Real Estate AI agent · Valuation support
Valuation-Support AI Assistant
Assemble candidate comparables and their provenance from licensed records, screen draft narrative for prohibited and proxy language, and hold the pool for the state-certified appraiser, who forms the opinion of value.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Appraiser review
- Pre-signature
- After launch
- Agent Care
What this agent does
Assembles the evidence, not the opinion of value
In
- 01
Retrieve licensed MLS, public and prior-file records for the subject and the market named in the assignment.
- 02
Deduplicate candidates across sources and keep the record identifier, the status and the retrieval time on each.
Reason
- 03
Normalise units, dates, living area and lot measures so records from different sources can be read together.
- 04
Retrieve the sale history behind each candidate, and mark the terms nobody has verified yet.
- 05
Normalise the subject's own attributes against the assignment file already open in the forms software.
Decide
- 06
Screen retained rationales and draft narrative for prohibited wording and protected-class proxies.
- 07
Screen the order file for a contract price or target value, and withhold it from the candidate ranking.
Out
- 08
Retain the retrieval log, the version stamps and what the appraiser accepted, replaced or rejected.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles candidates and evidence; the state-certified appraiser forms the opinion of value and signs the certification that carries it.
Example workflow
One assignment, order to workfile
- 1Assignment receivedOrder file, subject record, forms-software file or AMC request
- 2Candidates retrievedLicensed MLS and public records, sale history, status and retrieval stamps
- 3Pool assembledCandidate comparables, provenance, unverified fields and confidence
- 4Controls appliedLicence-scope checks, prohibited-wording and proxy screens, order-file redaction and confidence threshold
No human action required
Stages 1 to 4 run unaided, and no opinion of value is formed at any of them — the agent is assembling evidence, and the appraiser's lane opens at the confidence gate.
Goes to the signing appraiser to review.
Adds a review-appraiser read first.
Appraiser review
The pool is held with its provenance, its unverified fields and the confidence.
Accept · Replace · Send back for sourcing
Accepted — carried to the workfile
- 6Forms software updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedComp-set rejection, replacement reasons, screen hits and post-signature findings
Replacements
Every appraiser replacement is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Forming the opinion of value and reconciling it.
- Finalising the comparable set behind a value.
- Deriving and applying market adjustments.
- Signing the report and its SR 2-3 certification.
Automation boundary
Agent acts unaided
- ✓Retrieve and deduplicate licensed records into a candidate pool.
- ✓Normalise units, dates, living area and lot measures.
- ✓Screen draft narrative for prohibited and proxy wording.
- ✓Assemble the workfile with provenance, retrieval timestamps and version stamps.
Every retrieval, screen and export runs inside the boundaries agreed at implementation.
- Inspecting the subject or verifying sale terms.
- Setting scope of work or accepting for competency.
- Answering a Reconsideration of Value request.
- Submitting to UCDP, an AMC or the lender.
Example output
One candidate comparable, annotated
Everything the agent proposes carries the record it was retrieved from.
Candidate output · single comparable
Illustrative example- Comparable
- Closed sale, same tract
- Retrieved line
- Two-storey resale, four blocks from the subject
- Living area
- 1,712 sq ft
- Source of record
- MLS record, closed status
- Confidence
- 91%
- Verification
- Sale terms not yet verified
As received
Taken from the licensed record and its provenance log — nothing on this side is inferred by the agent.
Provenance shown
- MLS record identifier
- Retrieval timestamp
- Closed status and date
Why this candidate
Screened for proxy and prohibited language before the appraiser reads it.
What the score decides
Below the threshold the pool picks up a second sourcing pass before review.
Action
AcceptReplaceSend back for sourcingValue
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 assignment
From the assignment file
Retrieve against the file
Draw on licensed records and the search criteria set in the assignment file.
Widen the search, not the call
Routine retrieval, deduplication and normalisation arrive already done.
Send the read to the thin evidence
Unverified terms and low-confidence candidates are marked, so the appraiser's read starts where the evidence is thinnest.
Build an evidence trail
Provenance, retrieval timestamps and version stamps go into the workfile, not a side log.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Appraisal forms software
- ACI Sky
- ACI Report
- Bradford ClickFORMS
MLS data infrastructure
- RESO Web API
- MLS Grid
- licensed feeds
Property data
- CoreLogic
- HouseCanary
Agent
Valuation support
- Retrieves records
- Assembles candidates
- Holds for the appraiser
Order and review rails
- Reggora
- ValueLink
- Fannie CU
- Freddie LCA
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 workfile
Each control wraps the one inside it. What a layer leaves uncaught is named in the failure-mode map below it.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a candidate may stand
- L3
- Puts acceptance in an appraiser's hands
- L4 – L5
- Keep the record and the version honest
- L6
- Pulls retrieval back when signals degrade
How Nestack evaluates it
Evaluate the retrieval workflow — not only the accepted comp set.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the pool the appraiser reads
- E1Final-output evaluationDid each candidate match the record it was drawn from?
- E2Step-level evaluationDid the agent use the right licence scope, criteria and screen list?
- E3Tool evaluationDid it read the correct feed and write the correct file?
- E4Confidence calibrationDo low-confidence pools actually attract more replacements?
- E5Slice evaluationHow does performance change across specific property cohorts?
- E6Business outcomeHow many candidates were replaced, and what was found after signature?
Floor — the outcome the appraiser answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each placed where it starts rather than where it shows.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- VS-01
Stale-mirror comparable
A withdrawn and re-closed sale is read at its earlier price.
Stage gathers
Order file, licensed records and normalised measures
02 · Reasoning
2 modes
- VS-02
Tract covariate as proxy
Neighbourhood scoring tracks a boundary that stands in for race.
- VS-03
Contract-price anchoring
Candidates are shaped by a price sitting in the order file.
Stage proposes
Candidate pool, provenance and confidence
03 · Tool / write
2 modes
- VS-04
Adjustment written to the grid
Numbers reach the form that the appraiser did not derive.
- VS-05
Licence-scope export
Full records leave for a party the data licence excludes.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- VS-06
Prohibited narrative wording
Named unacceptable phrasing reaches the signed report.
Stage returns
The candidate pool the appraiser works from
05 · Change / Version
1 mode
- VS-07
Legacy mapping persists
Form and field mappings outlive the dataset version.
Stage tracks
Model, prompt, screen list and form mappings
- Sev-1 · a held act runs unaided
- Sev-2 · a bad candidate reaches the grid
- Sev-3 · a source degrades, pool routes to review
Affected slices
An acceptable overall rate can sit on one bad cohort
Comp-set rejection rate counts the candidates a signing appraiser removes or replaces before reconciliation. Nestack reports it by cohort, with the worst cohort read against the best.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Majority-Black and majority-Hispanic tracts | 7.6% | 3.6× | Review | |
| Rural non-disclosure states | 5.7% | 2.7× | Review | |
| Manufactured, ADU and mixed-use | 3.8% | 1.8× | Watch | |
| New construction and gut renovations | 1.9% | 0.9× | Normal |
Bar: comp-set-rejection-rate lift vs. all-order baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
A cycle closes when it leaves a test behind
Each closed cycle leaves one more regression case in the suite and one more entry in the screen list, and the next detection is measured against both.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Comp-set rejection rate rises in one cohort.
- 02
Diagnose
The provenance log behind the replaced candidates names the source and the retrieval stamp.
- 03
Improve
A fix lands against a named version, carrying the replacements that prompted it.
- 04
Verify
Nothing releases until the affected regression cases pass again.
- 05
Learn
The case joins the permanent suite and the screen list.
Learn → Detect
The return edge. The next detection runs against a longer suite than this one.
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, retrieval workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Appraisal workflow discovery and boundary definition.Week 1
- 02MLS, public-record and AVM assessment.Week 1, Week 2
- 03Search criteria and prohibited-wording list mapping.Week 1, Week 2
- 04Record retrieval and normalisation.Week 2
- 05Candidate assembly and provenance binding.Week 2, Week 3
- 06Confidence scoring and sourcing escalation.Week 3
- 07Appraiser review workflow.Week 3, Week 4
- 08Forms-software and order-platform integration.Week 5
- 09Evaluation suite and cohort regression cases.Week 4, Week 5
- 10Guardrails, screens and export controls.Week 4
- 11Observability and provenance 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 office, one market | ProductionProduction forms software | AdvancedMulti-panel and AMC |
|---|---|---|---|
| Introduced at Pilot | |||
| Retrieval to your criteria | ✓ | ✓ | ✓ |
| Appraiser review | ✓ | ✓ | ✓ |
| Comp-quality baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Cohort disparity reporting | — | ✓ | ✓ |
| Review workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Workfile provenance export | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-MLS licence scope | — | — | ✓ |
| Multi-stage panel review | — | — | ✓ |
| High order volume | — | — | ✓ |
| Enterprise panel controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transaction 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 assignment file and search criteria
→Criteria mapping and record retrieval
Week 102Representative completed workfiles
→Retrieval baseline, normalisation and provenance binding
Week 203Your MLS and data-licence agreements
→Licence-scope enforcement and automation-boundary definition
Week 104Access to relevant APIs, feeds or exports
→Feed, forms-software and order-platform assessment, then setup
Week 205Comparables your appraisers replaced, and why
→Evaluation suite, cohort regression cases and failure-mode testing
Week 406Your review thresholds and write-back boundaries
→Confidence scoring, screen routing, guardrails and review controls
Week 307Named certified appraisers to review pools
→Appraiser 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
Phases are drawn over the weeks they occupy. Week 5 carries evaluation and pilot at the same time.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Appraisal workflow discovery, licence scope and the automation boundary
W2Feed integration and the retrieval baseline
W3Assembly workflow, confidence logic and review controls
W4Evaluation suite, prohibited-wording screens and failure-mode testing
W5Forms-software integration, pilot assignments and targeted corrections
W6A live order returned to the appraiser, then Agent Care handover
- Reading the band
- A bar covers only the weeks its work is named in. The week 5 overlap is real, not padding.
- At the end of W6
- The verification pack is signed and Agent Care begins monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Real Estate AI agent
Build a valuation-support agent around your appraisal workflow.
Show us your forms software, your data licences and your search criteria. The appraiser who signs the certification keeps the value, the comp set and the adjustments, and the boundary sits short of all three.
- Typical delivery
- 4–6 weeks
- Deployment
- Your forms stack
- Appraiser review
- Pre-signature
- After launch
- Agent Care available
Nestack Agents · Valuation support
AGT-RE-03 · Agent Care available after launch