Best practices for agentic fund accounting and investment data: a field guide for investment data teams
How investment data teams move their data back office to agents, practice by practice: the state of the space, the old world and the new, what the agents do step by step, how to set autonomy, who owns what, and how to start.
Asset managers run on reconciliations, and the load grows with every fund and private asset.
Agents can reconcile continuously and trace breaks to cause, so the team ships portfolio data instead of working breaks.
The state of the space
Teams that own positions, transactions, holdings and performance data across custodians, administrators, order management and accounting books. Titles include Head of Investment Data, Investments Data Product Owner, IBOR Data Lead and Portfolio Data Manager.
Asset managers run on reconciliations between custodians, administrators and internal books. As products multiply and private assets grow, the reconciliation load grows faster than the team. Agents can run those reconciliations continuously, trace breaks to their cause and keep portfolio data products current.
The old world
A custodian restates two days; 30 funds break; operations works them by hand into the afternoon and the strike is late.
It is rarely one failure. Three patterns repeat:
- •A custodian restatement. A custodian restates two days; positions break across 30 funds.
- •Private assets out of spreadsheets. Capital calls and valuations arrive as spreadsheets from managers.
- •A PM copilot on official holdings. PMs see different exposure in the copilot and the risk report.
The new world
The restatement was detected on arrival, only the affected days reprocessed, reconciliations cleared and NAV inputs approved by 09:02.
Here is that new world as a chain of hand-offs between agents:
These are the impact metrics we move:
- •Holdings question answered: from about a day to seconds
- •Private asset template mapped: from about 3 days to hours
- •Position break traced: from about 4 hours to 8-15 min
The rest of this report walks through the 5 practices that make that possible. At a glance:
The practices, one by one
1. Reconcile continuously
End-of-day reconciliation leaves no time to fix.
Why it is harder than it looks. Custodians, administrators and internal books each have their own timing and identifiers.
The old world. Daily batch.
The new world. Quality Monitoring reconciles on every load.
What the agents do, step by step:
- 1.The Data Ingestion agent calls
get_connector_statusand confirms each custodian load. - 2.The Quality Monitoring agent calls
run_quality_checkand reconciles positions and cash. - 3.The Incident Debugging agent calls
diagnose_incidentand classifies each break.
In practice. Breaks surface at 08:11.
What to measure. Breaks open at strike time.
2. Detect restatements by fingerprint
Restated files are the top break source.
Why it is harder than it looks. Restated files look like new files unless you fingerprint them.
The old world. Reloads by hand.
The new world. Ingestion fingerprints files; only affected days reprocess.
What the agents do, step by step:
- 1.The Data Ingestion agent calls
get_connector_statusand fingerprints deliveries. - 2.The Incident Debugging agent calls
prepare_backfilland scopes affected days. - 3.The Incident Debugging agent calls
execute_backfilland reprocesses only those days.
In practice. Two days reprocessed in minutes.
What to measure. Time to clear a restatement.
Practices like these compound. Across the routine work they cover:
3. Trace breaks to cause
Working breaks one by one doesn't stop the next one.
Why it is harder than it looks. The same mapping error produces a new break every day until someone fixes the source.
The old world. Manual.
The new world. Incident Debugging traces to file, trade or mapping.
What the agents do, step by step:
- 1.The Incident Debugging agent calls
get_root_causeand finds the mapping. - 2.The Pipeline Building agent calls
validate_pipelineand tests the fix. - 3.The Autonomous Data-Conductor calls
verify_dataand confirms breaks clear.
In practice. A mapping break is fixed once.
What to measure. Repeat breaks.
4. Govern holdings for every consumer
PM tools and reports must agree.
Why it is harder than it looks. PM tools, risk and client reporting each pull holdings their own way.
The old world. Each tool has its own table.
The new world. Context Wizard serves governed holdings.
What the agents do, step by step:
- 1.The Data Context & Catalog agent calls
define_business_ruleand defines holdings and exposure. - 2.The Data Context & Catalog agent calls
mark_authoritativeand marks the source of truth. - 3.The Data Context & Catalog agent calls
trace_cross_platform_lineageand shows every consumer.
In practice. Copilot matches the book.
What to measure. Holdings discrepancies.
Autonomy where work is reversible, a named approver where it matters:
5. Hold NAV inputs for a person
NAV is accountable.
Why it is harder than it looks. NAV is the number clients and regulators hold you to.
The old world. All-or-nothing.
The new world. NAV inputs at L2.
What the agents do, step by step:
- 1.The Data Context & Catalog agent calls
blast_radius_analysisand flags NAV inputs. - 2.The Autonomous Data-Conductor calls
request_governance_reviewand holds at L2. - 3.The Data Access & Governance agent calls
generate_audit_reportand records approval.
In practice. Every change has an approver.
What to measure. NAV changes with approver.
Designing the autonomy ladder
Autonomy is set per task class, not per tool. This is where we would set each class for a mature deployment, and why:
| Task class | Closed autonomously | Escalated | Handled by a person | Where we set it | Why |
|---|---|---|---|---|---|
| Catalog and docs | 88% | 8% | 4% | L4 once the receipts are clean | The work is reversible and repeatable |
| Restatement reprocessing | 74% | 18% | 8% | L3: acts and keeps a rollback point | Most cases are routine; the rest escalate with evidence |
| Reconciliation reruns | 80% | 14% | 6% | L4 once the receipts are clean | The work is reversible and repeatable |
| Template mapping | 56% | 32% | 12% | L3: acts and keeps a rollback point | Most cases are routine; the rest escalate with evidence |
| NAV inputs | 0% | 86% | 14% | L2: always proposed to a named approver | The number carries accountability outside the data team |
The operating model: who owns what
Agents change who does the work, not who is accountable for it:
| Role | Owns | What the agents take off their plate |
|---|---|---|
| Head of Investment Data | Standards, autonomy levels per domain and which domain goes next | Assembling status, chasing owners, keeping the catalog current |
| Domain owners and approvers | NAV and performance inputs | Preparing the diff, blast radius and rollback plan for every proposal |
| Data engineers | Reviewing agent proposals and building new data products | Positions and cash are reconciled on every custodian load; restated custodian files are detected and only affected days reprocessed |
The swarm behind the practices
| Agent | What it does for you | Tools it calls |
|---|---|---|
| Data Ingestion | Custodian and administrator files, restatement detection | get_connector_status, trigger_airflow_dag, register_kafka_schema |
| Quality Monitoring | Position and cash reconciliations across books | set_sla, run_quality_check, get_anomalies, create_quality_tests_for_pipeline |
| Incident Debugging | Traces breaks to file, trade or mapping | diagnose_incident, get_root_cause, prepare_backfill, execute_backfill |
| Schema Evolution | Maps custodian and template changes | detect_schema_change, validate_schema_compatibility, generate_migration, rollback_migration |
| Data Context & Catalog | Governed holdings, exposure and performance definitions | trace_cross_platform_lineage, blast_radius_analysis, define_business_rule, generate_documentation |
| Data Access & Governance | Access by fund and client entitlement | provision_access, check_policy, request_governance_review, generate_audit_report |
Data Context Wizard keeps one governed graph of definitions, lineage and owners that every agent and person answers from. Autonomous Data-Conductor runs each job from detection to a verified result. Spellbook Data Catalog is where your team approves, audits and rolls back, and asks questions in plain English.
What disappears
- •Afternoon break-working sessions.
- •Late NAV strikes from restatements.
- •Private asset spreadsheets.
- •Copilots disagreeing with the book.
- •Repeat breaks from the same mapping.
The magic is not any single agent. It is that the back office keeps running: bad data is caught where it enters, every new table gets quality rules and documentation without anyone writing them, access follows policy, fixes are verified before anyone opens a dashboard, and every action leaves a signed receipt.
A product manager's scorecard
| Metric | How to read it |
|---|---|
| Breaks open at strike | Baseline it in the first week of the pilot; review it monthly. |
| Time to clear a restatement | Baseline it in the first week of the pilot; review it monthly. |
| Repeat breaks | Baseline it in the first week of the pilot; review it monthly. |
| Holdings discrepancies | Baseline it in the first week of the pilot; review it monthly. |
| NAV changes with approver | Baseline it in the first week of the pilot; review it monthly. |
Where this goes next
Investment operations will reconcile continuously and explain breaks before anyone asks. Data teams will deliver holdings as a governed product to every PM tool, report and copilot, and private assets will sit beside public ones on the same terms.
The climb: from assisted to autonomous to agentic
Our thesis is that the data platform climbs the way analytics did, in stages, and that the hard part of data work was never knowing what to do but doing it. For investment data teams, the climb looks like this:
- •Assisted. Agents work alongside your team. They read the whole stack, show what needs doing and propose every fix with a receipt; nAV and performance inputs always go to a named approver.
- •Autonomous, under your governance. Routine work runs on its own: positions and cash are reconciled on every custodian load, restated custodian files are detected and only affected days reprocessed, and private asset templates are landed and mapped. Your team governs the agents instead of doing the upkeep.
- •The agentic enterprise. Investment operations will reconcile continuously and explain breaks before anyone asks. Data teams will deliver holdings as a governed product to every PM tool, report and copilot, and private assets will sit beside public ones on the same terms.
You choose the altitude, every step is visible and reversible, and the pilot below is the first rung.
The pilot: how investment data teams start
The pilot is how your team proves the value on its own data before committing.
- •What we connect. Snowflake or Databricks; custodian and administrator SFTP and APIs; order management and accounting systems; Airflow; dbt; BI and PM tools, over MCP, inside your infrastructure. Nothing migrates.
- •Scope. Continuous reconciliation on one custodian at L2.
- •First week. Agents run at L1 across the scoped domain. They map every pipeline, definition and lineage path and show the fixes they would have made, with receipts, so your team sees the value before anything changes.
- •By week two. Positions and cash are reconciled on every custodian load and restated custodian files are detected and only affected days reprocessed start running on reversible work. NAV and performance inputs stay at L2 with a named approver.
- •At day 30. We review the impact metrics against your baseline:
- •Holdings question answered: from about a day to seconds
- •Private asset template mapped: from about 3 days to hours
- •Position break traced: from about 4 hours to 8-15 min
- •the share of catalog and docs closed autonomously, targeting 88% as the deployment matures
- •What you decide. Whether to expand to the next domain. The pilot fee is credited in full against the first year.
- •Cost. Start with a pilot ($7,500 one-time, credited in full against the first year). Scale is from $1,000 a month and Enterprise from $3,000 a month, billed annually, with unlimited seats, no usage meter and no markup on model spend. See pricing.
The pilot is the start of a longer curve:
Read the full playbook: Data Workers for investment data teams