Main site

Banking AI agent · Wealth advisory

Wealth-Advisory Research & Meeting-Prep AI Copilot

Assemble the research behind a holding, a fund or a market question, prepare the client's position and draft the meeting material — the recommendation and the suitability assessment stay with a licensed adviser.

Typical delivery
4–6 weeks
Deployment
Your stack
Client material
Approved first
After launch
Agent Care

What this agent does

Assembles what the adviser needs to advise

In

  1. 01

    Take the question as the adviser put it — a holding, a fund, a sector or something a client said.

  2. 02

    Read the client's position from the systems of record — holdings, share class, cost basis and mandate.

Reason

  1. 03

    Retrieve the firm's approved research and the product documentation in force, with its date and version.

  2. 04

    Set the holding against the portfolio the client actually has — weight, overlap and what the mandate allows.

  3. 05

    Draft the note with the balance the rules ask for — the case against, the risks, the charges, the conflicts.

Decide

  1. 06

    Check each figure against its source, and flag prices, holdings and documents that have gone stale.

  2. 07

    Check the requester's entitlement before research enters the pack, and hold advice-class questions for the adviser.

Out

  1. 08

    Hand the adviser a pack and a draft — every claim carrying its source, its date and its version.

  2. 09

    Retain what was read, what was drafted, what the adviser changed and who approved what a client saw.

Product statement

The agent researches, assembles and drafts. The recommendation, the suitability assessment and the approval of anything a client sees stay with a licensed adviser and your compliance function.

Example workflow

One adviser request, end to end

  1. 1Request receivedA question on a holding, a fund, a sector or a market move — or a review meeting falling due
  2. 2Position assembledHoldings, share class, cost basis, mandate and restrictions, taken from the systems of record
  3. 3Research retrievedThe firm's approved research, the house view and the product documentation in force, each with its date
  4. 4Balance and entitlement checkedRisk warnings, charges, past-performance wording, conflicts to disclose, and what this requester may see

No human action required

Stages 1 to 4 run before the adviser opens the pack — the position, the research and the balance checks are done first. Nothing in that stretch reaches a client or names a recommendation.

5DecisionSplits on how current the sources are and whether advice was asked for
Sources current, question in scope

Reaches the adviser as a draft pack.

Source moved, or advice asked for

Held for the adviser before anything is drafted.

Licensed adviser

Reads the pack, the sources and their dates, decides what the client is told, and owns the recommendation and the suitability assessment.

Approve · Amend · Send back

Adviser approved — handed back

  1. 6Adviser decides, agent filesOnly what the adviser approved — the pack, the note and any client material — written where access and policy allow
  2. 7Outcome evaluatedFigures traced to source, stale documents caught, required disclosures present, and adviser amendments

Corrections

A figure the adviser had to re-check against its source counts as a failure.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Making a recommendation to a client.
    • Assessing whether a product suits a client.
    • Approving a financial promotion.
    • Sending anything to a client.

  • Automation boundary

    Agent acts unaided

    • ✓Assemble the client's position, share class and mandate from the record.
    • ✓Retrieve the approved research in force, with its date.
    • ✓Draft the note against your own risk-warning, charges and conflicts rules.
    • ✓Flag a stale price or a superseded document.

    Write actions run only inside the approval boundaries agreed during implementation. Anything sent to a client is not one.

    • Placing, amending or cancelling an order.
    • Changing a mandate, model or risk profile.
    • Issuing research outside the firm.
  • Changing the approved source list or entitlements.

Example output

One adviser request, annotated

Everything the agent proposes is attached to the position and the document it came from.

Research pack · single request

Illustrative example
Adviser asks
Current view on a fund
Client holds
Two share classes
Review due
In 4 days
Pack assembled
House view, unchanged
Confidence
87%
Suitability
For the adviser to assess

As received

The question as the adviser put it and the position read from your systems — nothing on this side is inferred.

Evidence used

  • Approved fund research
  • Product document, dated
  • Custodian positions

Why both classes are shown

Same fund, different charges — the pack marks what does not carry across.

What the score decides

How much the adviser re-checks before using it — never whether the fund suits this client.

Action

ApproveAmendSend back

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 adviser request

Reviews, client questions and market moves

03Sources & dates

Work from the approved source

Read the firm's current research and the product documentation in force for the share class the client actually holds — not from what the model remembers.

01Approved path

Start from a pack, not a blank page

The position, the house view and the product documents arrive together and dated, so the adviser is not chasing a fact sheet the day before the meeting.

02Human review

Hold the draft when a source moves

A price, a holding or a document that has been superseded stops the draft, and the adviser is told which one moved and when.

04

Build an evidence trail

Retain the request, the sources read with their dates, the draft, the adviser's amendments and who approved what a client saw — on both paths.

Integrations

Typical integrations

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

  • Research & house view

    • Approved research notes
    • house view
    • Analyst models
    • research distribution

  • Product documentation

    • Prospectus and fact sheets
    • KIDs
    • Fund data feeds
    • share-class terms

  • Portfolio & custody

    • Portfolio management
    • custodian positions
    • Pricing and valuation feeds

Agent

Wealth-advisory research

  • Reads approved research
  • Assembles the position
  • Drafts for review

  • Approvals & entitlements

    • Approval workflow
    • restricted lists
    • Wall-crossing records
    • entitlements

  • 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 meeting

Each control wraps the one inside it. A pack clears every layer before an adviser reads it, and approving anything a client sees sits outside all six.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeReturn advisers to unaided research if evaluations or production signals degrade.Roll back
L5TraceabilityRecord the sources read, the draft, the amendments and the approval.Record
L4Adviser gateNo recommendation, suitability view or client item without a licensed adviser.Gate
L3Communication checksRisk warnings, charges and conflicts are checked against your rules.Check
L2Staleness checksPrices and documents are checked against your staleness window.Flag
L1Source and entitlementClaims carry their source, and entitlement is checked first.Bind
Model corePack proposed — the position, the sources read with their dates, the draft note and the confidence
L1 – L2
Decide whether a source may be used at all
L3
Decides what a client may be shown
L4 – L5
Keep the advice with a licensed person, on the record
L6
Pulls automation back when signals degrade

How Nestack evaluates it

Evaluate the whole pack — not only whether the facts were right.

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

Surface — the pack the adviser opens

  1. E1Source evaluationWas every figure taken from the approved document in force?
  2. E2Currency evaluationWere stale prices, holdings and superseded documents caught?
  3. E3Communication evaluationDid the draft carry the risk warnings, charges and conflicts required?
  4. E4Boundary calibrationWere advice and suitability questions actually held for the adviser?
  5. E5Slice evaluationHow does performance change across specific request cohorts?
  6. E6Business outcomeHow much of each pack did the adviser re-source or rewrite?

Floor — material an adviser could put in front of a client

Failure modes

Where each failure originates in the agent

Seven failure modes plotted against the five stages of the agent lifecycle.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    2 modes

    • WA-01

      Superseded document used

      A fact sheet replaced after a term change is read as current.

    • WA-02

      Restricted source crossed

      Research this requester is not cleared for is pulled into the pack.

    Stage gathers

    Approved research, product documents and entitlements

  2. 02 · Position

    2 modes

    • WA-03

      Stale price or holding

      The position is read before the feed repriced, so weights are wrong.

    • WA-04

      Held-away assets left out

      What is custodied elsewhere is missing, so the adviser rebuilds it.

    Stage assembles

    The client's holdings, share classes and prices

  3. 03 · Drafting

    1 mode

    • WA-05

      Past performance shown bare

      A return figure appears without its period, its basis or the warning.

    Stage drafts

    The note, its warnings, charges and disclosures

  4. 04 · Handover

    1 mode

    • WA-06

      Note reads as a recommendation

      The draft tells the client what to do, not what is known.

    Stage presents

    The pack an adviser reads, with nothing sent

  5. 05 · Change / Version

    1 mode

    • WA-07

      Research update lands late

      A house-view change arrives after the pack was already built.

    Stage tracks

    Research updates, document versions and prompts

  • Sev-1 · a client is told the wrong thing
  • Sev-2 · the adviser uses a wrong figure
  • Sev-3 · the pack is thin and rebuilt
See the complete Banking failure-mode catalogue

Affected slices

The fund that changed while nobody re-read it

A core model holding is close to a lookup — nothing moved and everybody quotes the same price. Failures land where terms changed mid-year, and where the assets sit elsewhere. Nestack reports by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Recently restructured funds4.0%2.6×Review
Structured and illiquid products3.3%2.1×Review
Held-away and legacy assets2.9%1.9×Watch
Core model-portfolio holdings1.4%0.9×Normal

Bar: mis-sourced-pack lift vs. model-holding baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

A superseded document does not announce itself

It surfaces when an adviser re-checks a figure the pack asserted, or when the fund house restates a change nobody upstream had picked up.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Re-sourced figures, adviser rewrites or late document versions rise in one request cohort.

  2. 02

    Diagnose

    Located in the document retrieved, its version, the position feed or the wording of the draft.

  3. 03

    Improve

    The source list, the staleness rule or the drafting template is re-approved by compliance and version-linked.

  4. 04

    Verify

    Packs already used in meetings are rebuilt against the corrected sources.

  5. 05

    Learn

    The document that had moved is kept beside the pack it broke, and the staleness window is narrowed.

  6. Learn → Detect

    The return edge. Product terms keep moving after a pack is written, so each cycle re-reads packs the last one had already signed off.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, approved sources, positions and drafting, evaluation, the adviser workbench, then production validation and handover.

Workstream

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

  1. 01Advice-boundary and request-type discovery.Week 1
  2. 02Research and document-source assessment.Week 1, Week 2
  3. 03Approved source list and version tagging.Week 2
  4. 04Portfolio and custody position assembly.Week 2, Week 3
  5. 05Staleness rules for prices and documents.Week 3
  6. 06Entitlement rules and information-barrier filters.Week 3, Week 4
  7. 07Drafting templates with required disclosures.Week 3, Week 4
  8. 08Adviser gate and promotion-approval routing.Week 4
  9. 09Slice evaluation suite and regression case packs.Week 4, Week 5
  10. 10Adviser workbench and CRM integration.Week 5, Week 6
  11. 11Observability and trace 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 desk, one product setProductionProduction advisory integrationAdvancedMulti-entity / cross-border
Introduced at Pilot
Research pack from approved sources✓✓✓
Client position, share class and mandate✓✓✓
Staleness and entitlement checks on sources✓✓✓
Adviser gate before anything reaches a client✓✓✓
Audit trail of sources, drafts and approvals✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Additional product sets and research sources—✓✓
Adviser workbench and CRM integration—✓✓
Observability and evaluation—✓✓
Introduced at Advanced
Multi-entity and cross-border product sets——✓
High volume and additional languages——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on the desks and product sets in scope, research and document sources, portfolio and custody integrations, request 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.

  1. 01The requests in scope, and who is licensed to advise on them

    Advice-boundary and request-type discovery

    Week 1

  2. 02Your approved research and the product documents you rely on

    Approved source list and version tagging

    Week 2

  3. 03Access to portfolio, custody and pricing systems

    Portfolio and custody position assembly

    Week 2

  4. 04How old a price or a document may be before you stop using it

    Staleness rules for prices and documents

    Week 3

  5. 05Your restricted lists, and who may see which research

    Entitlement and information-barrier filters

    Week 3

  6. 06The disclosures your client material has to carry

    Drafting templates and promotion-approval routing

    Week 4

  7. 07Named advisers and a compliance reviewer

    Adviser gate, then supervised packs in live meetings

    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 actually occupy. Week 5 runs on meetings already held, so no client reads agent-drafted material to prove the agent.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1The requests in scope, the advice boundary and who signs what

W2Approved source access, version tagging and the position feed

W3Staleness rules, entitlement filters and drafting templates

W4Promotion-approval routing, the evaluation suite and the slices

W5Packs rebuilt for meetings already held, nothing sent to a client

W6Advisers work live packs under review, then handover

Reading the band
Entitlement filters and staleness rules finish in week 3 and are tested in week 4. Nothing in week 5 runs on a source those checks have not already covered.
At the end of W6
A pack has been built, checked and used in a live meeting under supervision for each request type in scope, and every figure in it has been traced back to the document it came from, then Agent Care takes over monitoring.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Banking AI agent

Build a research copilot around the sources your advisers already trust.

Show us the questions your advisers ask, the research and product documents you already approve, and who signs what before it reaches a client. We'll rebuild a set of packs your advisers wrote by hand and show you where each figure came from — which sources had already been superseded, and which lines would have needed a compliance approval before a client saw them.

Typical delivery
4–6 weeks
Deployment
Your advisory stack
Client material
Approved before it is sent
After launch
Agent Care available

Nestack Agents · Wealth-advisory research

AGT-BK-06 · Agent Care available after launch

Build a research copilot

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

About you
Your setup

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