Accounting AI agent · Client onboarding
Engagement-Letter & Client-Onboarding Assistant
Resolve the parties, assemble the checks and draft the engagement letter from your approved templates — with acceptance, risk rating, screening decisions and signature left to the people authorised to make them.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Human review
- Before issue
- After launch
- Agent Care
What this agent does
Assembles the file, drafts from your templates
In
- 01
Ingest new-client and new-service requests from supported intake, CRM or practice-management sources.
- 02
Normalise the entity, the service actually requested and the jurisdiction it falls under.
Reason
- 03
Resolve the party list — entity, directors, beneficial owners and connected parties.
- 04
Assemble identity, ownership, conflict and screening evidence against each named party.
- 05
Draft the letter by selecting approved clauses for that service, entity and jurisdiction.
Decide
- 06
Score file completeness and flag every unresolved match, gap and expiring document.
- 07
Route acceptance, risk rating and screening matches to the people authorised to decide them.
Out
- 08
Retain the party list, searches run, hits, clause sources and every human decision.
- 09
Issue letters and open files only inside the approval boundaries agreed during implementation.
Product statement
The agent drafts, checks and assembles inside the approval boundaries agreed during implementation; acceptance, risk rating, screening decisions and signature stay with authorised people.
Example workflow
One new client, end to end
- 1Request receivedReferral, enquiry form, CRM record or a new service for an existing client
- 2Parties resolvedEntity, directors, beneficial owners, connected parties and the service actually requested
- 3Checks assembledIdentity and ownership evidence, conflict search, sanctions and PEP screening, adverse media
- 4Letter drafted and checkedScope, fees and exclusions from the approved template, then conformance, expiry and completeness checks
No human action required
Stages 1 to 4 run without a person in the loop — nothing has been accepted or sent yet, so the lane stays empty until the gate.
Goes forward as a pack ready to approve.
Held for the risk owner or the MLRO.
Acceptance decision
The pack is held with its party list, evidence, screening hits and clause sources.
Accept · Amend · Escalate
Accepted — handed back
- 6Letter issued, then file openedAn authorised approver accepts and the letter goes out; the file opens when the signed copy returns
- 7Outcome evaluatedTemplate conformance, scope accuracy, screening precision, time to onboard and re-acceptance completeness
Amendments
Clause edits and declined acceptances are counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Accepting the client or the engagement.
- The client's money-laundering risk rating.
- Clearing or accepting a screening match.
- Conflict and independence conclusions.
Automation boundary
Agent acts unaided
- ✓Resolve the party list and gather ownership evidence.
- ✓Run conflict, sanctions and PEP searches and present the hits.
- ✓Draft the letter from approved template clauses.
- ✓Flag gaps, expiries and unresolved matches for the approver.
Letters are issued and files opened only inside the approval boundaries agreed during implementation.
- Any clause outside the approved template set.
- Fee terms, caps and liability limits.
- Signing or countersigning the letter.
- Annual re-acceptance and disengagement.
Example output
One onboarding pack, annotated
Everything the agent drafts is attached to the clause it came from.
Onboarding pack · single client
Illustrative example- Prospect
- Newly formed holding company
- Service requested
- Annual accounts and tax
- Origin
- Referral
- Drafted from
- Accounts template v4
- Completeness
- 92%
- Held for decision
- Sanctions name match
As received
Taken from the enquiry as it arrived — the agent does not infer an entity or a service.
Evidence used
- Registry ownership filing
- Approved clause library
- Screening provider hit
Why it is held
A name in the ownership chain matched a list. Only a person may clear or accept it.
What the score decides
Below the threshold the pack goes back for missing evidence instead of to an approver.
Action
AcceptAmendEscalateValue
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
All new engagements
From your intake channels
Apply the firm's own templates
Draft from the approved clause library, the firm's acceptance policy and prior terms.
Shorten the onboarding wait
Routine, lower-risk onboarding arrives assembled instead of built by hand.
Put partners on the difficult files
Screening matches, layered ownership and non-standard terms reach a decision-maker instead of every file doing so.
Build an evidence trail
Retain the party list, searches run, hits, clause sources, evaluator result and the approver's decision — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Practice management & CRM
- Karbon
- Canopy
- TaxDome
- Client and billing records
AML / KYC screening
- SmartSearch
- Amiqus
- First AML
- Sanctions and PEP lists
- Adverse media
Registries & conflicts
- Company registries
- Ownership filings
- Your client and party register
Agent
Engagement letter & onboarding
- Resolves parties
- Assembles evidence
- Drafts from templates
Documents & e-signature
- SharePoint
- iManage
- NetDocuments
- DocuSign
- Adobe Acrobat Sign
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 a signed letter
Each control wraps the one inside it. A pack clears every layer before anyone is asked to accept it.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide whether the draft may stand
- L3
- Decides who accepts and who signs
- L4 – L5
- Keep the record and the templates current
- L6
- Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the full workflow — not only the drafted letter.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the pack the approver sees
- E1Final-output evaluationDid the letter match the service the firm actually sold?
- E2Clause-source evaluationDoes every clause trace to a current approved template?
- E3Step-level evaluationWas the party list complete before the checks were run?
- E4Screening calibrationAre true matches raised and false ones kept in bounds?
- E5Slice evaluationHow does performance change across specific client cohorts?
- E6Business outcomeHow many packs were amended or sent back before acceptance?
Floor — the outcome the client pays for
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 →
01 · Retrieval
1 mode
- EL-03
Incomplete party list
Conflict and screening run before every party is known.
Stage gathers
Parties, ownership, conflicts, screening and templates
02 · Reasoning
2 modes
- EL-01
Wrong-service scope clause
Scope drafted for a service the firm did not sell.
- EL-04
Ownership taken on trust
Declared beneficial owners are never traced to a source.
Stage proposes
The clause set, the party list and the checks it ran
03 · Tool / write
1 mode
- EL-02
Letter issued before acceptance
A binding letter goes out before an approver clears the client.
Stage writes
Only where acceptance is approved and policy allows it
04 · Output
2 modes
- EL-05
Screening match pre-judged
A name hit is presented as cleared, or a real one is buried.
- EL-06
Rating without consequence
The pack rates the client high but assembles standard evidence.
Stage returns
The pack the approver and the client file see
05 · Change / Version
1 mode
- EL-07
Silent clause drift
A withdrawn template stays in use after the firm replaces it.
Stage tracks
Template, list, model and configuration changes
- Sev-1 · takes a decision reserved for a person
- Sev-2 · wrong terms reach the client
- Sev-3 · input degrades, pack routes to review
Affected slices
Onboarding risk sits in a few client structures
Sole traders on standard terms clear without incident, while layered ownership, non-resident parties and trusts account for most of the clause and screening failures. Nestack reports performance by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Layered ownership structures | 5.4% | 3.3× | Review | |
| Non-resident clients | 4.1% | 2.5× | Review | |
| Trusts and partnerships | 3.0% | 1.8× | Watch | |
| Sole traders, standard terms | 1.0% | 0.6× | Normal |
Bar: lift vs. standard-terms baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
No change ships without the risk owner
A letter's wording is contractual and its thresholds are the firm's own policy, so the loop cannot close inside the agent — every fix is re-approved first.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Amendments, sent-back packs or dismissed matches rise in a cohort.
- 02
Diagnose
Traced to the party list, a clause mapping, a screening threshold or intake.
- 03
Improve
The clause or threshold is re-approved by the risk owner and version-linked.
- 04
Verify
Re-run over past onboardings, including every pack that was amended.
- 05
Learn
The amended clause becomes the template and a standing check.
Learn → Detect
The return edge. Next year's re-acceptance runs against the clauses this cycle approved.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, clause library, agent workflow, evaluation, integration, then a supervised onboarding and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Acceptance workflow and approval-boundary definition.Week 1
- 02Screening-provider and system assessment.Week 1, Week 2
- 03Service, entity and jurisdiction scoping.Week 1, Week 2
- 04Approved template and clause-library mapping.Week 2
- 05Party resolution and evidence assembly.Week 2, Week 3
- 06Conflict, sanctions and PEP search integration.Week 3
- 07Acceptance and sign-off workflow.Week 3, Week 4
- 08Document and e-signature integration.Week 5
- 09Evaluation suite and regression cases.Week 4, Week 5
- 10Guardrails and template-conformance controls.Week 4
- 11Observability and trace 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 service line | ProductionProduction integration | AdvancedMultiple offices / entities |
|---|---|---|---|
| Introduced at Pilot | |||
| Letters drafted from your templates | ✓ | ✓ | ✓ |
| Acceptance sign-off | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Conflict and screening assembly | — | ✓ | ✓ |
| Risk-flag routing | — | ✓ | ✓ |
| Approved issue and file opening | — | ✓ | ✓ |
| Observability and evaluation | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-jurisdiction letters | — | — | ✓ |
| Multi-stage acceptance | — | — | ✓ |
| Annual re-acceptance cycles | — | — | ✓ |
| Enterprise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on services and jurisdictions in scope, screening providers, onboarding 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
01The services, entities and jurisdictions in scope
→Service, entity and jurisdiction scoping
Week 102Your acceptance policy and who approves what
→Approval-boundary definition and acceptance controls
Week 103Your approved letter templates and standard terms
→Template and clause-library mapping
Week 204Access to practice management, registries and screening
→Source-system and screening-provider assessment, then integration
Week 205Your client risk-rating criteria and escalation rules
→Risk-flag routing, guardrails and the escalation path to your MLRO
Week 306Past onboardings, including the ones that went wrong
→Evaluation suite, regression cases and failure-mode testing
Week 407Named approvers and a signing authority
→Acceptance and sign-off workflow, then a supervised onboarding
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. No letter reaches a real client before week 6, and an approver clears it before it goes.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Acceptance workflow, approval boundary and who signs what
W2Template and clause library, registry and screening access
W3Party resolution, check assembly and the drafting workflow
W4Evaluation suite, conformance guardrails and failure modes
W5Packs rebuilt from clients you onboarded last year, nothing issued
W6One live client onboarded under approval, then Agent Care handover
- Reading the band
- Bars cover only the weeks their work is named in. Week 5 rebuilds files you closed last year, not live ones.
- At the end of W6
- One live client is onboarded, accepted and signed by your people, then Agent Care monitors.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Accounting AI agent
Build an onboarding assistant around your acceptance process.
Show us how work arrives, the templates you already approve and who accepts a client. We'll run a past onboarding through it, then agree what a person has to clear before a letter goes out.
- Typical delivery
- 4–6 weeks
- Deployment
- Your practice stack
- Human review
- Before acceptance and issue
- After launch
- Agent Care available
Nestack Agents · Engagement-letter & client onboarding
AGT-ACC-11 · Agent Care available after launch