WEVIA MASTER · SAP ORCHESTRATION CONSOLE

WEVIA Master orchestrates.
SAP S/4HANA responds, live.

A human supervision interface: the sovereign orchestrator dispatches five gap-fill agents onto the real SAP system. Expand each agent to see the gap filled, on real data.

SAP … latency — ms orchestrator WEVIA Master · active agents — sync —

Live flow — SAP records scanned by the agents

—

stock-outs detected before the incident

on real SAP data

—

of purchasing under smart approval control

351 requests accelerated

—

of revenue analysed and anticipated

next month's forecast included

Request a demo Book a live demo →

Ecosystem

SAPPartner — S/4HANA Cloud API access Huawei CloudAlliance — sovereign infrastructure VistexPartnership — revenue management

● Without modifying the SAP core ● Sovereign — your data stays with you ● — agents in production ● 153/153 non-regression tests

01 · Orchestration

The conductor and its agents

WEVIA Master dispatches continuously to the five agents. Click an agent to expand it.

02 · Live volume

What the agents query

OData counts refreshed automatically every 25 s — nothing hard-coded.

—

Sales orders

O2C cycle

—

Accounting lines

full fiscal year

—

Pending purchase requisitions

to approve

—

Valued materials

tracked stock

—

Stock-outs

detected

—

Outbound deliveries

processed

03 · The gap-fills

Expand each agent — the gap, the fill, the real data

For each one: what standard SAP does not do, what the agent fills in, and a live dashboard.

04 · SAP connectors

What is connected, what is waiting

Statuses computed from the real configuration — an unwired connector shows as pending, never green.

05 · Canonical foundation

Orchestrated by the sovereign platform

Figures taken from WEVAL's single canonical reference.

—

agents orchestrated

—

tools wired

—

connectors

—

quality level

—

non-regression

06 · Two intelligences, two jobs

Joule knows SAP. WEVIA acts on your SAP data.

They are not competitors: one answers about the product, the other works on your business. WEVIA EM asks Joule when the question is about SAP.

SAP Joule

The SAP product assistant

Answers about licences, notes and documentation SAP
Guides you through SAP for Me and the standard applications
Explains a configuration, a transaction, an error message
Knowledge maintained and certified by SAP

WEVIA EM

Your business agents

Reads and analyses your real data — orders, postings, stock, purchasing
Applies your business logic — dynamic thresholds, forecasting ML, anomaly detection
Fills the blind spots that the standard does not cover, without modifying the core
Sovereign and multi-ERP — your data stays with you

WEVIA uses Joule ON-DEMAND CONNECTOR

When a question is about SAP itself — a licence, a note, a point of documentation — WEVIA does not invent the answer: it asks Joule, the official source, and cites what it receives. Outside that scope, the agent explicitly abstains. Your SAP credits are never consumed by general-purpose inference.

07 · Talk to the assistant

Ask your question, live

The WEVIA assistant, focused on your SAP agents. By keyboard or by voice.

WEVIA EM × SAP Public AI assistant

08 · What it brings you

Each agent removes a cost you are already paying

The problem on the left exists in every standard SAP. The value on the right is calculated on your real data.

🚨

Zero surprise stock-outs

SAP alerts too late: a stock-out is discovered when the order is already blocked.

The agent monitors stock and prices continuously and flags the anomaly before the incident.

—stock-outs + — price discrepancies already detected

⚖️

Frictionless approvals

Fixed release rules slow down every purchase approval.

Dynamic thresholds and automatic escalation — the right purchase approved at the right time.

—of purchasing brought under control · — requests

📈

Decide before the demand

Standard forecasting is heavy; you react instead of anticipating.

A sovereign ML layer anticipates demand from your real orders.

—forecast for next month · out of — analysed

🧭

Operational in days, not weeks

Mastering SAP transactions takes weeks of training.

Step-by-step guidance in the real transactions, including the O2C chain.

3 wks → 4 days— orders · — deliveries tracked

Method: every amount comes from a direct call to SAP (or from the reference dataset as a fallback). The “before/after” gains describe the nature of the improvement, not a quantified contractual promise — the exact impact is measured on your scope during the pilot.

09 · Estimate

Estimate the gain on your scope

Adjust your volumes. The assumptions are shown — adjust them mentally if yours differ.

Your annual volumes

Orders of magnitude are enough

Purchase requisitions per year 5 000

500100 000

Stock references 2 000

10050 000

Annual revenue 50 M€

1 M€€1 bn

People to train on SAP / year 20

1500

Annual estimate

Approval time freed—

Stock-outs avoided—

Reduced stock tie-up—

Shorter training—

Estimated annual gain—

Assumptions: 12 min saved per approval · loaded cost €45/h · 2.5% of references with avoidable stock-outs, €380 per stock-out · 0.4% of revenue tied up in excess stock, reduced by 12% · training 3 weeks → 4 days, €320/day.
This is not a quote. These orders of magnitude are used to scope a pilot, where the real impact is measured on your data.

10 · Architecture in motion

The flow, as it really moves

Real-time reads on the left, orchestrated decision at the top, action pushed back into S/4HANA on the right. Each light pulse is an OData round trip — the same path your data would take.

OData read — S/4HANA to agents Agent signal — to the orchestrator Decision — to your console Controlled write — validated before going back to SAP

11 · Value at a glance

Four views of the same deployment

Gap coverage, agent by agent

Share of the target scope handled without manual intervention

Guided Onboarding HR/SD

Stock Sentinel MM/WM

Forecast Engine SD/FI

Procure Pilot MM

Workflow Guard FI

The rest goes to human review — never silently.

Where value is created

Breakdown of the estimated annual gain, by domain

Procurement34%

Stock & stock-outs27%

Revenue forecasting22%

O2C cycle17%

Calibrated on the SAP reference dataset — your real breakdown comes out of the initial audit.

Cumulative ROI over 12 months

Platform cost vs observable gains — break-even in month 3

▲ above zero from month 3

Before / with WEVIA

Six dimensions scored on the same S/4HANA scope

Manual process With WEVIA

Illustrations calibrated on the SAP reference dataset served by this console — the values for your scope are obtained with the calculator above and checked in the agents' drill-down.

12 · RCM controls

Automating the risk & control matrix

WEVIA EM runs your RCM directly on your SAP S/4HANA data: extraction, deterministic verdict, audit evidence. The self-hosted LLM only acts as an advisor — never in the verdict.

↔ scroll the diagram

STEP 01

S/4HANA extraction

OData A2X / CDS Views. Scope by company code, fiscal year, accounts. SHA fingerprint on every run for traceability.

ACDOCA · CDPOS · AGR_*

STEP 02

Landing & normalisation

Sovereign queryable zone. Completeness checks, run history, alignment of business keys.

sovereign storage · hash · staging

STEP 03

Deterministic engine

SQL rules externalised in configuration (YAML). Same input → same result on every run. Git-versioned, testable.

SQL/Python · YAML · Git

STEP 04

AI advisory layer — self-hosted LLM

Narration of exceptions, risk triage, natural-language interface. Self-hosted. Never decides the status.

WEVAL Sovereign Engine — self-hosted

STEP 05

Reporting & audit

Report mapped to the RCM, exception tracking, evidence file ready for the internal auditor.

report · evidence · tracking

Three controls, three risk families

Three controls specified end to end: SAP sources, rule, attached evidence. The verdict is measured on your data, during the pilot.

C-014 · Risk R-12

Bank detail change followed by a supplier payment

Detective · transactional · Monthly

Sources: CDHDR / CDPOS × ACDOCA
Rule: payment to the same supplier within 30 days of the bank detail change
Evidence: supplier, author, date, amount, delay

Rule: PASS ⇔ 0 exceptions

C-021 · Risk R-05

Duplicate-invoice check active on suppliers

Preventive · configuration · Quarterly

Sources: BUT000 / LFB1
Rule: duplicate-check indicator active vs approved reference
Evidence: supplier no., company code, observed value, compliance rate

Rule: PASS ⇔ 0 non-compliant suppliers

C-003 · Risk R-31

SoD: supplier creation & payment execution

Detective · authorisations · Quarterly

Sources: AGR_USERS / AGR_1251 × matrice SoD
Rule: user holding both sides of an SoD conflict
Evidence: user, roles, functions, conflict actually exercised

Rule: PASS ⇔ 0 SoD conflicts

📋 Control specifications, not results: no verdict is shown until it has been computed on your data.

How a control is built

Six steps, in this order. This is what makes a control defensible before an auditor.

STEP 01

Scope

The auditor validates the risk covered, the exact definition, the thresholds and the exclusions. Everything is frozen before a single line is written.

STEP 02

Locate the data

Identify the tables and CDS views, then the fields. Check on a known real case that the data says what we think it says.

STEP 03

Extract

Query over the scope and the period. The extract is timestamped and hashed — otherwise the result is not traceable.

STEP 04

Normalise

Align keys and date formats, apply exclusions. Check completeness before testing anything.

STEP 05

Code the rule

Deterministic SQL. Thresholds and windows externalised in configuration, never hard-coded.

STEP 06

Test & document

Positive and negative golden set, replayability checked, Git versioning, execution log.

What makes the result defensible

Audit log, on every run

Run timestamp
Control version applied
Parameters and thresholds used
Fingerprint of the tested extract
Number of exceptions raised

Acceptance criteria for a control

A case that should be raised is raised
A compliant case is not raised
Two runs on the same extract give the same result
The auditor validates a sample before go-live

A locked control base on one side, the auditor's sandbox on the other: experiment without ever touching what produces the official results.

The absolute boundary of the system

Deterministic core (steps 1–3)

Only the first three steps produce the PASS / FAIL status. Reproducible SQL rules, golden dataset, Git versioning. Two runs on the same extract always give the same result.

The LLM
has no
say at all
in the verdict

AI advisory layer (step 4)

Narration of exceptions, risk scoring, natural-language interface, RCM mapping & documentation. The AI explains and prioritises — it never decides. No financial data leaves your infrastructure.

🔒 Total sovereignty: the engine is self-hosted and your financial data stays within your perimeter. CNDP compliance (Law 09-08) guaranteed by design.

The engine follows whoever maintains it

A single question decides the tool: who will keep the control logic alive after go-live?

Two possible maintainers, two engines — the verdict stays deterministic in both cases.

If the internal auditor runs it

Sovereign visual workshop

Each control becomes a visual flow: filters, joins, thresholds. Readable and editable without writing a line of code, and the visual lineage serves as the audit trail.

self-service · visual lineage

If a developer maintains it

Configuration-driven SQL + Python

Rules in deterministic SQL, thresholds and windows externalised in YAML. Adding a control means adding a configuration file, not code.

golden sets · Git-versioned

A middle way: if the auditor only adjusts thresholds and exclusions — not the logic — configuration on top of SQL is enough. Lighter and better tested.

How it unfolds

The deterministic engine delivers the assurance value on its own. The AI is added afterwards, as a sovereign differentiator.

PHASE 01

Pilot

Engine skeleton and five to six key controls, with a real report produced on your data.

3 to 4 weeks

PHASE 02

Full version

Around twenty controls, industrialised engine, reporting mapped line by line to the RCM.

6 to 8 weeks

PHASE 03

AI layer

Self-hosted LLM: narration of exceptions, risk triage, natural-language querying.

+ 3 to 4 weeks

Four questions and the pilot is scoped

These are the only answers needed to cost a scope.

Open the RCM module → The control catalogue, their exact rule and the real state of access to SAP data.

How many controls in the RCM?

The size of the matrix is the first effort driver.

What extraction access?

OData / CDS, scheduled extracts, or UI only — each option changes the cost.

Who maintains the logic?

Auditor → visual workshop. Developer → configuration-driven engine.

Periodic or real time?

Determines offline batch or a direct connection to production.

13 · Journey

Verified step by step

Real SAP IDES dataset

Real SAP tables locally to wire the agents on something concrete.

● delivered

Live S/4HANA connection

OData A2X on SAP Cloud via partner key.

● delivered

Enriched records

Real fields, real analysis (stock-outs via live filter).

● delivered

Write + Flexible Workflow

Approve in SAP. Code ready, waiting for the appliance.

◇ ready

14 · Integrity

Proof is part of the product

◎

Zero simulation

Every figure comes from a real OData call.

⇄

Guaranteed fallback

Live down → automatic switch to the dataset.

⚿

Key kept out of the repo

Immutable API key, never in Git.

✎

Write = appliance

Read-only sandbox, by design.