Answer student questions from the institution's own published sources, route anything touching aid, immigration or wellbeing to the named professional, and keep each conversation on the record.
Ingest the catalogue, the aid policy pages and the registrar's deadline tables from supported SIS.
02
Normalise each passage and carry it forward with its source and the date it was published.
Reason
03
Answer from the institution's own published text, quoting the passage the answer rests on.
04
Apply the scope, disclosure and redisclosure rules the institution sets under its school-official agreement.
05
Emit answers a keyboard and a screen reader can reach — an answer a student cannot read is a discrimination problem.
Decide
06
Detect a question that turns on aid, immigration status or a wellbeing disclosure, and stop answering it.
07
Route those questions to the named aid administrator, designated school official or counsellor.
Out
08
Retain the question, the source it drew on, the routing and the outcome against the case.
09
Execute write actions only inside the approval boundaries agreed during implementation.
→Product statement
The agent answers from published sources; the named professional decides anything turning on the student's own record, and the institution keeps direct control of that record.
Example workflow
One question, message to hand-off
AgentHuman
1Question receivedStudent portal, chat, email or the advising queue
2Context assembledCatalogue passages, policy pages and deadline tables, each with its publication date
3Answer draftedPlain-language answer, the quoted passage, referral flags and confidence
4Controls appliedSource-currency checks, scope and disclosure rules, aid and wellbeing flags and confidence threshold
No human action required
The first four stages run unaided, and none of them applies a rule to the student's own record — the advisor's lane opens at the confidence gate.
5DecisionBranches at the confidence threshold
High confidence
Goes to the student with its source named.
Low confidence
Waits for the assigned advisor first.
Advisor review
The answer is held with its quoted source, its referral flags and the confidence.
Approve · Edit · Refer to the office
Approved — sent to the student▼
6Advising record updatedOnly where write access and approval policy allow it
7Outcome evaluatedEdit distance, referral outcomes, source currency and corrections made after the fact
Edits
Every advisor edit is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required8 items
Determining or stating aid eligibility or an award amount.
Advising on visa status, work authorisation or course load.
Approving a withdrawal, a drop or a leave of absence.
Responding to a disclosure of self-harm, crisis or abuse.
Automation boundaryAgent acts unaided
✓Quote a published, dated policy or deadline with its source named.
✓Explain a general process without applying it to.
✓Book an appointment with the student's assigned.
✓Route the student to the correct named office, with its hours for the named owner.
Any write happens inside the boundaries agreed at implementation, never ahead of the advisor.
Exercising professional judgement or resolving a verification conflict.
Granting a requirement substitution, waiver or exception.
Releasing a student's record to a parent or a third party.
Placing a student on, or removing one from, an intervention caseload.
Example output
One answer, annotated
Everything the agent answers is attached to the passage it was drawn from.
Advising output · single questionIllustrative example
Question
Answer line
Stated deadline
Source of record
Confidence
Routing
Withdrawal date
The published last day to withdraw from a full-term course
Autumn census date
Academic calendar
93%
No referral triggered
As receivedTaken from the academic calendar as published — nothing on this side is written by the agent.
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the student
Each control wraps the one inside it. What a layer does not catch is named in the failure-mode map below.
L6 · Outermost — last line of defenceInward → L1 · closest to the model
L6Rollback / safe modePull the agent back to referral-only when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, source-index and scope-configuration changes.Track
L4TraceabilityRecord the question, the source quoted, the flags, the routing and the advisor's edits.Record
L3Advisor reviewHold answers for the assigned advisor; it governs release, not whether a released answer is right.Gate
L2Policy guardrailsTest answers against scope and disclosure rules, and test that transcripts are not used for training, profiling, selling or advertising; a failure returns the answer.Restrict
L1Confidence thresholdsRoute low-confidence answers to the advisor before the student sees them.Require review
Model coreAnswer produced — plain-language reply, quoted source, referral flags and confidence
L1 – L2Test whether an answer may stand
L3Puts the release in an advisor's hands
L4 – L5Keep the answer and its source reconstructable
L6Pulls the agent back to referral only when signals degrade
How Nestack evaluates it
Evaluate the advising workflow — not only the answer that came back.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the answer the student reads
Depth of coverage ▼
E1Final-output evaluationDid the answer match the institution's published text?
E2Step-level evaluationDid the agent use the current catalogue, policy page and deadline table?
E3Tool evaluationDid it read the right record and write nothing it was not asked to?
E4Confidence calibrationDo low-confidence answers actually attract more advisor edits?
E5Slice evaluationHow does performance change across specific student groups?
E6Business outcomeHow many answers needed an advisor edit or a correction after the fact?
Floor — the outcome the student lives with
Failure modes
Where each failure originates in the agent
Seven failure modes, placed at the stage each one originates.
Agent lifecycleDirection of processing →
01 · Retrieval1 mode
GU-03
Superseded source quoted
Last year's catalogue or deadline table answers as current.
Stage gathersCatalogue, policy pages, with the source each came from
02 · Reasoning2 modes
GU-04
General rule read onto one student
A published rule is applied to a record the agent cannot see.
GU-06
Course-load advice to a visa holder
A workload answer ignores the full-course-of-study condition.
Stage proposesThe answer, the quoted source and confidence
03 · Tool / write2 modes
GU-02
Registration touched, not explained
A call changes a registration the student only asked about.
GU-05
Disclosure written to a shared note
Wellbeing content lands where the whole office can read it.
Stage writesOnly where write access and approval policy allow it
04 · Output1 mode
GU-01
Answer without a source
A confident reply arrives with no passage behind it.
Stage returnsThe answer the student reads and acts on
05 · Change / Version1 mode
GU-07
Silent policy regression
A policy changes mid-term and the source index is not rebuilt.
Stage tracksModel, prompt, source index and scope config
Sev-1 · acts outside the boundarySev-2 · a wrong answer reaches a studentSev-3 · source degrades, answer routes to review
The metric is the share of advising turns carrying a wrong institution-specific fact. In aggregate it holds steady; by group Nestack reports it by slice rather than in aggregate..
Slice performance — reported separately, not only in aggregateIllustrative example
Slice
Failure rate
Lift
Lift vs. threshold
Status
International students on a visa
4.8%
3.0×
Review
Aid verification and appeals
4.0%
2.5×
Review
Students using assistive technology
2.4%
1.5×
Watch
Continuing full-time students
0.8%
0.5×
Normal
Bar: error-rate lift vs. the all-conversation baseline · scale 0–4.0× · tick marks the 2.0× review threshold2 of 4 slices over threshold
Evidence-linked improvement
A cycle ends with a new test
Nothing closes on an explanation. The cycle ends when the failure exists as a regression case the next release has to pass.
Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn
01Detect
Error rate rises in one student group.
02Diagnose
The advising lead reads the conversations and the sources behind them until one cause survives.
03Improve
The fix is written against a version, with the conversations that motivated it attached.
04Verify
Release is blocked until the affected regression cases pass again.
05Learn
The case joins the permanent suite and the advising playbook.
Learn → DetectThe return edge. The next detection runs 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, advising workflow, evaluation, integration, then production validation and handover.
WorkstreamWeek 1Week 2Week 3Week 4Week 5Week 6
01Advising workflow discovery and boundary definition.
02SIS, CRM and content source assessment.
03Catalogue, policy and deadline source mapping.
04Source ingestion and passage normalisation.
05Answering logic and source binding.
06Confidence scoring and referral routing.
07Advisor review workflow.
08SIS and advising-CRM integration.
09Response regression cases.
10Guardrails and referral controls.
11Conversation-trail instrumentation.
12Deployment, documentation and Agent Care handover.
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallelFinal 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 tierPilotOne campus, one queueProductionProduction advising systemsAdvancedMultiple campuses / systems
Introduced at Pilot
Answering from your sources✓✓✓
Advisor review✓✓✓
Response-quality baseline✓✓✓
Introduced at Production
Reporting by student group—✓✓
Review workflow in your systems—✓✓
Approved write-back—✓✓
SIS integration—✓✓
Introduced at Advanced
Multi-campus policy sets——✓
Multi-stage advising approvals——✓
High conversation volume——✓
Enterprise advising controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build priceConfirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.
Separate from buildBuild 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 bringWe build with it
01Your catalogue, policy pages and deadline tables→Source ingestion and passage mappingWeek 1
02Representative advising conversations→Answering baseline, source binding and scope extractionWeek 2
03Your escalation list and the named offices behind it→Escalation, scope and referral-rule mappingWeek 1
04Access to relevant APIs, feeds or exports→SIS, CRM and content assessment, then integration setupWeek 2
05Conversations you would not want repeated→Response cases and failure-mode testingWeek 4
06Where an answer must stop and wait for the advisor→Confidence scoring, referral routing, guardrails and review controlsWeek 3
07Named advisors to review held answers→Advisor review workflow, then pilot and production validationWeeks 5–6
Nothing else is requiredDeployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Each phase sits on the weeks it actually occupies, and week 5 carries both evaluation and launch work.
PhaseW1W2W3W4W5W6
DiscoveryW1
BuildW2 – W3
EvaluateW4 – W5
Pilot & LaunchW5 – W6
Week focusW1Advising workflow discovery, scope mapping and the automation boundaryW2Source integration and the answering baselineW3Advising workflow, confidence logic and referral controlsW4Evaluation suite, accessibility checks and group error-rate testingW5SIS integration, a pilot advising queue and targeted correctionsW6A live advising queue answered under review, then Agent Care handover
Reading the bandBands sit on the weeks the work is named in and nowhere else. The doubled fifth week is real.
At the end of W6Validation closes on live conversations, and Agent Care picks up monitoring.
DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Education AI agent
Build an advising agent around your institution's referral chain.
Show us your catalogue, your deadline tables and the offices a student gets referred to. Bring one question that went wrong last term, and we will map the routing, set the automation boundary and name what stays with the advisor.
