signals degrade
How Nestack evaluates it
Evaluate the whole match — not only the slate that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the slate a manager works
- E1Final-output evaluationDid the match record the requirements it was actually built on?
- E2Step-level evaluationDid the agent read the right opening, the right profile version and the live weights?
- E3Tool evaluationDid it read and write the correct opening and the correct profile?
- E4Confidence calibrationDo low-confidence matches actually attract more manager corrections?
- E5Slice evaluationHow does performance change across specific opening classes?
- E6Business outcomeHow many slates needed a correction before the manager surfaced?
Floor — the profiles a match rests on
Failure modes
Where each failure originates in the agent
Seven failure modes, each at the point where it first becomes findable.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- UI-03
Profile inferred, never shown
The employee cannot see what the match read.
Stage gathers
The profiles, the openings and the move history
02 · Reasoning
2 modes
- UI-04
Proxy rides in on the taxonomy
Tenure or a career break stands in for a person.
- UI-06
Withdrawn guidance read as live
A removed document is worked as the current one.
Stage proposes
The requirements, the match and the considered set
03 · Tool / write
2 modes
- UI-02
Thin surfacing base passed on
A slate goes forward with most categories unknown.
- UI-05
Considered set not persisted
The population declined to surface is written nowhere.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- UI-01
Surfaced, considered set unrecorded
The record shows a slate but not who was weighed.
Stage returns
The slate a manager reads and the names on it
05 · Change / Version
1 mode
- UI-07
Silent narrowing after retrain
A rule change moves who is shown, and nobody is told.
Stage tracks
Model, prompt, match rules and profile dates
- Sev-1 · an employee never shown an opening
- Sev-2 · a skewed slate reaches a manager
- Sev-3 · signals degrade, opening shown wide
Affected slices
Cross-function moves absorb the corrections
An opening-level surfacing figure can read clean while cross-function lateral moves carry most of the corrections. Nestack reports the correction rate by opening class, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Cross-function lateral moves | 10.5% | 3.7× | Review | |
| Multi-country openings | 7.4% | 2.6× | Review | |
| Career-break returners | 4.6% | 1.6× | Watch | |
| Same-team progression | 2.1% | 0.7× | Normal |
Bar: correction-rate lift vs. same-team baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What a door not shown costs
A cycle shuts when the employee never surfaced for anything is a standing case. That suite is what the next opening posted is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on cross-function lateral moves.
- 02
Diagnose
The opening whose slate quietly stopped holding anyone returning from a career break is worked backwards until one cause is left standing.
- 03
Improve
Matches ship numbered, and the profiles behind them travel attached.
- 04
Verify
Each touched surfacing case is run again, and one red holds it back.
- 05
Learn
The case is kept, and the matching rules change in that same commit.
Learn → Detect
The return edge. The next slate is measured against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, matching and surfacing, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Requirement capture and automation-boundary scoping.Week 1
- 02HRIS, posting and skills-vendor sources.Week 1, Week 2
- 03Match-reconstruction and surfacing-coverage mapping.Week 1, Week 2
- 04Employee profile ingestion.Week 2
- 05Requirement, weight and match binding.Week 2, Week 3
- 06Surfacing scoring and review routing.Week 3
- 07Manager surfacing workflow.Week 3, Week 4
- 08HRIS-and-posting integration.Week 5
- 09Surfacing and inference cases.Week 4, Week 5
- 10Guardrails and matching controls.Week 4
- 11Match-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne opening, one cycle | ProductionProduction mobility workflow | AdvancedMultiple openings / countries |
|---|---|---|---|
| Introduced at Pilot | |||
| Matching to your written requirements | ✓ | ✓ | ✓ |
| Named manager surfacing | ✓ | ✓ | ✓ |
| Skills-inventory baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by opening class | — | ✓ | ✓ |
| Manager review workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Profile-and-requisition integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-source profiles | — | — | ✓ |
| Cross-opening slate packs | — | — | ✓ |
| Large employee populations | — | — | ✓ |
| Multi-population surfacing controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, profile volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your live openings and the requirements each is matched on
→Requirement capture and profile versioning
Week 102Representative profiles, move histories and vendor inferences
→Source binding, match logic and the skills-inventory baseline
Week 203Your posting calendar and the managers it names
→Requirement mapping, profile binding and the automation boundary
Week 104Access to the HRIS APIs, feeds or exports
→HRIS, posting and inference source assessment, then integration setup
Week 205Matches you would not want reconstructed
→Surfacing cases and the evaluation round
Week 406What no match may infer
→Surfacing scoring, review routing, guardrails and release controls
Week 307A named manager who surfaces the opening
→Release to the named manager, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
A band here is as wide as its phase actually costs, so week five shows a pair and not a neater blank.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Requirement discovery, profile versioning and the automation boundary
W2Source integration and the skills-inventory baseline
W3Matching, surfacing logic and release controls
W4Evaluation suite, surfacing cases and proxy testing
W5HRIS integration, pilot matches and targeted corrections
W6One posting cycle run under the talent-mobility lead, then Agent Care handover
- Reading the band
- A bar covers the weeks its own work is named for and no more, and week five is shared by design.
- At the end of W6
- When the match record validates, Agent Care adopts the agent.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Internal mobility
Build an internal mobility agent around the opening your last posting cycle never showed anyone.
Show us one opening you post each quarter and the people it reached. If nobody can list who the system weighed and did not show, the ones passed over were never told, and the 42 U.S.C. 2000e-5(e)(1) clock ran from the act rather than the day they found out.
- Typical delivery
- 4–6 weeks
- Deployment
- Your live openings
- Named manager
- Considered set
- After launch
- Agent Care available
Nestack Agents · Internal mobility
AGT-HR-08 · Agent Care available after launch