==================================================
 STORYGEN AI — STORY EXPORT
==================================================
Story ID:           30
Title:              Test2_AF_v2 (hybrid)
Owner:              Ang Chwee Geok Karen - Digital Innovation Consultant <karen.ang@educlaas.com>
App Type:           Agentic AI App
Input Type:         file
Status:             features_generated
Source File:        PIL_function_spec_procodevslowcode_nocode_29Apr2026.pdf
Created:            2026-04-29 13:52:02 UTC
Updated:            2026-04-29 14:12:21 UTC
Features Generated: 2026-04-29 14:12:21 UTC
Total Clusters:     8
Total Features:     16

--- ORIGINAL INPUT ---
Index
PIL's Agentic Project 
Scope
PIL Original Problem Statement
Feature Cluster 
(Master Catalogue)
Feature Name 
(Master Catalogue)
Description Actors/Users Expected Outcome
Human-in-the-loop (role & 
action)
Clarification Needed
Recommendation / Next 
Action
MoSCoW 
Priority
Potential 
MVP
6-12 Month 
Horizon Tag
Recommended 
Approach
Why
Comments (Angela)
1 Automated 
Workflow for 
Submission of AFS to 
GFRA/CAC
The current AFS submission 
process lacks a standardized 
workflow and centralized tracking, 
resulting in limited visibility, 
unclear accountability, and 
increased risk of late or missed 
submissions. Manual coordination 
and follow-ups also weaken 
auditability and compliance 
oversight.
Financial Close & 
Reporting
AFS Submission 
Agent
Standardises and tracks AFS 
submission end-to-end by 
capturing submission triggers, 
validating required packs, 
routing reviews, monitoring 
status and logging approvals 
through a governed workflow.  
Financial Reporting 
team, CAC, 
Controllership.  
AFS submissions become 
visible, standardised and 
traceable, reducing late or 
incomplete submissions and 
improving accountability.  
CAC / controllership reviewer 
approves submissions and 
resolves late or incomplete 
cases.
Confirm whether scope is 
only 
workflow/orchestration or 
also includes intelligence 
checks on completeness, 
anomalies, and submission 
quality.
Keep as Financial Close & 
Reporting. Position as 
workflow-led with 
optional intelligence 
checks for completeness, 
lateness risk, and reviewer 
prioritisation.
Must have Yes 0-6 months / 
Wave 1
Low code / no code 
first
Workflow-centric: task routing, 
reminders, submission tracking, 
approvals, escalations, and status 
visibility are strong fits for Power 
Automate, while Copilot/virtual 
agent can guide users on submission 
steps and FAQs, and Power BI can 
provide submission dashboards.
Referring to the column of Expected Outcome, 
1. Visible - does it mean to control the timeliness of the submission? What about 
version controls or revision that were update after first submission? 
2. Standardisation - what is the definition of standardisation - standardise in term of 
the definition of the account mapping, standardise format etc? How are these AFS 
prepared, from the Oracle GFS and CAC? What format?
3. Incomplete - which are the field that will define incomplete? 
4. Does it also require Quick Dashboard numbers consolidated to provide quick review 
and variance to enhance accuracy of submission?
2 Automated 
Prepayments 
Computations
Prepayments calculations are 
currently performed manually. 
There is high risk of formula 
errors, incorrect amortisation 
periods, and inconsistent 
assumptions. The manual process 
creates key-person dependency 
and a weak audit trail.
Financial Close & 
Reporting
Prepayments 
Computation Agent
Automates monthly 
prepayment amortisation and 
recognition by classifying 
prepayment items, computing 
schedules by policy, preparing 
journal proposals and routing 
exceptions for review.  
GL / Record-to-
Report 
accountants.  
Prepayment calculations 
become faster, more 
consistent and auditable, 
with lower spreadsheet and 
formula risk.  
R2R accountant reviews 
exceptions and approves 
prepayment journals before 
posting.
Clarify prepaid asset types 
in scope (e.g. subscriptions, 
insurance, rent), required 
validation rules, whether a 
workbench is needed, and 
whether Oracle GPS/GFS 
already supports 
amortisation schedules.
Keep as Financial Close & 
Reporting for PIL, but 
explicitly define policy 
rules, schedule logic, ERP 
touchpoints, and 
exception handling. Add 
ERP-native capability 
check.
Must have Yes 0-6 months / 
Wave 1
Pro-code core + low-
code wrapper
Prepayment schedules, amortisation 
logic, exception handling, and 
accounting-rule consistency are 
calculation-heavy and audit-sensitive, so 
the computation engine should be 
engineered in pro-code; Power 
Automate can trigger runs and 
approvals, Copilot can explain 
outcomes, and Power BI can expose 
schedules and exceptions. 
Need to clarify:
1. On vendor invoices, how is the period to amortise is captured? When is the start 
date? 
2. Are there fields to capture the classifications of such prepaid expense? 
3. Are all amortisation straight line or other methods eg reducing balance etc?
4. What are the exceptions that will arise that need to route for review?
5. Amortised expenses are offset against invoice in the prepayment account to reflect 
the outstanding yet to amortise prepayment for audit?
3 Automated GFS AP 
Payment Review
The AP payments review process 
is heavily manual and sample-
based, requiring the team to 
manually open each invoice to 
verify bank details. This limits 
coverage and increases the risk of 
errors or missed issues before 
execution.
AP Controls & 
Payments
AP Payment Review 
Agent
Reviews GFS AP payment 
proposals before execution by 
extracting vendor and bank 
details, comparing critical 
fields, flagging mismatches and 
prioritising exceptions for 
human review without 
executing payments directly.  
AP team, payment 
control reviewers.  
Payment review coverage 
expands beyond manual 
sampling, reducing bank-
detail error risk and 
improving pre-payment 
control traceability.  
AP payment controller 
validates flagged payments and 
confirms release, hold, or 
follow-up.
Confirm critical payment-
risk scenarios, what fields 
must be matched, how 
split payments are handled, 
and which exception types 
require reviewer decision.
Keep under AP Controls & 
Payments. Narrow MVP to 
pre-payment review of 
vendor/bank detail 
mismatches and high-risk 
exceptions, then expand 
to more scenarios later.
Must have Yes 0-6 months / 
Wave 1
Hybrid, leaning pro-
code
Bank/vendor master validation, 
duplicate detection, rule scoring, or risk 
checks before payment, the control 
logic should be pro-code; Power 
Automate is suitable for review 
workflow and evidence capture, 
while Power BI supports exception/risk 
reporting and Copilot supports reviewer 
Q&A.
Need to clarify:
1. Where are the bank details of vendors captured? Are there different bank accounts 
for a single vendor?
2. What are the critical fields that they are referring to?
3. What are the criteria on the prioritising requirement to consider?
4 AP Payments 
Dashboard
AP payment information is 
currently spread across multiple 
systems and reports, with limited 
consolidated visibility across 
entities. Finance relies on manual 
extraction and tracking to 
determine payments made to 
date and outstanding payables.
AP Controls & 
Payments
Payments Visibility 
Agent
Aggregates AP status feeds 
and payment files to provide a 
consolidated view of paid, 
outstanding, overdue and 
upcoming AP positions across 
entities through dashboards, 
alerts and drill-through views.  
AP finance leads, 
entity finance 
managers, 
Treasury.  
Finance gains a single, timely 
view of AP liabilities and 
payment status, improving 
monitoring, reporting and 
cash-planning decisions.  
AP lead or treasury reviewer 
investigates overdue or 
exception items surfaced in the 
dashboard.
Clarify whether MVP is 
dashboard-only or includes 
alerts and case 
management. Confirm 
source systems, frequency, 
entity coverage, and 
required due-date / 
overdue logic.
Keep under AP Controls & 
Payments. Start with one-
entity sample and a 
dashboard-led MVP, then 
scale with pro-code 
ingestion for multi-source 
/ multi-entity rollout.
Must have Yes 0-6 months / 
Wave 1
Low code / no code 
if data is already 
accessible; 
otherwise hybrid
If source systems can already feed a 
usable data model, Power BI is ideal for 
payment status visibility and Power 
Automate can refresh/alert; but if data 
must be stitched from multiple 
fragmented systems with complex 
integration and reconciliation rules, the 
integration layer should be pro-code. 
No comments, looks straight forward. So show them the Cluster view
5 Automated Lease 
Schedule 
Computation
Lease schedule computations are 
highly manual and spreadsheet-
driven, requiring significant effort 
to extract contract details, 
compute lease liabilities and right-
of-use assets, and maintain 
schedules over time.
Financial Close & 
Reporting
Lease Accounting 
Agent
Computes lease liabilities, ROU 
assets and schedules by 
ingesting lease terms, 
identifying amendments, 
applying policy rules and 
generating postings for review 
through a lease accounting 
cockpit.  
Lease accounting 
team, 
Controllership.  
Lease schedules and 
postings are produced more 
accurately and consistently, 
reducing manual effort and 
reporting risk.  
Lease accountant reviews 
schedule exceptions, 
amendments, and posting 
proposals before submission.
Confirm whether PIL 
already has a lease engine 
or if this is entirely 
spreadsheet-based. Clarify 
amendment handling, 
discount-rate assumptions, 
and posting integration.
Keep as Financial Close & 
Reporting. Treat as a 
schedule-computation 
and controlled-posting 
capability similar to 
prepayments, but with 
lease-specific policy logic.
Should have Yes 6-12 months / 
Wave 2
1. How are amendments tracked in the system? Does it provide update prompt from 
Oracle?
2. What are the policy rules? Are there different criteria for different rules?
6 Capital Accounting 
Dashboard
Capital accounting data is highly 
fragmented across sources. 
Significant manual effort is 
required to extract, validate, and 
consolidate data to obtain a 
complete view of capital assets, 
resulting in slow reporting and 
higher inconsistency risk.
Financial Close & 
Reporting
Capital Accounting 
Insight Agent
Consolidates asset, CIP and 
project data, performs 
integrity checks and presents 
review-ready capital views and 
exception queues for 
controllers.  
Capital accounting 
team, project 
controllers, 
finance managers.  
Capital accounting obtains a 
more complete and reliable 
view of balances and 
movements with less 
manual consolidation.  
Capital accountant or project 
controller reviews integrity 
breaks and resolves asset / CIP 
issues.
Clarify whether value is 
mainly dashboard visibility, 
data integrity checks, or 
journal / schedule support. 
Confirm sources for assets, 
CIP, and projects.
Keep as Financial Close & 
Reporting. Position as a 
review and exception 
dashboard first, with 
reusable validation 
patterns from 
prepayments / close 
review.
Should have Yes 6-12 months / 
Wave 2
1. How many systems sources that these data points from?
2. What is required to validate?
3. CIP (construction in progress) and Project - are these in Percentage of completion or 
build-up cost? Does it needs to compare to budget?
7 Automated Monthly 
SOAs to Customers 
Monthly SOAs are manually 
downloaded for each customer 
and emailed individually. This 
creates high manual effort during 
month-end closing and risk of 
incorrect customer selection or 
missed / delayed SOA delivery.
AR & Cash 
Application
SOA Dispatch 
Agent
Automates the generation, 
validation, dispatch and 
delivery tracking of monthly 
Statements of Account to 
customers through approved 
channels.  
AR team, 
customer service 
finance support.  
SOAs are delivered more 
accurately and on time, 
reducing manual month-end 
effort and communication 
errors.  
AR or customer finance 
support handles failed sends, 
invalid contacts, and resend 
decisions.
Confirm channel, schedule, 
customer-contact 
validation rules, bounce / 
failure handling, and 
whether approvals are 
required before send.
Keep under AR & Cash 
Application. Position as a 
workflow automation 
candidate with relatively 
fast time-to-value.
Must have Yes 0-6 months / 
Wave 1
Low code / no code 
first
Scheduled extraction, document 
dispatch, email delivery, status logging, 
and follow-up reminders are 
classic Power Automate use 
cases; Copilot can support 
customer/self-service queries, 
and Power BI can track dispatch 
timeliness and failure rates
No comments, looks straight forward. So show them the Cluster view
8 Automated Bank 
receipt classification 
(Trade vs Non-Trade)  
 
Bank statements are downloaded 
regularly and reviewed manually. 
Credit transactions require manual 
filtering and AR vs non-AR 
determination, creating high effort 
and misclassification risk.
AR & Cash 
Application
Bank Receipt 
Classification Agent
Ingests bank statements, 
classifies incoming credits as 
trade or non-trade using rules 
and remittance patterns, and 
routes ambiguous items for 
review before cash-application 
handover.  
Cash application 
team, AR, GL team.  
Receipt classification 
becomes faster and more 
consistent, reducing delays 
and misclassification risk in 
AR and GL.  
AR / GL reviewer confirms 
ambiguous receipts before 
cash-application handover.
Clarify classification rules 
for trade vs non-trade, 
handoff to GL, tolerance for 
ambiguous items, and 
whether bank recon 
overlap exists.
Keep under AR & Cash 
Application. Position as a 
classification and review 
queue that can later 
extend into Advanced 
Cash Application.
Should have Yes 6-12 months / 
Wave 2
1. Does Oracle perform the bank reconciliation? So the starting is from bank 
reconciliation or bank sttmt?
2. Where and how does the data on Trade and non-Trade receivables defined? 
3. to clarify remittance patterns

9Automated GFS data 
reconciliation for Tax 
Reporting
Manual adjustments and 
classifications of GFS output are 
required for tax reporting. This 
creates risk of delayed 
identification of errors, 
misclassifications, and operational 
inefficiencies, with increased 
scrutiny expected from IRAS.
Tax & Compliance Tax Reconciliation 
Agent
Pulls GFS outputs, applies tax 
mapping logic, identifies 
variances and unsupported 
adjustments, produces 
reconciliation evidence and 
routes unresolved breaks to 
tax reviewers.  
Group Tax, finance 
reporting, 
Singapore tax 
compliance team.  
Tax reporting becomes more 
standardised, auditable and 
timely, with lower 
misclassification and 
compliance risk.  
Group Tax reviewer 
investigates breaks and signs 
off reconciled outputs and 
evidence.
Confirm tax mapping logic, 
data sources, adjustment 
categories, reviewer 
workflow, and audit-
evidence requirements for 
Singapore entities.
Keep under Tax & 
Compliance. Position as a 
high-value reconciliation 
and evidence-generation 
use case with strong 
compliance justification.
Should have Yes 6-12 months / 
Wave 2
Are variances and unsupported adjustments referring to the same? If not, what's the 
difference.
10	Treasury & Liquidity Treasury 
Intelligence Agents
Provide advisory insights on 
cash positioning, FX exposure 
and short-term liquidity 
forecasting using treasury and 
ERP data feeds.  
Treasury, CFO, 
Finance 
leadership.  
Treasury gains earlier 
warning of funding, liquidity 
and FX risks and can make 
more proactive decisions.  
Treasury reviewer assesses 
alerts, validates assumptions, 
and decides follow-up actions.
Clarify whether PIL wants 
dashboard-only visibility or 
actionable advisory outputs 
on liquidity, FX, and short-
term cash moves.
Keep under Treasury & 
Liquidity. Treat as a later-
wave analytics use case 
unless treasury data 
readiness is already 
strong.
Won’t have 
now
No 12+ months / 
Strategic later 
wave
11	Ecosystem & Trade 
Validation
SGTraDex 
Validation Agents
Enrich AP, bunkering and 
document validation by cross-
checking finance transactions 
against trusted external trade 
events and ecosystem data 
from SGTraDex.  
AP, operations 
finance, bunkering 
teams.  
Transaction validation is 
strengthened with external 
corroboration, reducing 
reliance on internal records 
alone.  
AP or operations finance 
reviewer checks discrepancies 
and confirms disposition of 
flagged cases.
Clarify which PIL process 
should consume SGTraDex 
first: AP, bunkering, 
operational finance, or 
document validation.
Keep as ecosystem / 
operational extension. 
Position as future-wave 
unless a concrete high-
value validation scenario 
is selected.
Won’t have 
now
No 12+ months / 
Strategic later 
wave
12	AR & Cash 
Application
Advanced Cash 
Application Agents
Extend receipt classification 
into remittance matching, 
dispute routing and auto-
suggested cash allocation to 
reduce unresolved receipts.  
AR, Cash 
application, GL 
support.  
Cash application cycle time 
shortens and unapplied cash 
levels fall through better 
matching and routing.  
Cash application reviewer 
approves disputed or low-
confidence matches before 
posting.
Confirm volume of 
unapplied cash, remittance 
formats, dispute types, and 
whether PIL wants this in 
current scope or as next 
wave after classification.
Keep under AR & Cash 
Application. Treat as 
Phase 2 after Bank 
Receipt Classification 
unless data quality is 
already strong.
Could have Possible 6-12 months / 
Wave 2+
13	Cross-Process 
Governance
Cross-Process Risk 
Monitoring Agents
Consolidate abnormal activity 
and recurring control breaks 
across AP, AR, tax and close 
through shared risk flags, 
thresholds and dashboards.  
CFO, COO, Risk & 
Compliance, 
process owners.  
Leadership gets a single view 
of unusual finance activity 
and systemic control 
weaknesses across 
processes.  
Process owner, CFO, or COO 
reviews material risk patterns 
and assigns actions.
Clarify which cross-process 
KPIs and risk thresholds 
matter most to CFO/COO 
and which source agents 
will feed them.
Keep under Cross-Process 
Governance. Position as 
suite-level monitoring 
rather than first-wave 
MVP.
Won’t have 
now
No 12+ months / 
Strategic later 
wave
14	Cross-Process 
Governance
Agent Performance 
Governance Agents
Track throughput, exception 
rates, reviewer turnaround 
and overrides across all agents 
to govern scale-up and policy 
tuning.  
CFO, Finance 
Transformation 
Lead, IT.  
Finance and IT can scale 
agents in a controlled 
manner and tune policies 
based on evidence.  
Transformation lead or product 
owner reviews agent KPIs and 
approves tuning actions.
Clarify whether PIL wants 
governance of deployed 
agents now or only after 
several use cases go live.
Keep under Cross-Process 
Governance. Position as 
enablement for scale-up 
rather than immediate 
MVP.
Won’t have 
now
No 12+ months / 
Strategic later 
wave
15	Tax & Compliance Extended Tax 
Compliance Agents
Build on reconciliation outputs 
to generate audit-ready 
evidence packs, regulatory 
workpapers and recurring 
compliance schedules.  
Group Tax, 
compliance teams.  
Audit readiness improves 
and manual assembly of tax 
support files is reduced.  
Tax preparer / reviewer 
validates workpapers and signs 
off compliance packs.
Clarify which compliance 
packs, workpapers, and 
sign-off artefacts are in 
scope beyond tax 
reconciliation.
Keep under Tax & 
Compliance as an 
extension of Tax 
Reconciliation.
Won’t have 
now
No 12+ months / 
Strategic later 
wave
16	Treasury & Liquidity Working Capital 
Insight Agents
Combine AP, AR and 
statement-cycle data to show 
working-capital drivers and 
support cash-conversion 
decision-making.  
CFO, Treasury, 
Finance 
leadership.  
Leadership gains clearer 
visibility into payables, 
receivables and cash-
conversion performance.  
CFO, treasury, or finance lead 
reviews insights and confirms 
management actions.
Clarify target metrics, 
frequency, and whether 
leadership wants 
explanatory dashboards or 
scenario capability.
Keep under Treasury & 
Liquidity as a suite-level 
insight layer.
Won’t have 
now
No 12+ months / 
Strategic later 
wave
17	Operational Finance Operational 
Finance Agents
Reuse the same architecture 
for voyage cost checks, bunker 
reconciliation and operational 
billing validation across 
shipping operations 
workflows.  
Operations 
finance, shipping 
operations, 
Controllership.  
Finance-grade controls 
extend into shipping 
operations, improving end-
to-end visibility and 
validation without replacing 
core systems.  
Operations finance or 
controllership reviewer 
assesses exceptions and 
confirms case resolution.
Clarify which operational-
finance scenario matters 
first, e.g. voyage cost, 
bunker reconciliation, or 
billing validation.
Keep under Operational 
Finance as a future 
industry-specific 
extension.
Won’t have 
now
No 12+ months / 
Strategic later 
wave

--- USER STORY ---
# Automated AFS Submission to GFRA/CAC with Agentic Oversight

## Overview
This is an **Agentic AI App** designed to standardise and streamline PIL's Annual Financial Statement (AFS) submission process to GFRA/CAC. Today, submissions rely on manual coordination, ad-hoc follow-ups, and decentralised tracking — leading to limited visibility, late/missed submissions, and weak audit trails. The proposed solution introduces an autonomous **AFS Submission Agent** that orchestrates the end-to-end submission workflow, performs intelligence checks for completeness and anomalies, and routes exceptions to human reviewers for approval.

## Actors
- **Financial Reporting Team (Preparer)**: Prepares the AFS pack from Oracle GFS/CAC sources and uploads it for submission. Responds to clarification requests from the agent.
- **CAC / Controllership Reviewer (Approver)**: Reviews flagged or completed submissions, resolves late/incomplete cases, and provides final approval.
- **Finance Manager / Process Owner**: Monitors submission status across entities via dashboards and intervenes on escalations.
- **AFS Submission Agent (Autonomous AI Agent)**: Detects submission triggers, validates pack completeness against a defined checklist, applies anomaly/lateness risk scoring, routes packs for review, sends reminders, and logs the full audit trail. Defers to humans when confidence is low or thresholds are breached.
- **Notification & Reminder Sub-Agent**: Sends timed nudges, escalations, and status updates to relevant stakeholders.

## Goals
- **Primary**: Achieve 100% on-time, complete, and traceable AFS submissions to GFRA/CAC across all in-scope entities.
- Reduce manual coordination effort and eliminate key-person dependency.
- Strengthen auditability through a centralised log of submissions, versions, reviews, and approvals.
- Improve reviewer prioritisation by surfacing lateness risk and completeness gaps proactively.
- Provide leadership with real-time visibility into submission status and variance via dashboards.

## User Story
As a **Financial Reporting Lead**, I want an **AI agent to orchestrate, validate, and track AFS submissions end-to-end**, so that **submissions are timely, complete, auditable, and require minimal manual follow-up while preserving controllership oversight on exceptions and approvals**.

## Detailed Workflow

1. **Trigger Detection (Agent)**
   - The AFS Submission Agent monitors the financial calendar and detects upcoming submission deadlines per entity.
   - It pre-notifies the Financial Reporting team of the upcoming submission window with a checklist of expected pack components.

2. **Pack Ingestion & Validation (Agent)**
   - Once the preparer uploads the AFS pack (extracted from Oracle GFS/CAC), the agent ingests the document(s) and validates against a configurable checklist:
     - Required schedules and notes present
     - Account mapping conforms to the standard taxonomy
     - Format compliance (template, currency, period)
     - Mandatory fields populated (signatories, dates, entity codes)
   - The agent runs a **Quick Dashboard reconciliation** comparing key totals (e.g., revenue, expenses, net assets) against source GFS/CAC numbers to flag variances above tolerance.

3. **Anomaly & Completeness Scoring (Agent Decision Logic)**
   - The agent classifies each submission into one of three states:
     - ✅ **Clean** — passes all checks → routes directly to reviewer for approval.
     - ⚠️ **Needs Clarification** — minor gaps or variances → agent issues a structured query back to the preparer with specifics.
     - 🚩 **High Risk / Incomplete** — material gaps, lateness risk, or large variances → escalates to Controllership with a summary of issues.

4. **Review Routing & Reviewer Prioritisation (Agent)**
   - Routes submissions to the appropriate CAC / Controllership reviewer based on entity and approval matrix.
   - Prioritises the reviewer's queue using lateness risk, materiality of variances, and submission complexity.

5. **Human Review & Approval (Reviewer)**
   - The CAC reviewer examines the pack alongside the agent's validation report and variance dashboard.
   - The reviewer can approve, request rework (agent re-routes to preparer), or escalate further.

6. **Version Control & Resubmission (Agent)**
   - If the preparer resubmits a revised pack, the agent versions it, re-runs validation, and presents a diff/change summary to the reviewer.

7. **Reminders, Escalations & Logging (Agent)**
   - Automated reminders are sent at defined intervals before deadlines.
   - Escalations are sent to Finance Managers if a submission remains unresolved past defined thresholds.
   - All actions, decisions, communications, and approvals are logged in an immutable audit trail.

8. **Dashboard & Reporting**
   - A live submission dashboard shows status (Not Started / In Progress / Under Review / Approved / Late), variance summaries, and reviewer queue depth.

9. **Hand-off & Closure**
   - On final approval, the agent marks the submission complete, archives the pack and audit log, and notifies stakeholders.

## Acceptance Criteria

**Functional**
- The agent automatically detects all in-scope AFS submission deadlines and initiates the workflow without manual triggering.
- The agent validates uploaded packs against the configured completeness checklist and produces a validation report within 5 minutes of upload.
- The agent performs Quick Dashboard reconciliation between submitted pack totals and GFS/CAC source data, flagging variances exceeding defined tolerance.
- All resubmissions are versioned with a visible change summary.
- A consolidated submission dashboard is available to Finance Managers showing real-time status across entities.
- The full audit trail (uploads, validations, communications, approvals) is exportable for audit purposes.

**Agent Guardrails & Escalation**
- The agent **must not** mark a submission as "approved" — final approval is always a human reviewer action.
- If validation confidence is below a defined threshold (e.g., ambiguous account mappings, OCR uncertainty), the agent must defer to a human and explicitly state its uncertainty.
- The agent must escalate to Controllership if a submission is at risk of being late by a configurable threshold (e.g., T-3 days).
- The agent must stop and request human intervention if the pack format is unrecognised or if source-data feeds (Oracle GFS/CAC) are unavailable.
- The agent's reasoning (which checks passed/failed, which variances were flagged) must be transparent and auditable for every decision.
- Override actions by reviewers (e.g., approving despite a flag) must be captured with rationale.

**Performance**
- ≥ 95% of clean submissions routed to reviewers within 10 minutes of upload.
- Reduction in late submissions by at least 80% within 6 months of go-live.
- Reviewer queue reflects updated prioritisation in near real-time.

## Assumptions & Constraints
- AFS packs are sourced from Oracle GFS and CAC; data extraction interfaces are available or can be enabled.
- A standard AFS template and account-mapping taxonomy will be defined and maintained as configurable rules (definition of "standardisation" to be confirmed with PIL — covering account mapping, format, and mandatory fields).
- The fields that determine "incomplete" status will be documented and version-controlled by the Financial Reporting team.
- Quick Dashboard variance tolerance thresholds will be defined jointly with Controllership.
- The agent operates under a **workflow-led with intelligence checks** model — it does not autonomously approve or submit externally.
- Approach: **Low-code / no-code first** (Power Automate for routing/reminders/approvals, Power BI for dashboards, Copilot/virtual agent for user guidance), with pro-code extensions only if intelligence checks demand it.
- MVP scope is Wave 1 (0–6 months); priority is **Must Have**.
- Human-in-the-loop is mandatory at the approval step and at all high-risk exception points.
- Integration with downstream regulatory submission portals (if any) is out of scope for MVP.

--- FEATURE LIST SUMMARY ---
This solution enables PIL to standardise and automate the end-to-end Annual Financial Statement (AFS) submission process to GFRA/CAC through an autonomous AFS Submission Agent with human-in-the-loop oversight. Primary actors include the Financial Reporting Team (Preparer), CAC/Controllership Reviewer, Finance Manager, and the AFS Submission Agent with its Notification Sub-Agent. The flow spans calendar-driven trigger detection, pack ingestion, validation and anomaly scoring, reviewer routing, human approval, versioned resubmission, reminders, and dashboard reporting. The Master Data Configuration cluster holds users, roles, entities, approval matrices, AFS templates, checklist rules, and tolerance thresholds. The 16 features deliver autonomous validation, prioritised human review, and full auditability.

Feature Clusters & Features:
• Master Data Configuration
  - 1. User and Role Management — Maintains preparers, reviewers, and managers with role-based access to AFS submissions and approval rights.
  - 2. Entity and Approval Matrix Setup — Registers in-scope entities and maps each to its assigned CAC reviewer and approval hierarchy.
  - 3. AFS Checklist and Template Configuration — Defines the standard AFS template, mandatory fields, account taxonomy, and completeness checklist rules.
  - 4. Variance Tolerance and Threshold Rules — Configures reconciliation tolerance bands, lateness thresholds, and escalation triggers used by the agent.
  - 5. Financial Calendar Setup — Stores per-entity AFS submission deadlines and reporting periods that drive automated trigger detection.
• Submission Trigger and Pre-Notification
  - 6. Deadline Detection and Pre-Notification — Agent monitors the calendar and proactively alerts preparers of upcoming submission windows with the expected pack checklist.
• Pack Ingestion and Validation
  - 7. AFS Pack Upload and Ingestion — Allows the preparer to upload AFS packs sourced from Oracle GFS/CAC into the system for processing.
  - 8. Automated Completeness and Format Validation — Agent validates uploaded packs against the configured checklist and produces a validation report within minutes.
  - 9. Quick Dashboard Reconciliation — Agent compares pack totals against GFS/CAC source data and flags variances exceeding tolerance.
• Anomaly Scoring and Routing
  - 10. Anomaly and Completeness Risk Scoring — Classifies each submission as Clean, Needs Clarification, or High Risk based on validation results and lateness risk.
  - 11. Reviewer Routing and Queue Prioritisation — Routes submissions to the right CAC reviewer and prioritises queues by lateness, materiality, and complexity.
• Human Review and Approval
  - 12. Reviewer Workbench and Approval — Reviewer examines the pack alongside the agent's report and approves, requests rework, or escalates.
  - 13. Versioned Resubmission and Diff View — Tracks resubmitted packs with version history and presents change summaries to reviewers.
• Reminders and Escalations
  - 14. Automated Reminders and Escalations — Sub-agent sends timed nudges to preparers and escalates unresolved cases to Finance Managers past defined thresholds.
• Monitoring and Reporting
  - 15. Submission Status Dashboard — Provides Finance Managers a real-time view of submission status, variances, and reviewer queue depth across entities.
• Agent Orchestration and Audit
  - 16. Agent Audit Trail and Guardrail Logging — Captures every agent decision, confidence level, override rationale, and communication in an immutable, exportable audit log.

--- FEATURE LIST (16 features across 8 clusters) ---

#1 | Cluster: Master Data Configuration | Feature: User and Role Management
  Description: Central registry of users, roles, and access rights governing every step of the AFS submission workflow. Ensures only authorised actors can prepare, review, or approve submissions.
  Workflow:
    1. Admin creates user accounts for preparers, reviewers, and managers.
    2. Admin assigns roles and permissions per entity scope.
    3. System enforces role-based access on submissions and approvals.
    4. Audit log records all user/role changes.
  Table:       users_roles
  Columns:     id (bigint, pk), user_name (varchar 150), email (varchar 255), role_code (varchar 50), entity_scope (varchar 255), is_active (boolean), created_at (timestamp)
  Actor:       System Administrator
  AI Agent:    None
  ----
#2 | Cluster: Master Data Configuration | Feature: Entity and Approval Matrix Setup
  Description: Maintains the catalog of entities and the reviewer/approver matrix used to route submissions correctly. Drives all reviewer assignment and escalation decisions made by the agent.
  Workflow:
    1. Admin registers each in-scope legal entity.
    2. Admin maps reviewers and approvers to each entity.
    3. Defines approval hierarchy and escalation tiers.
    4. Saves matrix used by routing logic.
  Table:       entity_approval_matrix
  Columns:     id (bigint, pk), entity_code (varchar 50), entity_name (varchar 255), primary_reviewer_id (bigint, fk), approver_id (bigint, fk), escalation_user_id (bigint, fk), is_active (boolean)
  Actor:       System Administrator
  AI Agent:    None
  ----
#3 | Cluster: Master Data Configuration | Feature: AFS Checklist and Template Configuration
  Description: Stores the configurable AFS template, mandatory fields, account taxonomy, and completeness checklist rules. The AFS Submission Agent uses this configuration to validate uploaded packs.
  Workflow:
    1. Finance Reporting Lead defines AFS template and mandatory fields.
    2. Configures account-mapping taxonomy and required schedules.
    3. Versions the checklist with effective dates.
    4. Publishes for use by the validation agent.
  Table:       afs_checklist_config
  Columns:     id (bigint, pk), checklist_version (varchar 20), field_name (varchar 150), is_mandatory (boolean), taxonomy_code (varchar 100), effective_from (date), created_by (bigint, fk)
  Actor:       Financial Reporting Lead
  AI Agent:    None
  ----
#4 | Cluster: Master Data Configuration | Feature: Variance Tolerance and Threshold Rules
  Description: Captures tolerance percentages, lateness thresholds, and agent confidence cutoffs used in anomaly and risk scoring. Allows business to tune agent sensitivity without code changes.
  Workflow:
    1. Controllership defines variance tolerance bands per metric.
    2. Sets lateness risk thresholds (e.g., T-3 days).
    3. Configures confidence cutoffs for agent deferral.
    4. Saves rules consumed by scoring logic.
  Table:       threshold_rules
  Columns:     id (bigint, pk), rule_type (varchar 50), metric_name (varchar 100), tolerance_value (decimal 10,4), threshold_unit (varchar 20), effective_from (date), is_active (boolean)
  Actor:       Controllership Lead
  AI Agent:    None
  ----
#5 | Cluster: Master Data Configuration | Feature: Financial Calendar Setup
  Description: Maintains the AFS submission calendar with deadlines per entity and reporting period. The agent monitors this calendar to trigger pre-notifications and detect lateness.
  Workflow:
    1. Admin defines AFS reporting periods per entity.
    2. Sets submission deadline and reminder cadence.
    3. Publishes calendar to the agent.
    4. Updates calendar for new fiscal years.
  Table:       financial_calendar
  Columns:     id (bigint, pk), entity_code (varchar 50, fk), reporting_period (varchar 20), submission_deadline (date), reminder_offsets (varchar 100), fiscal_year (varchar 10)
  Actor:       Financial Reporting Lead
  AI Agent:    None
  ----
#6 | Cluster: Submission Trigger and Pre-Notification | Feature: Deadline Detection and Pre-Notification
  Description: Autonomously detects upcoming AFS deadlines and pre-notifies preparers with the expected checklist. Eliminates manual workflow initiation and ensures preparers start with full context.
  Workflow:
    1. Agent scans financial calendar daily.
    2. Identifies upcoming AFS submission windows per entity.
    3. Generates a checklist of expected pack components.
    4. Sends pre-notification to assigned preparer.
    5. Logs trigger event in audit trail.
  Table:       submission_triggers
  Columns:     id (bigint, pk), entity_code (varchar 50, fk), reporting_period (varchar 20), trigger_date (timestamp), preparer_id (bigint, fk), checklist_version (varchar 20), status (varchar 30)
  Actor:       AFS Submission Agent
  AI Agent:    AFS Submission Agent
  ----
#7 | Cluster: Pack Ingestion and Validation | Feature: AFS Pack Upload and Ingestion
  Description: Provides a secure upload interface for preparers to submit AFS packs sourced from Oracle GFS/CAC. Captures metadata that drives downstream validation and routing.
  Workflow:
    1. Preparer extracts AFS pack from Oracle GFS/CAC.
    2. Uploads pack documents to the portal.
    3. System tags upload with entity, period, and version.
    4. Agent acknowledges receipt and queues for validation.
  Table:       afs_submissions
  Columns:     id (bigint, pk), entity_code (varchar 50, fk), reporting_period (varchar 20), version_no (int), preparer_id (bigint, fk), uploaded_at (timestamp), file_path (varchar 500), status (varchar 30)
  Actor:       Financial Reporting Team (Preparer)
  AI Agent:    AFS Submission Agent
  ----
#8 | Cluster: Pack Ingestion and Validation | Feature: Automated Completeness and Format Validation
  Description: Agent runs configurable checklist validation on every uploaded pack and produces a transparent pass/fail report. Identifies missing schedules, incorrect mappings, or format gaps before review.
  Workflow:
    1. Agent ingests uploaded pack and parses sections.
    2. Validates against the active checklist (schedules, fields, format).
    3. Verifies signatories, dates, and entity codes.
    4. Generates a structured validation report within 5 minutes.
    5. Stores result against the submission.
  Table:       validation_results
  Columns:     id (bigint, pk), submission_id (bigint, fk), check_code (varchar 100), check_status (varchar 20), details (text), confidence_score (decimal 5,2), evaluated_at (timestamp)
  Actor:       AFS Submission Agent
  AI Agent:    AFS Submission Agent
  ----
#9 | Cluster: Pack Ingestion and Validation | Feature: Quick Dashboard Reconciliation
  Description: Performs automated reconciliation of pack totals against Oracle GFS/CAC source data. Flags material variances and provides reviewers with context for risk-based prioritisation.
  Workflow:
    1. Agent fetches source totals from Oracle GFS/CAC.
    2. Extracts key totals (revenue, expenses, net assets) from pack.
    3. Computes variance against tolerance bands.
    4. Flags variances exceeding tolerance with explanation.
    5. Attaches reconciliation summary to submission.
  Table:       reconciliation_results
  Columns:     id (bigint, pk), submission_id (bigint, fk), metric_name (varchar 100), pack_value (decimal 18,2), source_value (decimal 18,2), variance_pct (decimal 8,4), is_flagged (boolean)
  Actor:       AFS Submission Agent
  AI Agent:    AFS Submission Agent
  ----
#10 | Cluster: Anomaly Scoring and Routing | Feature: Anomaly and Completeness Risk Scoring
  Description: Combines validation and reconciliation results into a risk classification that drives the next routing action. Transparently records reasoning and confidence to support auditability.
  Workflow:
    1. Agent aggregates validation and reconciliation outcomes.
    2. Applies threshold rules to compute risk score.
    3. Classifies submission as Clean, Needs Clarification, or High Risk.
    4. If confidence below threshold, defers to human with rationale.
    5. Persists classification and reasoning.
  Table:       submission_risk_scores
  Columns:     id (bigint, pk), submission_id (bigint, fk), classification (varchar 30), risk_score (decimal 5,2), confidence (decimal 5,2), reasoning (text), scored_at (timestamp)
  Actor:       AFS Submission Agent
  AI Agent:    AFS Submission Agent
  ----
#11 | Cluster: Anomaly Scoring and Routing | Feature: Reviewer Routing and Queue Prioritisation
  Description: Routes each submission to the right reviewer and orders their queue by risk and lateness. Ensures controllership attention focuses on the most material and time-critical packs first.
  Workflow:
    1. Agent looks up reviewer from approval matrix by entity.
    2. Calculates priority using lateness, materiality, complexity.
    3. Assigns submission to reviewer queue with priority rank.
    4. For Needs Clarification, sends structured query to preparer.
    5. For High Risk, escalates to Controllership with summary.
  Table:       reviewer_queue
  Columns:     id (bigint, pk), submission_id (bigint, fk), reviewer_id (bigint, fk), priority_rank (int), routing_reason (varchar 255), assigned_at (timestamp), queue_status (varchar 30)
  Actor:       AFS Submission Agent
  AI Agent:    AFS Submission Agent
  ----
#12 | Cluster: Human Review and Approval | Feature: Reviewer Workbench and Approval
  Description: Provides reviewers a single workbench to inspect packs alongside agent findings and take approval actions. Final approval is always a human action, with overrides captured for audit.
  Workflow:
    1. Reviewer opens submission in workbench.
    2. Reviews pack alongside agent validation and variance report.
    3. Approves, requests rework, or escalates.
    4. If approving despite a flag, captures override rationale.
    5. System records final decision and notifies stakeholders.
  Table:       review_decisions
  Columns:     id (bigint, pk), submission_id (bigint, fk), reviewer_id (bigint, fk), decision (varchar 30), override_flag (boolean), rationale (text), decided_at (timestamp)
  Actor:       CAC / Controllership Reviewer
  AI Agent:    AFS Submission Agent
  ----
#13 | Cluster: Human Review and Approval | Feature: Versioned Resubmission and Diff View
  Description: Maintains version history for resubmitted packs with automated change summaries. Enables reviewers to focus on what changed rather than re-reading the entire pack.
  Workflow:
    1. Preparer uploads revised pack after rework request.
    2. Agent assigns new version number linked to original submission.
    3. Re-runs validation and reconciliation.
    4. Generates a diff/change summary versus prior version.
    5. Presents diff to reviewer for re-evaluation.
  Table:       submission_versions
  Columns:     id (bigint, pk), submission_id (bigint, fk), version_no (int), parent_version_id (bigint, fk), diff_summary (text), uploaded_at (timestamp), uploaded_by (bigint, fk)
  Actor:       Financial Reporting Team (Preparer)
  AI Agent:    AFS Submission Agent
  ----
#14 | Cluster: Reminders and Escalations | Feature: Automated Reminders and Escalations
  Description: Sub-agent autonomously sends timely reminders and escalations to keep submissions on track. Reduces key-person dependency and ensures no submission is overlooked.
  Workflow:
    1. Sub-agent reads reminder cadence and deadlines.
    2. Sends nudges to preparers at configured intervals.
    3. Notifies reviewers of pending queue items.
    4. Escalates to Finance Manager past lateness threshold.
    5. Logs every notification sent.
  Table:       notifications_log
  Columns:     id (bigint, pk), submission_id (bigint, fk), recipient_id (bigint, fk), notification_type (varchar 50), channel (varchar 30), sent_at (timestamp), status (varchar 20)
  Actor:       Notification & Reminder Sub-Agent
  AI Agent:    Notification & Reminder Sub-Agent
  ----
#15 | Cluster: Monitoring and Reporting | Feature: Submission Status Dashboard
  Description: Power BI-style dashboard giving Finance Managers real-time visibility into AFS submission status and risk across the enterprise. Surfaces lateness and variance trends for proactive intervention.
  Workflow:
    1. Dashboard aggregates status across all entities.
    2. Displays Not Started / In Progress / Under Review / Approved / Late.
    3. Shows variance summaries and reviewer queue depth.
    4. Allows drill-down to individual submission detail.
    5. Refreshes in near real-time.
  Table:       dashboard_metrics
  Columns:     id (bigint, pk), entity_code (varchar 50, fk), reporting_period (varchar 20), status (varchar 30), variance_summary (text), queue_depth (int), refreshed_at (timestamp)
  Actor:       Finance Manager / Process Owner
  AI Agent:    AFS Submission Agent
  ----
#16 | Cluster: Agent Orchestration and Audit | Feature: Agent Audit Trail and Guardrail Logging
  Description: Centralised, immutable audit log of all agent reasoning, guardrail events, communications, and human overrides. Provides full transparency and exportable evidence for audit and governance.
  Workflow:
    1. Agent logs every decision, check, and confidence score.
    2. Captures guardrail triggers (low confidence, unrecognised format, source unavailable).
    3. Records human overrides with rationale.
    4. Stores immutable timeline per submission.
    5. Supports export for internal/external audit.
  Table:       agent_audit_log
  Columns:     id (bigint, pk), submission_id (bigint, fk), event_type (varchar 50), agent_name (varchar 100), event_payload (text), confidence (decimal 5,2), guardrail_triggered (boolean), event_time (timestamp)
  Actor:       AFS Submission Agent
  AI Agent:    AFS Submission Agent
  ----

==================================================
 END OF STORY EXPORT
==================================================
