Main site

Education AI agent · Title IV returns

Title IV Withdrawal & Return-of-Funds AI Agent

Build the R2T4 file a financial aid administrator determines from — the withdrawal-date evidence, the return position per fund, the clocks — and be the desk our student-support agent hands anything touching aid to.

Typical delivery
4–6 weeks
Deployment
Your stack
FAA determines
Never dates
After launch
Agent Care

What this agent does

Builds the file, never the withdrawal date

In

  1. 01

    A student stops attending in week four and is noticed in week nine, and the file carries both dates, not one.

  2. 02

    A last date comes back from faculty, and a grade-entry date is refused as an attendance fact.

Reason

  1. 03

    A determination is made, and the file dates it and names the administrator who made it.

  2. 04

    A bare LMS login is not academic engagement under 34 CFR 600.2, and is refused.

  3. 05

    A student states the date they stopped, and it stays unsupported until a record carries it.

Decide

  1. 06

    A module registered but never begun leaves the denominator, from 1 July 2026.

  2. 07

    A loan post-withdrawal disbursement waits on an affirmative reply.

Out

  1. 08

    A calculation deadline is quoted from ED guidance, and the file does not claim the regulation states it.

  2. 09

    Execute write actions only inside the approval boundaries agreed during implementation.

Product statement

The agent assembles the evidence and runs the clocks; a named financial aid administrator determines the withdrawal date, and the institution makes the return.

Example workflow

One withdrawal, evidence to return

  1. 1Withdrawal signalledFaculty attendance record, LMS activity, a registrar drop or a bursar charge reversal
  2. 2Evidence gatheredEach attendance fact with the record it came from, its date and whether it meets academic engagement
  3. 3File assembledLast-date position and its source, calculation inputs, return positions per fund and confidence
  4. 4Controls appliedEngagement-definition checks, module-begun checks, clock arithmetic and confidence threshold

No human action required

Stages 1 to 4 run unaided, and none of them determines a withdrawal date — the agent is assembling, and the administrator lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the financial aid administrator.

Low confidence

Adds a director-of-aid read first.

Administrator review

The file is held with its evidence, the source of each date and the confidence.

Review · Correct · Send to the director

Reviewed — the administrator determines

  1. 6Aid and bursar systems updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedEvidence completeness, date corrections, return timeliness and findings raised at audit

Corrections

Every administrator correction is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Determining the withdrawal date, or the date of determination the clocks hang off.
    • Making the return, or moving money to the Department or a lender.
    • Packaging aid, or releasing a post-withdrawal disbursement.
    • Applying the full-refund withdrawal exemption to a student.

  • Automation boundary

    Agent acts unaided

    • ✓Gather attendance facts and name the record each one came from.
    • ✓Run the clocks from the determination and show the days left.
    • ✓Assemble the calculation inputs and the position per fund.
    • ✓Show what the file is missing, and hold it open.

    Wrong after the determination and the return already went out — on the wrong amount, or late.

    • Approving a leave of absence, or judging that one held.
    • Student status reporting — the registrar's act, and our enrollment and records agent.
    • Resolving conflicting information, or referring anything to the OIG.
  • Changes to clock, evidence or determination-control configuration.

Example output

One withdrawal, annotated

Everything the agent assembles is attached to the record it came from.

Withdrawal-file output · single student

Illustrative example
Student
Term withdrawal
Evidence line
Last graded discussion post sits four weeks before the LMS login the report used
Days to return
Day 31 of 45
Rule applied
34 CFR 600.2
Confidence
88%
Determination status
Undetermined — evidence conflict

As received

Taken from the attendance records and the registrar drop as received — nothing on this side is the agent's.

Record used

  • Graded discussion post
  • LMS activity report
  • Registrar drop record

Why this flag

A login with nothing after it is not academic engagement; the graded post is the fact.

What the score decides

Below the configured threshold the file picks up a director read before the administrator sees it.

Action

ReviewCorrectSend to the director

Value

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

Every withdrawal

From the attendance and registrar records

03Assembly

Assemble against the record

Draw on faculty attendance, registrar and bursar records and the module map. The rules at 34 CFR 668.22 were rewritten with a general effective date of 1 July 2026, and what moved was the evidence rather than the arithmetic.

01Approved path

Late is priced, wrong is not

Routine evidence gathering and clock arithmetic arrive already done.

02Human review

Point people at the clock

A return outside the timeframe at 34 CFR 668.173, across enough of an audit sample, is priced in an irrevocable letter of credit; a calculation error is not. And R2T4 is not the top audit finding — in each ranking ED has published, student status reporting leads it.

04

Build an evidence trail

The date, the evidence it rests on and the administrator who determined stay on the file.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • Student systems

    • Banner
    • Colleague
    • Workday Student
    • Jenzabar

  • Financial aid systems

    • COD and G5
    • PowerFAIDS
    • Regent Award
    • CampusLogic

  • Attendance and LMS

    • Canvas
    • Blackboard
    • Moodle
    • class rosters

Agent

Title IV withdrawal and returns

  • Reads the records
  • Assembles the file
  • Holds for the administrator

  • Bursar and ledger

    • TouchNet
    • Nelnet
    • Cashiering ledgers

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and the return

The six are taken outward to inward. What none of them stops appears in the map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeRevert to withdrawal-evidence assembly when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, clock-rule and calculation-input configuration changes.Track
L4TraceabilityRecord the evidence, the source of each date, the flags and the review time.Record
L3Administrator gateHolds the file for the named administrator; it governs what is shown, not whether the date chosen is right.Gate
L2Policy guardrailsTest each attendance fact against the academic-engagement definition; a failure returns the fact.Restrict
L1Confidence thresholdsRoute low-confidence dates and module maps to a director read first.Require review
Model coreFile produced — last-date position, calculation inputs, return positions and confidence
L1 – L2
Test whether a date may stand
L3
Puts the file in the administrator's hands
L4 – L5
Keep the date and the evidence behind it
L6
Reverts to withdrawal-evidence assembly when signals degrade

How Nestack evaluates it

Evaluate the whole file — not only the percentage at the end.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the file the administrator reads

  1. E1Final-output evaluationDid each date in the file rest on an institutional record?
  2. E2Step-level evaluationDid the agent use the right term, module map and clock configuration?
  3. E3Tool evaluationDid it read the correct student and write the correct record?
  4. E4Confidence calibrationDo low-confidence dates actually attract more administrator corrections?
  5. E5Slice evaluationHow far apart do evidence rates sit across specific withdrawal types?
  6. E6Business outcomeHow many dates were corrected, and how many surfaced only at audit?

Floor — the return the institution answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, placed where in the lifecycle each one starts.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • HI-03

      Grade-entry date as the LDA

      A faculty return gives the date the grade was entered.

    Stage gathers

    Attendance, LMS, registrar and bursar records

  2. 02 · Reasoning

    2 modes

    • HI-04

      Login taken as engagement

      A bare LMS login is carried as an attendance fact.

    • HI-06

      Registered module counted

      A module never begun is left in the denominator.

    Stage proposes

    Last-date position, returns per fund and confidence

  3. 03 · Tool / write

    2 modes

    • HI-02

      Determination date drifts

      The date the calculation ran is recorded as the determination.

    • HI-05

      Second-period hours accrue

      Hours accrue where the prior period was never completed.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • HI-01

      Self-report stands alone

      The only support for a date is the student's own statement.

    Stage returns

    The file the administrator determines from

  5. 05 · Change / Version

    1 mode

    • HI-07

      Freeze-date logic persists

      An inherited sheet still fixes the denominator in advance.

    Stage tracks

    Model, prompt, clock rules and module config

  • Sev-1 · a date is set without a person
  • Sev-2 · a return goes out late or wrong
  • Sev-3 · source degrades, file goes to review
See the complete Education failure-mode catalogue

Affected slices

One withdrawal type can carry most of the gaps

A term-level attendance-evidence figure can read sound while one fund and one withdrawal type hold most of the unsupported dates. Nestack reports the gap rate by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Unofficial withdrawals in modules9.5%3.7×Review
Clock-hour second-period exits6.9%2.7×Review
Direct Loan post-withdrawal cases4.7%1.8×Watch
Pell-only official withdrawals2.3%0.9×Normal

Bar: gap-rate lift vs. the Pell-only official-withdrawal baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

A late return ends as a test

The loop shuts when the late return is a regression case. That suite is what the next withdrawal worked is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Gap rate rises in one withdrawal type.

  2. 02

    Diagnose

    The student who stopped attending in week four and was noticed in week nine is read back through the attendance records until the cause narrows to one.

  3. 03

    Improve

    Any change goes out numbered, with the withdrawals that caused it attached.

  4. 04

    Verify

    One withdrawal case still failing is enough to hold the release back.

  5. 05

    Learn

    One case joins the suite, one line joins the student record.

  6. Learn → Detect

    The return edge. The next withdrawal is detected 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, withdrawal workflow, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Withdrawal workflow discovery and refusal boundary.Week 1
  2. 02Registrar, bursar and aid source assessment.Week 1, Week 2
  3. 03Clock, evidence and calculation-input rule mapping.Week 1, Week 2
  4. 04Attendance and enrolment ingestion.Week 2
  5. 05Clock arithmetic and evidence binding.Week 2, Week 3
  6. 06Confidence scoring and finding routing.Week 3
  7. 07Administrator review workflow.Week 3, Week 4
  8. 08Aid and bursar-system integration.Week 5
  9. 09Date and calculation cases.Week 4, Week 5
  10. 10Guardrails and determination controls.Week 4
  11. 11Withdrawal-trail instrumentation.Week 5, Week 6
  12. 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 tierPilotOne campus, one termProductionProduction aid and student systemsAdvancedMultiple campuses / systems
Introduced at Pilot
Evidence and clocks against your records✓✓✓
Administrator review✓✓✓
Attendance-evidence baseline✓✓✓
Introduced at Production
Reporting by fund—✓✓
Review workflow in your systems—✓✓
Approved record write-back—✓✓
Registrar-and-bursar integration—✓✓
Introduced at Advanced
Multi-programme calculation rules——✓
Multi-stage aid-office review——✓
High withdrawal volume——✓
Multi-campus return controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, withdrawal volume, review 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.

  1. 01Your attendance sources and where drops land

    Attendance and enrolment ingestion, and evidence binding

    Week 1

  2. 02Representative withdrawals from a past term

    Attendance-evidence baseline, date normalisation and rule binding

    Week 2

  3. 03Your calculation inputs and your module map

    Clock, evidence and calculation-input rule mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Registrar, bursar and aid-source review, then integration setup

    Week 2

  5. 05Returns you would not want audited

    Clock cases and the evaluation suite

    Week 4

  6. 06What no determination may assume

    Confidence scoring, finding routing, guardrails and determination controls

    Week 3

  7. 07The named financial aid administrator

    Administrator review workflow, 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

Widths on this chart come from worked weeks, not from layout, and that is why the fifth doubles.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Withdrawal workflow discovery, rule mapping and the refusal boundary

W2Record integration and the attendance-evidence baseline

W3Clock arithmetic, calculation inputs and review controls

W4Evaluation suite, clock cases and failure-mode testing

W5Aid and bursar integration, a pilot term and targeted corrections

W6One term run under the financial aid administrator, then Agent Care handover

Reading the band
A bar covers only the weeks its own work is named in. The fifth-week doubling is real, not padding.
At the end of W6
Once the term validates, Agent Care assumes the running agent.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Education AI agent

Build a Title IV returns agent around the administrator who determines.

Show us your attendance sources, your module map and who determines a withdrawal date. That determination belongs to one named financial aid administrator, it is dated, and every clock in 34 CFR 668.22 runs from it.

Typical delivery
4–6 weeks
Deployment
Your aid and student stack
FAA determines
Never dates
After launch
Agent Care available

Nestack Agents · Title IV withdrawal and returns

AGT-ED-10 · Agent Care available after launch

Build a Title IV returns agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your withdrawals

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.