Entelli AI · Supply Chain Intelligence · Q2C · Core Finance · Strategic Finance

Agentic AI for Supply Chain,
Q2C, Core Finance & Strategic Finance.

Entelli AI is an agentic intelligence layer over your ERP, WMS, CRM, billing, and planning data. 39 supply chain signals across 8 domains, autonomous SC Ops Copilot, hypothesis‑tested root‑cause reports, and a three‑model Anomaly Feed — every recommendation human‑approved, every action auditable. Every answer shows the SQL, the rows, and the reasoning.

See Entelli AI · Supply Chain See Fintelli AI · Q2C
  • 39 SC signalsacross 8 supply chain domains
  • 18 ML use casesO2C · P2P · R2R · Treasury · FP&A
  • 7‑node LangGraphplan · hypothesise · critique · cite
  • 100% HITLhuman approves every write

Our products

Two agentic AI products, one investigation engine.

Entelli AI is the foundational agentic investigation engine. Fintelli AI packages that engine into finance‑domain modules. Fintelli is Entelli applied to Q2C and core finance; Entelli's Supply Chain suite runs the same engine over your ERP, WMS, and planning data.

Product

Fintelli AI

AI Copilot for Q2C, Core Finance & Strategic Finance — over Salesforce, Zuora, NetSuite, Oracle EBS.

A self‑hosted AI copilot over your Quote‑to‑Cash and finance stack. A semantic layer teaches it your data model; a state‑machine agent plans, runs SQL, self‑critiques, and cites. Analysts get answers in seconds; finance leaders get investigations in minutes instead of days. Five copilot modes, five O2C ML models, and modular FP&A forecasting — one integrated app.

Copilot modes

  • Chat — plain‑English Q&A; SQL expanded for audit; follow‑ups keep context
  • Analyze — plan‑first, execute‑many, narrative‑first Business Review briefs
  • Investigate — LangGraph agent plans multi‑hypothesis, tests each with SQL, cites evidence
  • Anomaly Feed — SARIMAX + Prophet + XGBoost consensus over Bookings, Revenue, AP Expense
  • Data Quality — Q2C Tie‑Out — reconciles CPQ · Billing · REBS · GL with nine break types

O2C ML models

  • Churn Prediction — retain the accounts that pay the bills
  • Dunning Risk Scoring — lower DSO by focusing on best recovery odds
  • Revenue Forecast (Segment) — board‑defensible forecast in minutes
  • Dunning‑Letter Agent — collections cadence at scale; rep signs off
  • Cash Application — auto‑apply payments; unapplied cash days → hours (PII‑safe)
LangGraphClaude Sonnet 4.6 / Haiku 4.5 / Opus 4.7DuckDB (read‑only)Chroma RAGPhoenix tracingSARIMAX · Prophet · XGBoostPydanticStreamlitSQLite checkpoints

Product

Entelli AI · SC Suite

Supply Chain Intelligence — 39 signals across 8 domains, one autonomous SC Ops Copilot, every action HITL‑gated.

Entelli AI's Supply Chain suite is an agentic intelligence layer that sits over your ERP, WMS, and planning data. Thirty‑nine AI signals scan demand, inventory, shipping, contract manufacturing, costing, work‑in‑process, bill‑of‑materials, and data quality — continuously, automatically, and with every recommendation routed through a human‑in‑the‑loop gate before any action is taken. One autonomous orchestrator — the SC Ops Copilot — coordinates them all, triages findings by financial impact, and delivers an executive brief every cycle.

Coverage

  • 39 signals · 8 domains · 100% HITL · Haiku‑narrated for cost efficiency
  • Sits over your ERP · WMS · planning stack; no data movement
  • Append‑only audit trail — approve / defer / dismiss with reason codes
  • Urgency triage (high / medium / low) prioritises by financial impact
  • SC Ops Copilot orchestrates all 39 signals into an executive brief
  • Every recommendation cites the SQL and rows behind it

Example outcomes

  • Excess Redistribution proposes an inter‑site transfer with cost/benefit analysis — a controller approves in one click, freeing trapped cash.
  • Yield Variance (SPC) flags an out‑of‑control point on a new CM ramp; the AI opens a CAPA with narrative root cause.
  • Freight Audit surfaces invoices 2%+ over contracted rates and generates a ready‑to‑file credit claim proposal.
LangGraphClaude Haiku 4.5 (cost‑optimised)DuckDB (read‑only)Chroma RAGPhoenix observabilitySPC · SARIMAX · analog matchingAppend‑only auditSC Ops CopilotStreamlit

Fintelli AI · Q2C deep dive

Every dollar of revenue lives in four systems. Fintelli proves they agree.

Opportunities in Salesforce, subscriptions and invoices in Zuora, GL in NetSuite, AP and expense in Oracle EBS. Every meaningful question crosses two or three of those systems — and waits. Fintelli AI answers them in seconds, investigates in minutes, and reconciles CPQ → Billing → REBS → GL with a SOX‑safe audit trail.

~5sper Chat question · ~$0.01 / question
~10sper Analyze BRR · ~$0.05–$0.10 / analysis
5–10 minper Investigation · ~$0.15+ / run · cited
~$0.02per DQ break narrated · ~1,000 orders scanned in seconds
Chat

Data Chat — plain‑English Q&A

Ask a question. The agent generates SQL, runs it, and answers — with the SQL expanded for audit. A semantic model translates business vocabulary ("open invoice", "at‑risk account") into physical columns and joins before SQL is generated, so terms resolve consistently across proprietary schemas.

  • Business term → column mapping (no LLM guessing on your schema)
  • Value dictionaries (status=1 → 'unpaid') so predicates read cleanly
  • Declared joins — eliminates cross‑product bugs that silently inflate numbers
  • Every turn shows tokens, cost, and how many SQL statements ran
Insights

Analyze — narrative‑first Business Reviews

Give the agent a topic. Sonnet plans 3–5 sub‑questions and issues them in parallel; Haiku generates each SQL; a synthesizer writes the narrative first, then renders KPI tiles and charts. The story drives the evidence — not the other way around.

  • Plan‑first, execute‑many, synthesize‑last agent loop
  • Every KPI tile expands to show its generating SQL — auditable end‑to‑end
  • Saved analyses persist per user with run ID — compare this week's BRR to last week's
  • Best for weekly business reviews, exec summaries, exploring a metric openly
Investigate

Investigate — hypothesis‑tested root cause

A 7‑node LangGraph state machine. Sonnet 4.6 plans 3–5 root‑cause hypotheses; Haiku 4.5 writes and executes SQL per hypothesis in a loop; a critic gates the final synthesis so a confident report only writes when evidence is solid; Sonnet writes the cited report.

  • Model tiering per node — Opus/Sonnet where reasoning depth matters; Haiku for mechanical SQL
  • Circuit breakers: budget cap (95% of MAX_COST_USD_PER_RUN), SQL error streak, max iterations (8)
  • HITL interrupt on ambiguous critique — run saved; Resume replays from LangGraph checkpoint (no re‑bill)
  • Runs are pinnable and comparable — track how a metric evolves
Anomaly Feed

Auto‑Detect — three‑model ensemble consensus

A SARIMAX + Prophet + XGBoost ensemble runs over every registered (metric, period) time series. A period is flagged only when at least two of the three models agree — consensus cuts the false‑positive rate significantly without needing a labelled anomaly dataset.

  • SARIMAX catches stable drift and seasonal breaks; misses sudden shocks
  • Prophet catches structural changepoints and holiday spikes; misses smooth non‑event trends
  • XGBoost catches multi‑factor non‑linear anomalies; misses pure seasonality — together they cover blind spots
  • One click hands the item to Investigate with metric + period + observed + expected pre‑filled
Data Quality

Q2C Tie‑Out & Reconciliation

The same dollar of revenue lives in four systems — CPQ, Billing, the Revenue Sub‑Ledger (REBS), and the General Ledger. They drift apart constantly. Fintelli automates the reconciliation discipline; a human approves every correction that touches the books.

  • Nine break categories with deterministic detectors and a fixed remediation menu
  • Duplicate · Orphan (CPQ / Bill / REBS / GL) · Timing Lag · Tax Mismatch · Partial Recognition · Unknown Diff
  • Auto‑cleared bucket: differences inside tolerance (max $1 or 0.5%) logged as noise, not breaks
  • SOX‑safe: LLM never emits a dollar figure — deterministic tools compute all amounts
FP&A

Rolling forecasts that auto‑explain their own variance

A modular subgroup for the monthly finance rhythm: rolling forecasts for Revenue, Expenses, and Headcount that keep an immutable snapshot history and auto‑generate a written variance narrative when actuals miss the forecast by more than a per‑metric threshold.

  • Ensemble forecast (SARIMAX + Prophet + XGBoost) with 80/95% CI bands; snapshots persisted per run
  • Bridge buckets: Base · Expected growth · Volume · Rate · Mix · Timing
  • Auto‑narrative (Haiku, RAG‑seeded from prior narratives) inherits the team's voice
  • One‑click deep‑dive escalation to the full Investigate agent — same cited root‑cause report

Fintelli AI · Beyond Q2C

P2P, R2R & Treasury modules.

The same investigation engine, HITL controls, and audit trail — extended across procure‑to‑pay, record‑to‑report, and treasury. Every module is agent‑led, cited, and human‑approved.

P2P

Procure‑to‑Pay AI

Catch duplicates before payment; score vendor risk; automate 3‑way match.

4 modules
  • Duplicate Invoice Detector — deterministic matching with LLM narration of why each suspected pair is suspicious
  • Vendor Risk Scoring — composite 0–100 (geopolitical · financial · concentration · quality)
  • 3‑Way PO Match Agent — PO / Goods Receipt / Invoice reconciliation with variance narratives
  • Payment Terms Optimizer — early‑pay discount vs. cash‑flow trade‑off, per‑vendor recommendations
R2R

Record‑to‑Report AI

Close the books faster and defensibly — anomaly detection, tie‑out, flux narrative.

5 modules
  • JE Anomaly Detector — 7‑signal detection scoring ≥ 70; ~6 hrs of manual review saved per close
  • Subledger ↔ GL Reconciliation — CPQ → Billing → REBS → GL tie‑out with break classification
  • Close Checklist — task orchestration for month / quarter close with owner & SLA tracking
  • Manual JE Risk (SOX) — reviewer risk scoring on high‑value manual journal entries
  • Flux Commentary (RAG‑Seeded) — auto variance narrative from prior‑period templates, in the team's voice
TRE

Treasury AI

13‑week cash forecast, FX exposure matrix, hedge advisory, working‑capital levers.

4 modules
  • Cash Forecast (13‑Week Rolling) — SARIMAX AR projections with scenario overlays
  • FX Exposure (Entity × Currency × Tenor) — exposure matrix by legal entity with tenor bucketing
  • Hedge Advisor (HITL‑Gated) — hedge recommendations pre‑vetted for board‑defensible decisions
  • Working Capital Optimizer — AP · AR · inventory levers with cash‑release estimates

Entelli AI · Supply Chain deep dive

39 signals. 8 domains. One autonomous SC Ops Copilot.

AI‑powered decision support across demand planning, inventory optimisation, shipping, contract manufacturing, costing, WIP, bill‑of‑materials, and data quality. Every recommendation is routed through a human‑in‑the‑loop gate; every action is auditable. HITL responses — Approve, Defer, Dismiss — are logged to an append‑only audit trail that feeds the Admin governance page.

01

Demand & NPI

Eliminate Excel forecasting; detect demand deviations early.

3 signals
  • Demand Planner — ML statistical 12‑month rolling forecast by SKU; HITL approve/adjust
  • Consensus Forecast — aggregates statistical forecast with sales/finance overrides + reason codes
  • NPI Forecast — new‑product forecast via top‑3 analog SKU matching + editable ramp curve
02

Inventory

Free trapped cash; right‑size safety stock; cut write‑offs.

4 signals
  • ABC × XYZ Segmentation — 9‑cell matrix; revenue contribution × demand variability
  • Safety Stock Optimizer — from demand history, lead time, target service level (default 95%)
  • SLOB Detector — slow‑moving / obsolete score 0–100 (flag ≥ 70); dead = 9+ months no movement
  • Excess Redistribution — inter‑site / channel transfers with cost/benefit analysis
03

Shipping

Recover freight overcharges; flag late deliveries proactively.

4 signals
  • Carrier Selection — multi‑carrier cost vs. transit matrix with lane scoring narrative
  • Freight Audit — invoice vs. contract rate; flags >2% tolerance; auto credit‑claim draft
  • OTIF Analyzer — On‑Time In‑Full waterfall by carrier and lane (default target 95%)
  • ETA Predictor — predicted vs. actual per shipment; detects carrier drift
04

Contract Manufacturing

Score CMs; catch capacity squeezes; detect yield drift.

4 signals
  • CM Scorecard — trailing 12‑mo quality / cost / delivery heatmap with composite score
  • Capacity Signal — utilisation vs. order load; flags >85% with risk summary + actions
  • Yield Variance (SPC) — Statistical Process Control charts per CM/product; min yield 90%
  • CM Risk — composite geopolitical / financial / quality / concentration score (flag ≥ 70)
05

Costing

Keep standard costs current; detect cost spikes before close.

4 signals
  • Landed Cost — material + duties + freight + handling roll‑up with prior‑period compare
  • PPV Analyzer — Purchase Price Variance waterfall (rate / mix / FX); reclass proposals
  • Standard Cost Roll — full BOM cost explosion, multi‑level with labor and overhead
  • Item Cost Anomaly — Z‑score on item cost time series (flag |Z| ≥ 2.5) with context narrative
06

Work‑In‑Process

See inside the factory; surface stalled jobs and scrap trends.

4 signals
  • WIP Aging — days‑in‑process by work centre; at‑risk jobs + root cause narrative
  • Cycle Time & Drift — actual vs. standard by operation; systematic slowdown detection
  • Scrap Trend — by material / operation / work centre; LLM narrative with root‑cause hypotheses
  • Job Cadence — work order release vs. capacity plan; batch bunching + starvation patterns
07

Bill of Materials

Single BOM source of truth; simulate ECR impact before sign‑off.

4 signals
  • BOM Viewer — multi‑level exploded parent → assembly tree; filterable by level / revision
  • Where‑Used Analyzer — reverse lookup: which finished goods use a given component
  • Change Impact Simulator — component swap: cost delta, lead‑time delta, cascade effects
  • BOM Change Approval (ECR) — AI‑populated impact queue; sequential / parallel multi‑approver
08

SC Data Quality

Catch data corruption that distorts every downstream report.

12 signals
  • Chain Reconciliations — 4 checks across sourcing, receiving, and inventory movements
  • Master Data Hygiene — 4 checks on item master, vendor master, and location data
  • Integrity Checks — 4 checks on BOM structure, routing, and cost coherence
  • All feed the SC Ops Copilot and the SC Data Quality tile on the executive brief

Production‑grade foundations

Evaluation HarnessStructured LoggingPhoenix TracingAppend‑Only Audit TrailAI Guardrails (HITL · prompt sanitize)Cost Analytics

Product Foundations

Built to survive audit. Deployed on your infrastructure.

✦

Agentic Investigation Engine

7‑node LangGraph state machine — plan hypotheses, execute SQL, critique evidence, cite the report. Model tiering per node keeps cost low without losing rigor.

  • Sonnet · Haiku · Opus routing per node
  • Circuit breakers: budget cap, SQL streak, max iterations
  • HITL interrupt + LangGraph checkpoint resume
◈

Three‑Model Anomaly Ensemble

SARIMAX + Prophet + XGBoost run in parallel over every registered time series. A period flags only when ≥2 of 3 agree — consensus cuts false positives without a labelled dataset.

  • Sparkline + per‑model vote breakdown
  • One‑click hand‑off to Investigate
  • Batch investigations for N flagged periods
▲

Q2C Tie‑Out (SOX‑Safe)

Reconciles CPQ → Billing → REBS → GL with nine deterministic break types. LLM never emits a dollar figure; the auditor gets a clean append‑only chain of evidence.

  • Duplicate · Orphan · Timing Lag · Tax · Partial
  • Deterministic remediation menu, HITL gate
  • Corrections log — JSONL, tamper‑evident
◆

Semantic Model + Read‑Only Data

Business term → column mapping resolves consistently across proprietary schemas. Runs read‑only on your infrastructure via DuckDB; no data movement.

  • Entity · join · synonym catalog
  • Value dictionaries for readable predicates
  • Connectors: SFDC · Zuora · NetSuite · Oracle EBS · SAP
●

Observability & Cost Governance

Phoenix traces per node, per‑user cost log per capability, and hard spend caps that short‑circuit any run at 95% of the budget.

  • Full audit trail (question · SQL · rows · answer)
  • MAX_COST_USD_PER_RUN circuit breaker
  • Per‑user, per‑capability cost analytics
◇

Multi‑Tenant + RBAC + HITL

Tenant isolation, role‑based access, OAuth 2.0 / OIDC, and human‑in‑the‑loop gates on every recommendation. AI proposes, humans approve — no autonomous write‑backs.

  • Approve · Defer · Dismiss with reason codes
  • Append‑only decision log
  • PII prompt sanitisation before any LLM call

How we work

Small, senior, and outcome‑anchored.

  1. 01

    Discover

    A short, focused engagement to map the data estate, the real business question, and the shortest credible path to value.

  2. 02

    Design

    Reference architecture, KPI contracts, and a delivery plan that names owners, dates, and success metrics.

  3. 03

    Build

    Production‑grade pipelines, models, and dashboards — versioned, tested, and instrumented from day one.

  4. 04

    Hand off

    Runbooks, docs, and a mentored team. We leave when your people can extend the platform without us.

About Entelli AI

Agentic AI, engineered for enterprise finance.

Entelli AI is the agentic investigation engine that powers Supply Chain, Q2C, Core Finance, and Strategic Finance workflows on your own infrastructure. Every recommendation is cited, every dollar is computed deterministically, and every action is human‑approved before it touches the books.

Entelli AI is a product of ViTec Business Solutions, Inc. — a boutique data and AI studio whose senior team has led platforms at global technology enterprises. Our roots are in enterprise finance analytics, which is why we're opinionated about governance, KPI discipline, and auditability from the first commit.

Work with us

Contact

Tell us what you're trying to figure out.

A short note is enough. We reply within one business day.

By sending, you agree to be contacted by Entelli AI, a product of ViTec Business Solutions, Inc.