BCGOrchestration, not assistance
Agentic AI orchestrates the close
BCG’s 2026 finance perspective is that agentic AI moves from task assistance to orchestrating multi-step workflows such as the accounting close.
Program 16 · Domain specialisation
SCIKIQ Certified Data and AI EngineerFinance, Accounting & Auditing
Finance is the function where an agent is a control, not a tool. If it touches anything in scope for internal controls over financial reporting, your auditor will ask how it works, who reviews it and what evidence exists. Build for that question and finance becomes the fastest domain to deploy in.
- 80 hrstaught
- 40hands-on AI labs
- 8assessment tests
- +20 hrs2 E2E capstones
>50%Of financial close tasks automatable with agentic workflows
~30%Reduction in days-to-close reported by AI-enabled finance teams
100%Transaction coverage replacing sampling in continuous controls
2026Big Four staff trained specifically to scrutinise AI-generated evidence
01
Part 1 · The business
The value chain, the people, the trends
Finance is a chain of controls as much as a chain of activities. Part 1 maps record-to-report, order-to-cash, procure-to-pay and the assurance layer over all of them.
The finance value chain
Six stages from transaction to assurance. Select one.
01
Transact & capture
Procure-to-pay, order-to-cash and the capture of every accounting event.
Decisions made here
- Is this invoice valid?
- Does it match the order and receipt?
- Which account and cost centre?
Where it breaks today
- Three-way match exceptions worked manually
- Coding decisions inconsistent across shared services
- Duplicate and fraudulent invoices caught late
What is measured
- Cost per invoice
- Touchless rate
- Days payable
Where AI landsInvoice extraction, match-exception resolution, coding suggestion and duplicate detection.
Cost per invoiceTouchless rateDays payable
02
Reconcile & substantiate
Account reconciliation, intercompany matching and balance substantiation.
Decisions made here
- Does this balance reconcile?
- What explains the difference?
- Is the supporting evidence sufficient?
Where it breaks today
- Reconciliations prepared in spreadsheets each month
- Intercompany differences chased across entities
- Evidence attached inconsistently
What is measured
- Reconciliation completion
- Ageing of open items
- Preparer effort
Where AI landsAutomated matching, difference explanation and evidence assembly with source links.
Reconciliation completionAgeing of open itemsPreparer effort
03
Close & consolidate
Period-end close, journals, accruals, consolidation and flux analysis.
Decisions made here
- Is the close complete?
- Is this variance expected?
- Does this journal need review?
Where it breaks today
- Close checklists tracked manually
- Flux explanations written from scratch monthly
- Anomalies found after the numbers went out
What is measured
- Days to close
- Post-close adjustments
- Manual journal volume
Where AI landsClose orchestration, anomaly detection, journal review triage and drafted flux commentary.
Days to closePost-close adjustmentsManual journal volume
04
Report & disclose
Statutory reporting, management reporting, disclosures and narrative.
Decisions made here
- What do the numbers say?
- What must we disclose?
- Is this consistent with last period?
Where it breaks today
- Narrative rewritten each cycle
- Consistency checked by reading
- Report assembly consuming the team
What is measured
- Reporting cycle time
- Restatements
- Review iterations
Where AI landsDrafted commentary tied to governed metrics, consistency checking and disclosure retrieval.
Reporting cycle timeRestatementsReview iterations
05
Plan & forecast
FP&A, budgeting, forecasting, scenario analysis and business partnering.
Decisions made here
- What will happen?
- What if?
- Where should we spend?
Where it breaks today
- Forecast cycles too slow to inform decisions
- Driver analysis done manually
- Business questions answered days later
What is measured
- Forecast accuracy
- Cycle time
- Decision support turnaround
Where AI landsDriver-based forecasting, scenario generation and natural-language analysis over governed data.
Forecast accuracyCycle timeDecision support turnaround
06
Control & assure
Internal controls, SOX testing, internal audit and external audit support.
Decisions made here
- Is this control operating?
- Is this an exception?
- What evidence supports this conclusion?
Where it breaks today
- Sample-based testing that misses population issues
- Evidence requests consuming the business
- Walkthroughs repeated annually
What is measured
- Control failures
- Testing effort
- Audit findings
Where AI landsContinuous control monitoring, exception triage and evidence assembly with full lineage.
Control failuresTesting effortAudit findings
Who you are building for
Five roles, and the auditor is the one who decides whether it stays. Select one to light the stages they own.
What the strategy houses are seeing
Finance is adopting quickly — and being audited on it just as quickly.
McKinseyDeployed, with oversight
Finance teams are past experimentation
Finance functions are using AI for forecasting, real-time working capital monitoring and faster reporting cycles — with human checks retained on judgement.
Industry~30% faster close
The close is the measurable win
Agentic workflows can automate more than half of close tasks and cut days-to-close by around 30%, with anomaly detection running continuously.
PCAOBAI-specific audit scrutiny
Auditors are now trained to look
Big Four firms have trained staff specifically to scrutinise AI-generated evidence, AI-touched controls and AI-driven exception resolution.
Industry100% coverage
Sampling gives way to full population
Continuous evidence across 100% of transactions lets auditors shift from sampling to reviewing exception reports — a structural change in how assurance works.
HBRSupervised assistance
Human-supervised, not agent-led
The prudent 2026 posture is agentic assistance under human review for documentation, evidence collection and analysis — not autonomous conclusion-forming.
Where these numbers come from — 4 sources
Consulted when this program was written, in July 2026. Each entry names the claim on this page that it backs, so the pairing stays checkable if the copy is later edited.
- BCGApplying Agentic AI in the Finance Function for Transformative ImpactAgentic AI orchestrating the accounting close
- McKinseyHow finance teams are putting AI to work todayForecasting, working capital monitoring, faster reporting
- IndustryWhat Your SOX Auditor Will Ask About Your AI AutomationAuditors trained to scrutinise AI-touched controls
- IndustrySOX and AI Controls: The 2026 Governance FrameworkHuman-supervised agentic assistance as the prudent posture
02
Part 2 · Data and meaning
Domains, ontology, taxonomy and governed terms
In finance the architecture question is not “can it compute the number” but “can we evidence how the number was produced”. Lineage and immutability are the load-bearing layers.
The data landscape
Six families, all of which an auditor may ask you to trace end to end.
ERP & general ledger
Journals, balances, master data and the sub-ledgers behind them.
SAP S/4HANA, Oracle, NetSuiteTransactional sub-systems
AP, AR, payroll, treasury, expenses and procurement.
Ariba, Coupa, ConcurConsolidation & reporting
Consolidation, disclosure management and management reporting.
EPM, consolidation toolsControls & audit
Control matrices, test results, findings and evidence repositories.
GRC platforms, audit toolsContracts & policy
Contracts, accounting policy, standards and internal guidance.
CLM, policy repositoriesOperational context
The non-financial data that explains a variance — volume, headcount, activity.
Operational systems, HRThe semantic layer
Finance already has a semantic layer — the chart of accounts and the close calendar. The work is making it machine-usable, so an agent quotes the number the business reports rather than one it derived itself.
Data domains
Six subject areas, each with the entities it holds, its critical data elements and the function accountable for it.
Organisation
The legal and management structures the numbers roll up through.
Legal entityCost centreProfit centreSegment
- Critical data elements
- Entity code; Cost centre code; Consolidation group
- Owned by
- Group finance
Chart of accounts
The account structure giving every posting its meaning.
AccountAccount groupStatement line
- Critical data elements
- Account code; Account type; Statement mapping
- Owned by
- Financial control
Transaction
Everything that posts, and the documents behind it.
JournalInvoicePaymentPurchase orderGoods receipt
- Critical data elements
- Document number; Posting date; Amount and currency
- Owned by
- Shared services
Close & reporting
The period-end process and what it produces.
PeriodClose taskReconciliationDisclosure
- Critical data elements
- Period identifier; Task status; Reconciling item age
- Owned by
- Financial control
Control & assurance
The control environment and the evidence it generates.
ControlTestExceptionFinding
- Critical data elements
- Control identifier; Test result; Exception severity
- Owned by
- Internal audit & SOX
Planning
Budgets, forecasts and the drivers behind them.
BudgetForecastDriverScenario
- Critical data elements
- Version identifier; Driver value; Scenario assumption
- Owned by
- FP&A
Ontology & knowledge graph
The entities and the typed relationships between them — what the agent traverses instead of guessing joins. Select any entity.
Core entityEventReference / classification
Core entity
Legal entity
A reporting unit within the group.
- Legal entity consolidates into Legal entity
- Journal posted by Legal entity
- Legal entity consolidates into Legal entity
Reference / classification
Account
A chart-of-accounts node giving a posting meaning.
- Journal posts to Account
- Reconciliation substantiates Account
- Control mitigates risk in Account
- Disclosure derived from Account
Reference / classification
Cost centre
The management unit a cost belongs to.
- Journal charged to Cost centre
Event
Journal
A posting to the general ledger.
- Journal posts to Account
- Journal posted by Legal entity
- Journal falls in Period
- Journal charged to Cost centre
- Invoice generates Journal
Event
Invoice
A billed or received demand for payment.
- Invoice generates Journal
- Invoice settled by Payment
- Purchase order commits to Invoice
Event
Purchase order
A commitment to buy.
- Purchase order commits to Invoice
Event
Payment
Settlement of an obligation.
- Invoice settled by Payment
Reference / classification
Period
The accounting period a posting falls in.
- Journal falls in Period
- Reconciliation performed for Period
- Control tested for Period
Event
Reconciliation
Substantiation of a balance at period end.
- Reconciliation substantiates Account
- Reconciliation performed for Period
Core entity
Control
A procedure mitigating a financial reporting risk.
- Control mitigates risk in Account
- Control produces Exception
- Control tested for Period
Event
Exception
A control test result outside tolerance.
- Control produces Exception
Reference / classification
Disclosure
A reported statement or note.
- Disclosure derived from Account
Taxonomies
The classification hierarchies that make records comparable across systems.
Chart of accounts
- Statement
- Account group
- Account
- Sub-account
P&L → Operating expense → Travel → Airfare
Control framework
- Process
- Risk
- Control
- Test step
Order to cash → Revenue cut-off → Manual review of period-end shipments → Sample 25
Management hierarchy
- Group
- Segment
- Entity
- Cost centre
Group → Industrial → EU Manufacturing Ltd → Plant maintenance
Governed business terms
The definitions an agent must use rather than invent. Most wrong answers in this industry are a term used loosely.
- Legal entity vs cost centre
- A legal entity reports statutorily; a cost centre reports managerially. They do not align.
- Posting date vs document date
- Posting date drives the period; document date drives the business event. Confusing them misstates cut-off.
- Reconciling item
- A difference between a balance and its supporting detail, with an age and an owner.
- Operating effectiveness
- Evidence a control worked throughout a period, not merely that it was designed.
- Cut-off
- The rule determining which period a transaction belongs to — the most common audit adjustment.
- Materiality
- The threshold above which a misstatement could influence a user of the accounts.
- Accrual
- Recognition of an expense or revenue before the cash moves.
- Days to close
- Working days from period end to reported results — the headline close metric.
03
Part 2 · The architecture
How it is actually built
The layers a deployment needs, what each one holds, and what the engineer owns there.
The reference architecture
Seven layers, built so that every figure resolves to a source.
01/07
Sources & systems of record
ERP, sub-ledgers, consolidation and evidence repositories — read through controlled paths.
Takes in
Nothing. This is the upstream edge.
Hands on
Period-locked balances, sub-ledger detail and the policy corpus, each pinned to a close period.
SystemWhat it doesKey business entities
General ledger
The book of record for the entity. Balances by account, period and dimension, closed and locked on a calendar that does not move for you.
JournalAccountBalancePeriod
Sub-ledgers
Receivables, payables, fixed assets and revenue. Detail that must reconcile to the ledger, and the place reconciliation breaks are actually caused.
InvoicePaymentAssetRevenue contract
Consolidation and close
Group structure, intercompany elimination, currency translation and the close checklist that governs the whole cycle.
EntityEliminationTranslationClose task
Procure-to-pay and expense
Requisitions, purchase orders, receipts and expense claims. The highest-volume transaction population and the usual target for testing.
Purchase orderReceiptExpense claim
Audit and controls systems
Control descriptions, test plans, evidence and findings. Where an assertion has to be supported by something a reviewer can open.
ControlTestEvidenceFinding
Policy and standards documents
Accounting policy, group manuals and the standards themselves — the corpus a treatment question is grounded on.
PolicyStandardManual
What the engineer owns here
Establishing read access that IT and audit both accept.
Where this layer fails
Reading a period that is still open. A number that moves after you have cited it is worse than no number, so pin the period and record which close you answered from.
02/07
Ingestion & immutable staging
Extracts landed immutably with timestamps, so evidence cannot be quietly rewritten.
Takes in
Read-mostly access paths, change feeds and document streams from the systems of record.
Hands on
Replayable, resolved, quality-checked records with their lineage back to the source row.
SystemWhat it doesKey business entities
Change data capture
Reads the source’s own change log rather than polling it, so the platform sees every state a record passed through instead of only where it ended up.
Change eventOperation typeCommit timestamp
Landing and raw zone
Every extract kept as it arrived, immutable and timestamped. The thing you replay from when a downstream definition turns out to have been wrong.
Raw extractBatchWatermark
Document processing
Layout-aware extraction over the unstructured half of the estate: sectioning, tables, signatures, and the page reference every later citation depends on.
DocumentSectionPage reference
Entity resolution
Decides that two records are the same real-world thing, with a survivorship rule and a confidence, so downstream joins are a decision rather than an assumption.
Match candidateSurvivorship ruleCluster
Data quality and contracts
Schema, freshness and volume expectations asserted at the boundary, so a bad load fails loudly here rather than quietly two layers later.
ContractExpectationViolation
What the engineer owns here
Proving the data the agent saw is the data that existed at that moment.
Where this layer fails
Snapshot loads instead of change capture. It looks identical in a demo and it silently loses every intermediate state, which is exactly what an audit later asks you for.
03/07
Governance & the metric layer
Chart of accounts semantics, governed metrics and entitlements by entity and role.
Takes in
Replayable, resolved records with lineage from ingestion.
Hands on
Typed, classified, defined data an agent can be grounded on without inferring meaning.
SystemWhat it doesKey business entities
Business glossary
One agreed definition per term, owned by a named person, so the number an agent quotes means what the business means by it.
TermDefinitionOwnerSteward
Ontology and knowledge graph
The typed entities and relationships the domain actually has, so an agent can traverse "which customers are exposed to this" rather than guess from adjacent text.
EntityRelationshipNodeEdge
Semantic and metrics layer
Metrics defined once, in one place, with their filters and grain. Removes the class of error where the agent computed something plausible and wrong.
MetricDimensionGrainFilter
Classification and policy
Sensitivity labels and the purpose each classification permits, applied at the field level and inherited by everything downstream.
ClassificationPurposePolicy
Lineage and catalogue
Where a value came from and what it touched on the way. The layer that makes an answer defensible rather than merely correct.
Lineage edgeDatasetColumn
What the engineer owns here
One definition of revenue, margin and headcount that an agent must use.
Where this layer fails
Letting the model infer meaning from column names. It will, it will be plausible, and nobody will notice until the number reaches a regulator or a board pack.
04/07
Retrieval & grounding
Retrieval over accounting policy, contracts, prior filings and control documentation.
Takes in
Typed, classified, defined data from the governance layer.
Hands on
Ranked, permission-trimmed, citable evidence scoped to the caller and the moment.
SystemWhat it doesKey business entities
Hybrid index
Vector and lexical retrieval together, because exact identifiers, codes and clause numbers are the thing semantic search is worst at.
ChunkEmbeddingTermScore
Chunking and metadata
Splits on the document’s own structure and carries effective dates, version and source into every chunk, so a retrieved passage knows when it was true.
ChunkEffective dateVersionSource reference
Security trimming
Applies the caller’s permissions inside the query rather than filtering results afterwards, so the model never sees what the user may not.
PrincipalEntitlementFilter
Graph retrieval
Traverses the knowledge graph for questions that are joins rather than similarity — exposure, genealogy, ownership, causation.
PathHopTraversal
Reranking and citation
Reorders candidates on relevance and returns the passage identifier behind every sentence, so the answer can be checked rather than trusted.
CandidateRankCitation
What the engineer owns here
Grounding treatment questions in the standard and the policy, not the model.
Where this layer fails
Filtering results after retrieval instead of constraining the query. The model has already seen the rows you removed, and it will use them.
05/07
Agent & orchestration
Close orchestration, reconciliation, exception triage, commentary drafting and control testing.
Takes in
Ranked, permission-trimmed, citable evidence from retrieval.
Hands on
Actions taken or proposed, each with the evidence, the identity and the trace behind it.
SystemWhat it doesKey business entities
Agent runtime
Plans, calls tools and holds the loop. Where the step budget, the timeout and the stopping condition are enforced rather than hoped for.
TaskPlanStepTrace
Tool and MCP servers
Typed, permissioned tools with declared schemas. An agent’s real capability surface is this list, which is why the list is a security artefact.
ToolSchemaScopeInvocation
State and memory
Durable task state with checkpoints, so a run that dies mid-way resumes instead of restarting and re-doing side effects.
CheckpointSessionMemory record
Human-in-the-loop gates
Approval steps on the actions that need one, carrying enough context for the approver to actually decide rather than rubber-stamp.
ApprovalThresholdDecision
Workflow integration
Writes back into the systems people already work in, under a service identity with its own audit trail.
Work itemWrite-backActor
Segregation of dutiesthis domain adds
Separate preparer and reviewer agents with disjoint tool sets, so no single identity can both propose and approve a journal.
PreparerReviewerApproval
What the engineer owns here
Segregation of duties: the agent that prepares must not be the one that approves.
Where this layer fails
One agent with both propose and approve tools. It is a control failure the moment it exists, whether or not it is ever used.
06/07
Controls, SOX & guardrails
Control design for AI-touched processes, review evidence, access control and audit trail.
Takes in
Proposed actions and drafted responses from the agent layer.
Hands on
Permitted, grounded, logged output — or a refusal with a stated reason.
SystemWhat it doesKey business entities
Policy enforcement
The rules that decide whether an action is permitted at all, evaluated before the action and independently of the model that proposed it.
PolicyDecisionObligation
Input and output screening
Injection detection on the way in, and on the way out the checks for leakage, unsupported claims and content the domain forbids.
PromptResponseViolation
Groundedness checking
Verifies each assertion resolves to retrieved evidence, and fails the response rather than shipping the sentence that does not.
AssertionSupportVerdict
Model risk management
Model inventory, intended use, validation evidence and the sign-off that lets a model be used for a purpose. Not optional in a regulated estate.
ModelIntended useValidationApproval
Audit log
Immutable record of what was asked, retrieved, decided and done — the artefact a reviewer reads when they do not take your word for it.
Audit entryActorEvidence reference
What the engineer owns here
Designing the compensating control that makes the agent auditable.
Where this layer fails
Answering from an open period. The number moves after you cite it, and every downstream working paper is now wrong.
07/07
Evaluation & monitoring
Accuracy evals, exception-rate monitoring, drift and cost per transaction.
Takes in
Permitted, grounded, logged output from the guardrail layer.
Hands on
Measured quality, cost and latency — and the evidence to change any of the three.
SystemWhat it doesKey business entities
Evaluation harness
Held-out sets and graded runs on every change, so a prompt edit is a measured change rather than a hopeful one.
Eval setCaseScoreBaseline
Regression gate
Blocks a release when quality drops, in CI, on the same evidence for everyone. The difference between a system and a demo.
ThresholdRunGate decision
Tracing
End-to-end spans across retrieval, model and tool calls, so a bad answer can be opened and read rather than argued about.
TraceSpanLatencyToken count
Cost and capacity
Cost per task and per tenant, against throughput and quota. The number that decides whether the pilot can become the rollout.
Cost per taskQuotaThroughput
Drift and feedback
Watches quality against production traffic rather than the test set, and routes real corrections back into the eval suite.
Drift signalFeedbackIncident
What the engineer owns here
Evidence that the control operated effectively throughout the period, not just today.
Where this layer fails
Evaluating once, before launch. Quality moves with the data, the model and the traffic, and a system with no live measurement has no idea which of the three moved.
Typical tech stack
Models
- Enterprise frontier models
- Extraction models for documents
- Deterministic rules where determinism is required
Grounding
- Accounting policy index
- Contract retrieval
- Prior-period comparison
Agents
- Close orchestration agents
- MCP over ERP and GRC
- Preparer/reviewer separation
Data
- Immutable timestamped staging
- Governed metric layer
- Full lineage capture
Assurance
- Control design documentation
- Continuous evidence generation
- Exception reporting
Runtime
- Entity-scoped entitlements
- Segregation of duties enforced
- Immutable audit trail
Constraints that shape the design
ICFR scope
If the agent touches a process in scope for internal controls over financial reporting, it is part of the control environment and will be tested.
Segregation of duties
An agent that both prepares and approves breaks a fundamental control. Separation must be architectural.
Determinism where required
Calculations must be deterministic and reproducible. Use the model for language and judgement support, not arithmetic.
Evidence immutability
Evidence that can be regenerated differently later is not evidence. Stage immutably and timestamp everything.
04
Part 3 · Getting it done
Where the business and the technology meet
These use cases sit deliberately on the assist-and-approve side of the line, because that is what an auditor will accept in 2026 — and because it is where the value actually is.
Use cases, mapped to the chain and the architecture
Filter by stage or by earned autonomy. Selecting a use case jumps the value chain and the architecture to the stage and layer it depends on.
Value chain stage
Autonomy earned
12 use cases
No use cases match that combination.
The delivery playbook
How a finance deployment actually runs.
01
Week 1 — establish ICFR scope
Determine whether the process is in scope for internal controls. That answer sets the documentation and review burden for everything that follows.
02
Week 2 — design the control, then the agent
Decide who reviews what and how evidence is produced before writing the agent. Retrofitting a control around a working agent never goes well.
03
Weeks 3-5 — immutable staging and lineage
Build the evidence path first. If you cannot show the auditor what data the agent saw, nothing else you build will survive.
04
Weeks 6-8 — run parallel
Run the agent alongside the existing process for a full cycle. Compare outputs and document every difference. This is your operating effectiveness evidence.
05
Weeks 9-12 — walk the auditor through it
Do the walkthrough before go-live, not after. Hand over the control documentation, runbook and the continuous evidence pipeline.
The curriculum
8 modules, 80 taught hours, 40 hands-on labs and 8 assessments — every lab provisioned and graded by the SCIKIQ Agentic AI Playground. Open a module to see its labs.
01The SCDAI delivery model in financeDeploying into a function where your system becomes part of the control environment.10 hrs · 5 labs
Delivery mandateICFR scope determinationControl design thinkingStakeholder mapValue sizingThin-slice scoping
01
ICFR triageClassify five finance processes as in or out of ICFR scope.
02
Control designDesign the review control for an AI-assisted reconciliation.
03
Value sizingQuantify close days and cost per transaction for a process.
04
Stakeholder mapMap controller, internal audit, external audit and IT.
05
Thin sliceScope a two-week slice that produces auditable evidence.
Assessment 01 — Operating model & control scope
02Finance data foundationsERP, sub-ledgers, consolidation and the evidence trail that ties them together.10 hrs · 5 labs
GL and sub-ledger structuresChart of accountsConsolidation dataEvidence repositoriesLineageEntity entitlements
01
GL extractionExtract journals and balances with complete lineage metadata.
02
CoA semanticsModel a chart of accounts so an agent can reason over it.
03
Immutable stagingBuild timestamped, immutable staging that reconciles to source.
04
Sub-ledger tie-outReconcile a sub-ledger to the GL programmatically.
05
Entity entitlementsScope agent access by legal entity and role.
Assessment 02 — Finance data landscape
03Context engineering for financial reportingNumbers must come from the metric layer; the model supplies language and judgement support only.10 hrs · 5 labs
Metric-grounded generationDeterministic calculationStructured outputsAccounting policy groundingRefusalContext budget
01
Metric groundingForce every figure in generated commentary to come from the metric layer.
02
Deterministic splitSeparate calculation from narration and prove the split.
03
Policy groundingAnswer a treatment question citing the standard and internal policy.
04
RefusalMake the agent refuse to opine where judgement is required.
05
Context budgetReduce context cost on a high-volume reconciliation agent.
Assessment 03 — Context engineering
04Grounding on policy, contracts and prior periodsThe three corpora that decide whether a finance answer is defensible.10 hrs · 5 labs
Accounting policy retrievalContract retrievalPrior-period comparisonStandards updatesCitationsRetrieval evals
01
Policy indexIndex accounting policy and standards with version awareness.
02
Contract retrievalExtract revenue-relevant terms from contracts with citations.
03
Prior-period diffCompare current disclosures to prior periods and surface changes.
04
Citation contractAttach a resolvable source to every assertion.
05
Retrieval evalsMeasure recall on labelled accounting treatment questions.
Assessment 04 — Retrieval & grounding
05Agent design under segregation of dutiesDesigning agents that prepare and propose but architecturally cannot approve.10 hrs · 5 labs
Workflow vs agentPreparer/reviewer separationClose orchestrationException triageMulti-agent finance flowsContainment
01
Reconciliation agentMatch transactions and explain residual differences with evidence.
02
SoD enforcementProve the preparing agent cannot approve its own work.
03
Close orchestratorTrack close tasks and dependencies and report blockers.
04
Exception triageScore and route exceptions with explainable risk reasoning.
05
ContainmentCap posting authority and prove the cap holds.
Assessment 05 — Agent design & segregation of duties
06Integrating with ERP and GRCSAP, Oracle and GRC platforms — where a write is a journal and a journal is evidence.10 hrs · 5 labs
MCP over ERPPosting safetyGRC integrationIdentityPeriod-end loadError handling
01
MCP over ERPExpose GL query tools with entity-scoped typed schemas.
02
Posting safetyDesign a journal-proposal tool that cannot post without approval.
03
GRC integrationWrite control test results into a GRC platform with evidence.
04
Period-end loadHandle concurrency and load during close without failures.
05
Error handlingEnsure a retry never creates a duplicate journal.
Assessment 06 — Systems integration
07SOX, audit and securityAnswering the questions a SOX auditor will actually ask about your agent.10 hrs · 5 labs
AI in the control environmentOperating effectiveness evidenceWalkthrough preparationAccess controlOWASP LLM Top 10Fraud risk
01
Control documentationDocument an AI-touched control to audit standard.
02
Operating effectivenessProduce evidence the control operated throughout a period.
03
WalkthroughRun a mock auditor walkthrough of your agent and fix the gaps.
04
Injection defenceDefend an invoice pipeline against instructions in a submitted document.
05
Fraud scenariosTest whether the agent can be manipulated into approving a bad payment.
Assessment 07 — SOX, audit & security
08Production, evidence and costContinuous evidence, exception volume management and the economics of finance automation.10 hrs · 5 labs
Eval suitesContinuous evidenceException volumeDriftCost per transactionHandover
01
Eval gateGate deployment on matching and extraction accuracy regression.
02
Evidence pipelineGenerate control evidence continuously rather than at period end.
03
Exception tuningTune thresholds so exception volume is actually reviewable.
04
Cost per transactionMeasure and halve cost per processed transaction.
05
HandoverProduce runbook, control documentation and owner assignment.
Assessment 08 — Production & operations
Two capstones — 20 hours
End-to-end agent design, development, deployment and testing, 10 hours each. Reviewed by a Senior SCDAI Engineer against a published rubric.
Capstone A — Reconciliation and close agent
Build an agent that reconciles an account, explains residual differences with attached evidence and reports close status — with preparer/reviewer separation enforced architecturally.
Capstone B — Continuous control monitoring agent
Build continuous testing of a control across the full transaction population, producing an exception report and an evidence trail an external auditor could rely on.
What you deliver, and how it is marked
- A running agent, deployed on the program's target stack, reachable by a reviewer.
- An architecture note: boundaries, tools, autonomy level and the trade-offs you took.
- A versioned eval set with a baseline and your final scores.
- A trace walkthrough of one success and one deliberate failure.
- Cost per task, measured — not estimated.
- A runbook and rollback path a client team could operate on Monday.
Problem fit15%Is this a decision worth automating, scoped to a defensible boundary?
Architecture20%Right pattern for the constraints; autonomy earned, not assumed.
Grounding & accuracy20%Measured groundedness against the eval set, with the failures named.
Safety & compliance20%Domain controls actually enforced, and demonstrably so.
Operability15%Traces, runbook, rollback, cost — could someone else run this?
Communication10%Can a client executive follow the reasoning in ten minutes?
One lab, start to finish
Every lab in this program follows the shape below. This is lab 05 in full — the brief you are given, the environment that is provisioned for you, the code you start from and the assertions that decide whether you passed.
Module 05 · Agent design under segregation of duties
Segregation of duties in code
Make it architecturally impossible for the preparing agent to approve its own work.
The brief
Build a reconciliation agent that matches transactions, explains residual differences and proposes a journal. A separate reviewer path approves. Then demonstrate to a mock auditor that the preparer cannot approve, by capability rather than by configuration.
Provisioned for you
- GL and sub-ledger extracts, 50k lines, 12 seeded breaks
- Immutable timestamped staging
- Two scoped service identities
- Mock auditor walkthrough script
You start from this
# The preparer holds propose_journal. It does not hold post_journal.
PREPARER_TOOLS = [read_gl, read_subledger, propose_journal]
REVIEWER_TOOLS = [read_proposal, approve_journal]
def reconcile(account, period):
raise NotImplementedErrorWhat the grader asserts
- Preparer identity cannot invoke post_journal under any prompt
- All 12 seeded breaks are found and explained with attached evidence
- Every explanation is verifiable against staged data, not plausible prose
- Retry of a proposal does not create a duplicate
- Evidence trail reconstructs exactly what data the agent saw and when
05
Assessment
Assessed, not attended
Every module ends with a timed, randomised assessment delivered through the SCIKIQ Agentic AI Playground. The certificate requires a pass on all of them plus two reviewed capstones.
Assessment blueprint
What each test covers, how long it runs, and how many items are currently in the versioned bank behind it.
| # | Assessment & coverage | Items | Time | In bank |
|---|---|---|---|---|
| 01 | Operating model & control scopeDelivery mandateICFR scopeControl designValue sizing | 25 | 35 min | 4 |
| 02 | Finance data landscapeGL and sub-ledgersChart of accountsImmutable stagingEntitlements | 25 | 35 min | 4 |
| 03 | Context engineeringMetric groundingDeterminismPolicy groundingRefusal | 25 | 35 min | 4 |
| 04 | Retrieval & groundingPolicy retrievalContract retrievalPrior-period comparisonRetrieval metrics | 25 | 40 min | 4 |
| 05 | Agent design & segregation of dutiesPreparer/reviewer separationClose orchestrationException triageContainment | 25 | 40 min | 4 |
| 06 | Systems integrationMCP over ERPPosting safetyGRC integrationResilience | 25 | 35 min | 4 |
| 07 | SOX, audit & securityAI in the control environmentEvidenceWalkthroughsLLM security | 30 | 45 min | 4 |
| 08 | Production & operationsEval gatingContinuous evidenceException managementEconomics and handover | 30 | 45 min | 4 |
Single best answerOne defensible option among plausible distractors drawn from real field mistakes.
Scenario judgementA client situation with constraints; you pick the action a Senior SCDAI Engineer would take.
Architecture selectionGiven non-functional requirements, choose the pattern and justify the trade-off.
Failure diagnosisA trace, an eval report or a cost curve — identify the root cause.
- Each module ends with a timed assessment; the program certificate requires a pass on every one.
- Items are drawn at random from a versioned bank, so no two learners sit an identical paper.
- Pass mark is 70%. Two retakes are included, each with a fresh draw.
- Scenario items carry double weight — they are the ones that predict field performance.
- Capstones are reviewed against a published rubric by a Senior SCDAI Engineer, not auto-graded.
Try four questions from the bank
Real items, drawn from 32 in this program's bank — weighted toward the scenario and diagnosis types, because those are the ones that predict field performance. Instant feedback, nothing saved.
Four sample items — one attempt each, then the reasoning is shown.
Q1Why is ICFR scope the first question on a finance engagement?
In-scope means documentation, review evidence and auditor walkthroughs. Discovering that late is expensive.
Q2Why must staging be immutable and timestamped?
Evidence that can be regenerated differently later is not evidence. Immutability is what makes it auditable.
Q3Where should a figure in generated commentary come from?
Model arithmetic is unnecessary risk. Separating calculation from narration is the core discipline here.
Q4Why must accounting policy retrieval be version-aware?
Standards change and periods differ. Version-aware retrieval is the same discipline as wording versioning in insurance.
06
Career & outcomes
What this makes you, and where it goes next
The same ladder whichever specialisation you enter through — what changes is the domain you go deep in. Below: how the program is delivered, the skills it moves, the roles it leads to, and the specialisations closest to this one.
Available two ways
The same labs, assessments and capstones, delivered to an enterprise cohort or to individual professionals.
B2B
Direct to enterprise
Cohorts of 20 to 2,000+ on your own tenancy, with your data patterns and your cloud. Skill-gap baselining up front, per-team mastery reporting throughout, and capstones scoped against your real backlog so the output is deployable work.
- Private cohort on your tenancy
- Baseline skills assessment before kick-off
- Capstones scoped to your backlog
- Per-team mastery dashboards for L&D
- Optional Senior SCDAI Engineer architecture reviews
B2C
Direct to individual professionals
The same labs, assessments and capstones for individual engineers and analysts, run on shared infrastructure with a fixed cohort calendar. You leave with a graded portfolio, not a certificate of attendance.
- Fixed-date cohorts, self-paced within them
- Same graded labs and item bank
- Portfolio-ready capstones you own
- Certificate on a pass in every module
- Community and office-hours access
Skill matrix
Find your row and aim one column right. The Playground scores you against this after every module.
| Skill | Beginner | Intermediate | Advanced |
|---|---|---|---|
| Finance fluency | Knows the cycles. | Maps the chain to controls and owners. | Sizes value and designs the control change. |
| Control engineering | Aware of SOX. | Designs review controls around agents. | Produces audit-ready operating effectiveness evidence. |
| Evidence & lineage | Logs outputs. | Stages immutably with lineage. | Designs an evidence path that survives external audit. |
| Context engineering | Writes clear prompts. | Structures retrieval, tools and state deliberately. | Designs context strategy for reliability and cost at scale. |
| Retrieval & grounding | Builds basic vector search. | Tunes chunking, hybrid search and reranking. | Designs graph + vector grounding with measured recall. |
| Agent orchestration | Runs a single tool-calling agent. | Builds supervised multi-step and multi-agent flows. | Designs autonomy boundaries and failure containment. |
| Tool & system integration | Calls a documented API. | Writes an MCP server over a system of record. | Designs a least-privilege tool estate across systems. |
| Evaluation | Eyeballs outputs. | Builds labelled eval sets and regression gates. | Runs online evals with drift and judge calibration. |
| Observability & cost | Reads logs. | Traces runs, tracks tokens and latency. | Owns cost per task and capacity planning in production. |
| Security & guardrails | Adds output filters. | Mitigates the OWASP LLM Top 10 in a build. | Threat-models an agent estate and proves controls. |
| Client delivery | Takes notes in a workshop. | Runs discovery and scopes a thin slice. | Owns the account technically, from scope to handover. |
Where it takes you
The SCDAI ladder is the same whichever specialisation you enter through — what changes is the domain you go deep in.
01
Data and AI EngineerEntry
Python, LLM APIs, first RAG build, prompt and context basics.
02
Associate Data and AI EngineerPractitioner
Client-facing discovery, tool integration, agent build under supervision.
03
SCIKIQ Certified Data and AI EngineerSpecialist
Owns a deployment end to end — scoping, build, evals, handover.
04
Senior Data and AI Engineer / Domain LeadSenior
Multi-agent architecture, regulated deployments, reference patterns for the practice.
05
AI Architect / Practice LeadLeadership
Portfolio of accounts, delivery standards, hiring and capability strategy.
Terms worth knowing
- ICFR
- Internal Controls over Financial Reporting — the control scope SOX tests.
- SOX
- Sarbanes-Oxley — the US legislation driving financial control requirements.
- R2R / O2C / P2P
- Record-to-report, order-to-cash and procure-to-pay — the three core finance cycles.
- Segregation of duties
- The control principle that preparation and approval sit with different parties.
- Flux analysis
- Variance analysis explaining movement between periods.
- Three-way match
- Matching invoice, purchase order and goods receipt before payment.
- Operating effectiveness
- Evidence that a control worked throughout the period, not just at a point in time.
- Continuous control monitoring
- Testing controls against the full population continuously rather than by sample.
- SCDAI
- SCIKIQ Certified Data and AI Engineer — the credential this program awards.
- SCDAI engineer
- An engineer embedded with the customer who turns their constraints into a working system.
- Thin slice
- The narrowest end-to-end path through a workflow that still produces business value.
- Agentic RAG
- Retrieval where the agent decides when, what and how often to retrieve.
- MCP
- Model Context Protocol — open standard connecting an agent to tools and data.
- A2A
- Agent2Agent — protocol for agent-to-agent delegation across vendors.
- Eval set
- A labelled, versioned set of cases an agent must pass before release.
- LLM-as-judge
- Using a model to score outputs, calibrated periodically against human ratings.
- Guardrail
- An enforced constraint on an input, output or action — not a prompt request.
- Human-in-the-loop
- A required human approval step before a consequential action executes.
- Cost per task
- Total token, tool and compute cost of one completed unit of work — the engineer's unit economic.
- Golden path
- The reference architecture a practice standardises on so deployments stay reviewable.
- Handover
- The point at which the client can operate, evaluate and change the system without the engineer.
Related programs
Skill a team, or join a cohort
B2B cohorts run on your tenancy with capstones scoped to your backlog. B2C cohorts run on a fixed calendar.
07
Enrolling
What you need, what you get, and what it costs
Stated plainly enough to rule yourself in or out without a sales call: the prerequisites, how the program runs, exactly what the credential is worth, and the questions everyone asks.
Before you start
Stated plainly so you can rule yourself in or out without a sales call. Nothing here is a formal qualification — it is what the first lab assumes you can already do.
Technical
You should already be able to do these
- Comfortable in Python: functions, typing, virtual environments, reading a stack trace
- Have called an LLM API and handled a structured response
- Git, the command line, and reading someone else’s code without panic
Domain
What we assume, and what we teach
- Know what a journal, a ledger and a reconciliation are
- No accounting qualification needed; controls thinking is taught
Time and access
What the program asks of your week
- A machine that can run a browser and a terminal — labs run in the Playground, not locally
- Roughly 8 hours a week for the taught weeks, plus 20 hours for the two capstones
How it runs
Cohort dates and pricing are confirmed on enquiry rather than printed here, because both move with the intake.
- Format
- Self-paced within a fixed cohort window, delivered in the SCIKIQ Agentic AI Playground
- Taught hours
- 80 hours across 8 modules
- Capstones
- 20 hours — two end-to-end builds, reviewed not auto-graded
- Typical duration
- 10–12 weeks at around 8 hours a week
- Level
- Intermediate → Advanced
- Built for
- Engineers and analysts deploying AI in finance, accounting and audit functions
- Next cohorts
- — confirmed on enquiry
- Pricing
- — confirmed on enquiry
The credential
The credential is awarded per specialisation, so it names the domain or stack you were assessed in rather than claiming general competence. On this program the badge reads SCDAI — Finance, Accounting & Auditing.
How it is earned
- A pass at 70% or above on every module assessment — not an average across them
- Both capstones reviewed against the published rubric by a Senior SCDAI Engineer
- All labs submitted and machine-graded green
- Awarded as
- A verifiable digital badge with a public credential record
- Valid for
- Two years from award
- Recertification
- One current-year assessment plus a refreshed capstone
- Retakes
- Two included per assessment, each with a fresh item draw
- Named specialisation
- The badge states the domain or stack, e.g. SCDAI — Healthcare
- Verification
- Third parties confirm a holder from the credential record, no login required
A certificate that cannot be checked is decoration. Every award resolves to a record showing the specialisation, the award date and the assessments passed.
The certificate you leave with
This is the credential the program awards, shown exactly as it is issued — with the specialisation named, the assessment record attached and a verification link anyone can check without an account.
SCIKIQ Data Academy
Certificate of Professional Certification
This is to certify that
Your name
has been assessed and certified as
SCIKIQ Certified Data and AI Engineer
Finance, Accounting & Auditing
- Specialisation
- Finance, Accounting & Auditing
- Assessments passed
- 8 of 8, each at 70% or above
- Labs graded
- 40 of 40 machine-graded green
- Capstones reviewed
- 2, against the published rubric
- Taught hours
- 80 across 8 modules, plus 20 capstone hours
AwardedOn completion
Valid untilTwo years from award
Credential IDSCDAI-FINANCE-ACCO-<issued>
Senior SCDAI Engineer, Certification Board
- 01The specialisation is on the certificate
Not “Data and AI Engineer” but the domain or stack you were actually assessed in. A general claim would be a weaker one.
- 02The assessment record travels with it
Modules passed, labs graded and both capstones reviewed — so the credential states what was measured rather than that you attended.
- 03Anyone can verify it, without a login
The credential ID resolves to a public record showing the specialisation, the award date and the assessments passed.
Share it on LinkedIn
Add it to your LinkedIn profile in one step. The link pre-fills the certification fields from the credential record, so the entry on your profile matches the record a reader can check.
Add to LinkedIn profile The button is live on your real certificate; here it opens LinkedIn pre-filled with this specialisation so you can see exactly what the profile entry will say.A certificate that cannot be checked is decoration. Every award resolves to a record showing the specialisation, the award date and the assessments passed.
Questions we get asked
The objections that come up in every conversation about this program, answered without the brochure voice.
Is this a course or a certification?
Both, and the second is the point. Eight timed assessments and two reviewed capstones stand between you and the credential, so a pass means someone measured the skill rather than recorded your attendance.
How is it self-paced without falling apart?
Every lab is provisioned, graded and unblocked by the Playground rather than by an instructor. That is what lets a cohort of 2,000 cost the same faculty time as a cohort of 20 — and why you are never waiting on someone to mark your work.
What if I fail an assessment?
Two retakes are included per assessment, each drawing a fresh item set from the bank, so a retake is a genuinely new paper rather than the same questions again.
Do I need my own cloud account?
For the domain specialisations, no — labs run in provisioned sandboxes. For the four tech-stack programs you will want access to that platform, since deploying into a real subscription is much of the point.
Can we run this privately for our team?
Yes. B2B cohorts run on your own tenancy with your data patterns, a skills baseline before kick-off, per-team mastery reporting, and capstones scoped against your actual backlog so the output is deployable work rather than an exercise.
Which specialisation should I take?
Take the domain you deploy into. If you move across industries, take a tech-stack program instead and pick up domain context on the engagement. The chooser on the programs page will narrow it.
How current is the material?
The trends, platform capabilities and regulatory positions are reviewed each quarter, and every external claim on these pages links to its source so you can check the date yourself.
What do I actually leave with?
A graded portfolio: forty machine-graded labs, two reviewed end-to-end agent builds with measured evaluation and cost per task, and a verifiable credential naming your specialisation.
Still deciding?
Tell us the systems you deploy into and we will say plainly whether this specialisation is the right one — or which of the 21 is.