Accounting

Pilot wave 3 (M9). Impact: medium. See the hub. Accounting runs late-wave deliberately: statutory data demands the highest accuracy bar, and its use cases reuse extraction patterns proven in procurement/logistics pilots.

Current state (assumptions to validate)

Accounting keeps PRIME PRODUCTS’ books in SoftOne under Greek statutory requirements: supplier invoice entry and matching, myDATA e-books transmission, VAT treatment complicated by the ship-supply business (export-equivalent zero-rated supplies to vessels, bonded transactions, intra-EU and third-country imports), fixed assets, and the statutory close with external accountants/auditors.

Assumptions — to validate in discovery: team ~4–6; supplier invoices arrive as paper/PDF and are keyed manually into SoftOne; three-way matching (PO–receipt–invoice) manual and partial; myDATA reconciliation discrepancies chased by hand; month-end close takes well over a week; document archive part-paper. Pain points: invoice-entry volume, VAT-regime classification errors (costly in audit), myDATA mismatch investigation, close-period crunch.

Target AI-first operating model

By M12 supplier invoices flow scan-first: an agent extracts header/lines, proposes the SoftOne entry with PO match and VAT-regime classification (vessel-supply zero-rating flagged with the supporting document reference), and an accountant reviews and posts. myDATA discrepancies come as a worked list with likely causes, not raw mismatches. Close runs against a live checklist with the assistant drafting reconciliation narratives. Every posting, every VAT judgment, every statutory submission stays human.

AI use cases

Use case	Pain addressed	Data needed	Complexity	Impact	Pilot
Supplier invoice extraction → draft SoftOne entry with PO match	Manual keying, matching gaps	Invoice scans/PDFs, SoftOne POs/receipts	M	H	Y
VAT-regime classification suggestion (vessel supply, bonded, import, domestic) with rule citation	Misclassification risk	Transaction context, VAT rule reference set	M	H	Y
myDATA reconciliation triage (mismatch → likely cause → fix suggestion)	Manual discrepancy chasing	myDATA reports, SoftOne ledgers	M	M	N
Month-end close checklist + reconciliation narrative drafting	Close crunch	GL data, close checklist, prior narratives	L	M	N
Accounting Q&A over internal policies and Greek GAAP references	Interruptions to seniors	Policy docs, curated reference set	L	L	N

Process transformation opportunities

• Invoice pipeline: scan-first intake with extraction and matching, exceptions-only human work.
• VAT assurance: every zero-rated/bonded transaction linked to its supporting document at entry time, making audits document-complete by construction.
• Continuous close: reconciliations run weekly with agent drafts, flattening the month-end peak.

Required data sources

• SoftOne: AP, GL, POs, goods receipts, VAT setup; myDATA transmission reports.
• Invoice scans/PDFs (accounting mailbox + paper → scan); customs documents (from logistics) supporting zero-rating.
• Close checklists and prior reconciliation workpapers (Excel).

Potential AI agents

• Invoice Entry Agent — extracts and drafts AP entries with PO match; accountant approves every posting.
• VAT Classification Agent — suggests regime with rule citation; accountant decides; systematic sampling review by chief accountant.
• Reconciliation Agent — drafts myDATA and account reconciliations; accountant validates.

Automation opportunities

• n8n: invoice mailbox/scanner → extraction queue; myDATA report pull and diff; close-checklist tracker with reminders.
• Scheduled AP-aging and accrual-candidate reports.

Required integrations

• SoftOne read (AP, GL, POs) and draft-entry write — see integrations; myDATA report access; scan pipeline.

KPIs

KPI	Baseline	M12 target
Invoice processing time (receipt → posted)	TBD (M2)	−60%
Invoices entered via extraction pipeline	0	≥80%
VAT classification corrections at review	n/a	≤2% of suggestions
Month-end close duration	TBD	−40%
myDATA unresolved mismatches at period end	TBD	~0

Risks

• Statutory error from over-trusted extraction/classification → human posting mandatory; sampling audits; external accountant briefed on the workflow.
• VAT rule set goes stale after regulatory change → rule corpus owned by chief accountant with compliance; review on every circular.
• Sensitive financial data scope → same segregated access tier as finance.

Training needs

• Accountants: extraction-review workflow, classification-verification discipline (M8–M9).
• Chief accountant: sampling routine, rule-corpus maintenance (M9).

Deliverables

• Invoice pipeline (pilot M9, production M11); VAT rule reference corpus.
• myDATA triage reporting; close checklist system.

12-month execution milestones

Month	Milestone
M1–M2	Discovery: invoice volume/format analysis, VAT-regime inventory, close walkthrough
M4	Assistant onboarding (general drafting/Q&A)
M6	VAT rule corpus drafted and validated
M8	Training; extraction accuracy tested offline on historical invoices
M9	Wave-3 pilot: invoice extraction + VAT suggestions on live intake
M10	Pilot evaluation; myDATA triage live
M11	Agents in production; continuous-close cadence begins
M12	KPI review; external-auditor walkthrough of the workflow