Overview · positioning
This isn't a dashboard. It's an agentic decision system grounded in a business ontology.
Reporting
What happened?
→
Analytics
Why did it happen?
→
Agentic
What should we do — and who owns it by when?
You are here
Section 1 · The difference
Management reporting vs. an agentic decision system
Management reporting
"Show me last quarter."
- DirectionRetrospective
- OutputA chart, a table, a PDF
- KnowledgeRows in a fact table
- AI roleOptional caption / nice-to-have
- AudienceOne report for everyone
- ActionImplied — left to the reader
- DrillFilter / pivot
- StateRefreshes nightly
VS
Agentic decision system
"What should we do, framed for the board?"
- DirectionForward-looking · recommends
- OutputImplication + evidence + actions w/ owner & date
- KnowledgeTyped ontology + linked KB graph
- AI rolePrimary reasoning agent · grounded
- AudienceSame KB, read through CEO / CFO / COO personas
- ActionExplicit · owner · by-when · ties to a decision
- DrillClick any evidence chip → back into the graph
- StateListens to news + signals · re-tags impact live
The litmus test
Ask the system: "What does PSA mean for Avery Dennison in the next 90 days? Cite specific evidence and recommend one action."
A dashboard can't answer that. An agent grounded in a business ontology can — because PSA isn't a string, it's an entity linked to a domain (MG.POLY), a value-chain stage (Inbound), a category (Materials), customers, KPIs and owners. The agent traverses those links, then a CEO-persona writes the answer.
Section 2 · The agent loop
How ontology + KB + taxonomy turn a question into a grounded answer
Each step below is a real component in the app. The grey boxes are knowledge artifacts (static, curated). The blue boxes are agent operations (live, per-question). The orange box is the LLM call. The arrows show how grounding flows in — that's what stops the model from hallucinating.
①
User question
"Frame the Walmart ramp for the next earnings call."
↓
②
Intent + entity extraction
Pulls Walmart, earnings call, Solutions Group out of the prompt
↓
⇲
③a
Ontology lookup
62 typed terms · finds RFID, atma.io, HVC; resolves RBIS = Solutions Group
Open ontology →
③c
Taxonomy / typing
Knows Walmart is a Customer, RFID is Tech, EVA is Finance — so the agent reasons at the right level
Open domains →
⇱
↓
④
Grounded context assembled
Definitions, current values, owners, related entities — packed into the prompt so the model can only cite real facts
↓
⑤
LLM reasons as a persona
CEO · CFO · COO · CHRO — same KB, different voice and decision lens
Meet the copilots →
↓
⑥
Structured agentic answer
Implication
Evidence (clickable chips → back into KB)
Actions · owner · by-when
Follow-up questions
↻
User drills any evidence chip back into step ③ — the system is a loop, not a pipeline. Drilling enriches the next question.
Section 3 · The grounding stack
Three knowledge layers that make AI act like an agent, not a chatbot
01
Taxonomy
The typing system
17 categories — Brand, Person, Materials, Tech, KPI, Regulation, …
Why it matters for agentic AI:
- Without it, the LLM treats "RFID" and "Walmart" as the same kind of string.
- With it, the agent knows to look up financial impact for a KPI, resolve owner for a Person, check regulation date for a Regulation.
02
Ontology
The vocabulary
62 typed terms with definitions, domain refs, and a "what this means for AVY" implication.
Why it matters for agentic AI:
- Stops jargon hallucination — the LLM quotes the curated definition rather than inventing one.
- Enables semantic search: ask "EU compliance terms" → ontology returns DPP, ESPR, Scope 1/2/3, even though none contain the word "EU".
03
Knowledge base / graph
The relationships
9 domains, sub-domains, customers, competitors, KPIs, value chain, journeys — all linked.
Why it matters for agentic AI:
- Lets the agent traverse: "Walmart" → Top Customer → SG segment → Francisco Melo → atma.io → DPP impact.
- Every answer can cite a path through the graph; every claim is back-traceable.
Taxonomy
types the data
+
Ontology
defines the terms
+
KB / graph
links the entities
=
Agentic AI
that can reason, cite, and act
Section 4 · What it unlocks
Six things only work because of the grounding stack
①
Acronym-aware answers
"What does PSA mean for AVY in the next 90 days?"
Ontology resolves PSA → Pressure-Sensitive Adhesive · domain
MG.POLY · category Materials. Agent cites that definition verbatim and recommends an action grounded in the real product portfolio.②
Persona-switching on one KB
"Frame the Walmart ramp for the next earnings call." — asked of CEO, then CFO, then COO.
Same facts, three lenses: CEO talks growth narrative, CFO talks revenue uplift & margin, COO talks rollout risk. No three databases — one ontology, three personas.
③
Semantic, not keyword
"Show me terms tied to EU compliance."
Returns DPP, ESPR, Scope 1/2/3, SBTi — none of which contain "EU". LLM ranks by meaning against the typed ontology, not a substring match.
④
Cross-entity comparison
"Compare Smartrac vs TexTrace on the differentiation lens."
Both resolve to acquired-brand entities under
SG.RFID / SG.EMBEL.TEX. Graph traversal pulls year, parent domain, and overlap — then the LLM scores the comparison.⑤
Action with ownership
"Recommend one action on HVC margin."
Agent returns: "Lift HVC mix from 40 → 45% in Wine & Spirits" — Owner: SVP & GM LPM · By: end of Q3. Owners come from the domain hierarchy, not invented.
⑥
Live signals tagged to entities
DDGS news feed → ontology auto-tag → cockpit ticker.
A headline about "Walmart RFID expansion" is auto-routed to Walmart (Customer), RFID (Tech), SG.RFID (Domain) — and surfaces on the relevant 360 page, not in a generic feed.
Section 5 · Where the value lands
Three pillars for Data & AI big bets — and how agentic adds value to each
Every agent in the registry must trace its value to one of these three P&L levers. If it doesn't move revenue, monetize data, or expand margin, it doesn't ship.
AD value chain → where each pillar lives
R&D · NPI
HVC labels · RFID inlays · CleanFlake · DPP substrates
Pillar 01
→
Make
MG coaters · slitters · SG inlay lines · 50+ plants
Pillar 03
→
Sell · Serve
Diageo · L'Oréal · Walmart · Inditex · Pfizer
Pillar 01
→
Connected items
atma.io · RFID · QR · DPP events
Pillar 02
Pillar 01 · Top line
Revenue Growth
NPI for HVC labels, RFID inlays & sustainable substrates · R&D acceleration
Brand brief
→
DDGS signal
→
Ontology match
→
NPV rank
→
Pilot launch
6 wks
vs 6 mo today
Brief → pilot compression
⚙ The lever — for AD
⚡
Compress NPI
Brief → pilot → launch on HVC W&S, beauty, pharma labels
🛡
De-risk R&D bets
RFID Gen-3 · CleanFlake (PET-recyclable PSA) · DPP substrates
🌐
Open whitespace
Walmart · Inditex · H&M apparel → food, pharma, beauty
◆ Agents on the job
01 NPI Scout
Unmet jobs from Diageo · L'Oréal · Pfizer — CRM + DDGS + win/loss
02 R&D Triage
Weekly rank: RFID Gen-3 vs CleanFlake vs HVC PSA by NPV × fit
03 Voice-of-Customer
Clusters tickets & calls — "liner waste", "fast-fashion velocity"
04 Patent Watch
Flags Smartrac · TexTrace · Checkpoint filings vs SG.RFID / MG.POLY
▶ AD case in action
- 1Diageo brief — tamper-evident HVC label
- 2NPI Scout finds parallel asks from Pernod Ricard & LVMH Wines
- 3R&D Triage lifts NPV · VoC attaches live counterfeit incidents
- ★Pilot in 6 weeks — instead of 6 months
Pillar 02 · New revenue
Data Monetization
Turn AD's full data estate — connected items, materials science, plant & market signal — into products
Data source
→
Consent check
→
Aggregate
→
Narrative
→
Buyer API
4 SaaS lines
net-new — from data we already have
DPPaaS · indices · APIs · narratives
⚙ Raw material → shippable product
AD data estate
📡
Connected-item eventsatma.io · RFID · QR · DPP
🧪
Materials & sustainabilityCleanFlake · lifecycle · carbon
🏭
Plant & processyield · scrap · OEE · 50+ plants
💼
Market & pricingquote · win-rate · channel · CRM/ERP
→
Products that ship
📊
Industry indicesApparel Velocity · W&S Auth · Pharma Cold-Chain
🌱
Sustainability dashboardsPSA labels · packaging · carbon
📜
DPPaaS subscriptionapparel · footwear · EV battery · cosmetics
🔌
Data-as-a-Service APIsbrands · retailers · regulators · insurers
◆ Agents on the job
01 Data Product
Assembles, governs & prices a product from any source — atma.io, lab, MES, CRM
02 Benchmark
Anonymizes across customer × plant × SKU into sellable indices
03 Contract & Consent
Enforces purpose-of-use, brand consent, EU residency & rev-share
04 Insight-as-a-Service
Narrative answers from any product to brand · retailer · regulator buyers
▶ AD case in action
- 1L'Oréal sustainability buyer asks for packaging carbon benchmark across PSA families
- 2Contract & Consent validates AVY's right to publish anonymized dataset
- 3Benchmark aggregates lifecycle data from materials lab + MG plants · IaaS wraps narrative + API
- ★Net-new sustainability subscription ships in days — no atma.io scan needed
Pillar 03 · Bottom line
EBITDA Margin Improvement
Coater yield · slitter throughput · drying-oven energy · SKU-tail profitability
MES sensor
→
Ontology resolve
→
Agent rec
→
Operator action
→
Telemetry back
$180k/mo
scrap avoided · per line
Mentor Line 7 · W&S coating run
⚙ The lever — for AD
🎯
Coating yield
First-pass quality on PSA-coated webs · edge & defect scrap
⏱
Throughput
Coater line speed · slitter changeover · SG inlay OEE
⚡
Cost efficiency
Drying-oven energy · freight on heavy rolls · liner working capital
📉
Cost-driver
30k+ SKUs · customer profitability · plant-by-plant unit cost
◆ Agents on the job
01 Yield Optimizer
Setpoint shifts (oven temp, web tension, coat wt) on Mentor OH Line 7
02 OEE / Throughput
Root-causes changeover downtime between W&S SKUs on slitter lines
03 Cost-Driver
COGS by SKU × plant × customer · flags negative-margin HVC tails
04 Procurement & Energy
Hedges paper · silicone · adhesive · reroutes freight on fuel signals
▶ AD case in action
- 1Yield Optimizer spots 60% scrap spike on Mentor Line 7 during W&S coating
- 2Cross-checks recipe v3.2 in the KB · simulates oven-temp + tension changes
- 3Sends shift supervisor a one-click fix
- ★Yield recovers in 2 shifts · ~$180k/mo scrap avoided on that line
SIPOC — how each pillar runs, end-to-end
Same lens, three pillars: Suppliers · Inputs · Process (the agents) · Outputs · Customers. Each row is what the agent fleet actually consumes and produces — not abstract, all AD-real.
⬢S · Supplierswho feeds us
→
▤I · Inputswhat flows in
→
◆P · Processthe agent fleet
→
★O · Outputswhat comes out
→
◉C · Customerswho consumes
01
Revenue
Growth
Growth
NPI & R&D acceleration
Owner · CGO
⬢ Suppliers
- Diageo
- L'Oréal
- Pfizer brand teams
- Salesforce CRM
- DDGS news feed
- USPTO / EPO
- Win-loss notes
→
▤ Inputs
- Brand briefs
- Sell-through signal
- Competitor filings
- Support tickets
- NPS verbatims
→
◆ Process · Agents
- 1 NPI Scout
- 2 Voice-of-Customer
- 3 R&D Triage
- 4 Patent Watch
→
★ Outputs
- Ranked NPI pipeline (NPV × fit)
- Pilot brief
- IP-risk flag
- Prioritized feature set
→
◉ Customers
- R&D
- Product Marketing
- MG / SG GMs
- Account teams
- CGO
💡
Brand signals in → ranked, IP-cleared NPI pipeline out
02
Data
Monetization
Monetization
atma.io as a product
Owner · CDO
⬢ Suppliers
- atma.io platform
- Brand owners (consent)
- Retailers (sell-through)
- EU regulators (DPP / ESPR)
→
▤ Inputs
- Serialized RFID / QR scan events
- Brand consent records
- Contract terms
- Residency rules
→
◆ Process · Agents
- 1 Contract & Consent
- 2 Benchmark
- 3 Data Product
- 4 Insight-as-a-Service
→
★ Outputs
- DPP-as-a-Service
- Apparel Velocity Index
- Authentication API
- Narrative insights
→
◉ Customers
- Brand owners
- Retailers
- Sustainability / ESG buyers
- Regulators
- Partner CDOs
💡
Scan events in → priced data products & compliance services out
03
EBITDA
Margin
Margin
Yield · throughput · cost
Owner · COO / CFO
⬢ Suppliers
- AVEVA Historian
- Rockwell PLC
- SAP CO / PP
- BW
- Coupa
- Commodity & freight feeds
→
▤ Inputs
- Process tags (temp, tension, coat-wt)
- Downtime codes
- COGS bridge
- Paper / silicone / adhesive prices
→
◆ Process · Agents
- 1 Yield Optimizer
- 2 OEE / Throughput
- 3 Cost-Driver
- 4 Procurement & Energy
→
★ Outputs
- Setpoint rec. (oven temp · tension)
- Downtime root cause
- Negative-margin tail
- Hedge plan
→
◉ Customers
- Shift supervisors
- Plant managers
- COO
- CFO
- CPO
💡
Plant + market signals in → setpoints, root causes & hedges out
Agent collaboration — how the fleet hands off across pillars
Real value compounds when an insight in one pillar triggers an action in another. SCIKIQ's A2A bus lets agents post events on the shared ontology — any other agent subscribed to that entity reacts. Below are the live cross-pillar handoffs.
Revenue · Pillar 01
NPI Scout
R&D Triage
Voice-of-Customer
Patent Watch
Data $ · Pillar 02
Data Product
Benchmark
Contract & Consent
Insight-as-a-Service
EBITDA · Pillar 03
Yield Optimizer
OEE / Throughput
Cost-Driver
Procurement & Energy
Cross-pillar A2A handoffs (live in SCIKIQ)
Patent Watch
→
R&D Triage
Smartrac files a competing RFID inlay patent — Triage re-ranks SG.RFID Gen-3 ahead of CleanFlake this sprint.
Voice-of-Customer
→
Data Product
Three brands ask for "wine authentication" data — Data Product packages a W&S Authentication Index from atma.io.
Benchmark
→
NPI Scout
Apparel sell-through spikes in fast-fashion velocity — Scout opens a brief for higher-density RFID inlays.
Yield Optimizer
→
Cost-Driver
Mentor Line 7 scrap drops 18% — Cost-Driver re-bridges COGS and flags HVC unit margin uplift to CFO.
Cost-Driver
→
R&D Triage
Negative-margin tail in low-volume HVC variants — Triage moves SKU-rationalization R&D items up the queue.
Contract & Consent
→
Insight-as-a-Service
Brand consent & EU residency green-lit — IaaS exposes the Carrefour-DPP API and starts metering.
Procurement & Energy
→
R&D Triage
Acrylic adhesive cost up 9% on hedge curve — Triage elevates CleanFlake (less adhesive load) in the pipeline.
Section 6 · Agent map & registry
SCIKIQ Agent Mesh — registry, identity & control plane for the agent fleet
The SCIKIQ Agent Mesh treats every agent as a first-class workforce identity — registered, discoverable, governed and observable. Every agent is mapped to a pillar, owned by a persona, grounded in the ontology, and wired into the same guardrails as a human user.
SCIKIQ
Agent Registry & Control Plane
IdentityCatalogPolicyTelemetry
Pillar 01 · Revenue Growth
NPI Scout
R&D Triage
Voice-of-Customer
Patent Watch
Pillar 02 · Data Monetization
Data Product
Benchmark
Contract & Consent
Insight-as-a-Service
Pillar 03 · EBITDA Margin
Yield Optimizer
OEE / Throughput
Cost-Driver
Procurement & Energy
Connects to
- M365 / Copilot · Teams, Outlook, Excel, SharePoint
- ERP & MES · SAP, Oracle, Rockwell, AVEVA
- CRM · Salesforce, Dynamics
- Data plane · Fabric, Snowflake, atma.io
- Other agents · A2A handoff via shared ontology
Agent registry — like a workforce directory, for agents
| Agent | Pillar | Owner persona | Grounding | Tools / systems | Trigger | Guardrails | Status |
|---|---|---|---|---|---|---|---|
| NPI Scout | Revenue | CGO | Ontology · Customer 360 · DDGS | CRM · DDGS · KB | Weekly + event | Obs · Gov · Sec | Live |
| R&D Triage | Revenue | CTO | R&D pipeline · IP graph | PLM · Patents API | Weekly | Obs · Gov · Sec | Live |
| Voice-of-Customer | Revenue | CMO | CRM notes · Tickets · NPS | Salesforce · Zendesk | Daily | Obs · Gov · Sec | Live |
| Patent Watch | Revenue | CTO | IP ontology · Competitor map | USPTO · EPO feeds | Daily | Obs · Gov · Sec | Pilot |
| Data Product | Data $ | CDO | Data catalog · Contracts | Fabric · atma.io | On request | Obs · Gov · Sec | Live |
| Benchmark | Data $ | CDO | Anonymized aggregate KB | Snowflake · DP layer | Monthly | Obs · Gov · Sec | Live |
| Contract & Consent | Data $ | CLO | Consent ledger · Purpose-of-use | OneTrust · Vault | Every call | Obs · Gov · Sec | Live |
| Insight-as-a-Service | Data $ | CDO | Data products + ontology | API gateway · Stripe | On request | Obs · Gov · Sec | Pilot |
| Yield Optimizer | EBITDA | COO | MES · process tags · recipe lib | AVEVA · Historian | Per shift | Obs · Gov · Sec | Live |
| OEE / Throughput | EBITDA | COO | Downtime codes · scheduling | Rockwell · SAP PP | Per shift | Obs · Gov · Sec | Live |
| Cost-Driver | EBITDA | CFO | COGS bridge · SKU × plant | SAP CO · BW | Weekly | Obs · Gov · Sec | Live |
| Procurement & Energy | EBITDA | CPO | Commodity index · contracts | Coupa · Energy APIs | Daily | Obs · Gov · Sec | Pilot |
Every row in the registry is an identity in SCIKIQ — provisioned, scoped, observable, revocable. Just like an employee record.
Section 7 · End-to-end protection
Guardrails for every agent and every user — Observability, Governance, Security
Agents are workforce. Treat them like one — with identity, policy, audit, and a kill switch. SCIKIQ wraps every agent in the registry with the three rings below, so the same controls apply whether the requestor is a human or an agent acting on a human's behalf.
Observability
See what every agent does, in real time
- Per-agent telemetry — calls, latency, tokens, cost
- Decision trace — prompt → grounding → tool calls → output
- Drift & quality — eval scores against gold answers
- Health dashboard — green / amber / red by agent & pillar
- Issue resolution — replay any session, diff against baseline
Governance
Who's allowed to do what, with which data
- Agent registry — owner, pillar, purpose-of-use on file
- Policy engine — RBAC, data classification, residency
- Approval workflows — high-impact actions need human sign-off
- Audit log — immutable record of every agent action
- Lifecycle — onboard / promote / retire like a person
Security
Protect the agent, the data, and the action
- Workforce identity — every agent has its own credential
- Least-privilege tools — scoped tokens per system
- Prompt-injection defense — input/output filtering & isolation
- Secrets & PII redaction — at ingress and egress
- Kill switch — revoke any agent in one click
Security
Governance
Observability
Agent
grounded in ontology + KB
End-to-end protection
Every request — from a human or another agent — passes through all three rings. Nothing reaches a tool, a system, or the LLM without identity, policy, and trace attached.
- Inputauth · prompt-injection scrub · PII redact
- Reasoninggrounded prompt · cited evidence only
- Actionscoped token · policy check · approval if needed
- OutputPII filter · audit log · telemetry emit
Section 8 · Demo
See the SCIKIQ Agent Mesh in action — a request flowing through the registry, guardrails, and back
Follow one real ask — "Why is HVC margin slipping in Wine & Spirits?" — as it traverses identity, the agent map, grounding, tools, guardrails, and telemetry.
1
User asks
CFO persona · authenticated · SSO + MFA
"Why is HVC margin slipping in Wine & Spirits — and what should I tell the board?"
→
2
Guardrail check
SCIKIQ policy engine
- Identity verified ✓
- Data scope: WS-Finance ✓
- Prompt-injection scrub ✓
- Audit log opened ✓
→
3
Agent map routing
Registry · pillar selection
- Cost-Driver agent (EBITDA)
- + Yield Optimizer (EBITDA)
- + Voice-of-Customer (Revenue)
- A2A handoff via ontology
→
4
Grounded tool calls
Scoped tokens · per agent
- SAP CO → COGS bridge by SKU
- AVEVA → scrap rate, last 8 wks
- Salesforce → top-5 WS accounts
- Ontology → HVC, PSA, LPM
→
5
Structured answer
CFO-persona narrative
Implication: −180 bps, driven by 60% scrap on Line 7
Action: shift Line 7 recipe v3.2 · Owner: SVP LPM · By: end of Q3
Evidence: SAP, AVEVA, CRM chips → drill back
…meanwhile, observability streams the whole run
Latency
8.4s
P50 across 3 agents
Tokens
42k
in · 6.1k out
Cost
$0.31
per answer
Eval
0.94
vs gold answer
Policy
0 deny
all checks passed
Audit
trace #4821
replayable
Same pattern for any of the 12 agents in the registry — one ontology, three pillars, three guardrails, end-to-end traceable.
Section 8 · Data architecture
From Oracle & Databricks → metadata twin → ontology → agents — with Gemini and a self-hosted vLLM safety lane
AD is an Oracle shop (Fusion ERP / SCM / HCM / EPM on OCI + Exadata) with Databricks as the analytics lakehouse and Google Gemini as the enterprise LLM (already rolled out to 22,000+ employees — AD has the world's largest Gemini install base). Partner-side forecast data flows in from Azure Databricks and Oracle ADW. For PII, IP, contract, and recipe data we add a self-hosted LLM served by vLLM on GPU so sensitive prompts never leave the AD perimeter. The glue is a metadata twin (technical + semantic) feeding a knowledge graph + ontology the agents reason over.
07Use cases
Revenue
NPI ranking · VoC mining · IP clearance · win-loss
Data $
DPP-as-a-Service · Apparel Velocity Index · Auth API
EBITDA
Yield optimizer · OEE root-cause · hedge plan · cost driver
▲
06Agent fleet
NPI Scout
VoC
R&D Triage
Patent Watch
Contract & Consent
Benchmark
Data Product
IaaS
Yield Optimizer
OEE
Cost-Driver
Procurement & Energy
▲
05LLM routing
safety-gated
safety-gated
Path A · Default
Google Gemini Enterprise
Used when: general reasoning, brand briefs, summarization, code, marketing copy, public NPI ideation, supplier news triage
21 production GenAI pilots · 10 hrs/employee/month saved · India plant predictive-maintenance pilot cut unplanned downtime 25%
⬥
Policy router
per prompt
per prompt
- PII / employee data → vLLM
- Recipe / coating IP → vLLM
- Contract / pricing → vLLM
- EU residency (DPP) → vLLM
- Everything else → Gemini
Path B · Safety lane
Self-hosted LLM · vLLM on GPU
Used when: PII, employee data, customer contracts, coating recipes, supplier pricing, DPP records under EU residency, RFID-tag-linked consumer data
Data never leaves AD perimeter · prompts & completions logged for audit · model weights pinned · no vendor egress
▲
04Knowledge graph
+ ontology
+ ontology
◆ Ontology · entities & relationships
Brand
owns
Product
made on
Line
at
Plant
Recipe
used by
Product
RFID Tag
scanned at
Retailer
Customer
buys
Product
Supplier
delivers
Raw material
consumed by
Line
~120 entity types · ~400 relationships · stored in Neo4j / OCI Graph · synced from Unity Catalog
▤ Taxonomy · business hierarchy
- MG · Materials Group
- Label & Graphic Materials (LGM)
- Wine & Spirits
- Beauty & Personal Care
- Pharma
- Industrial & Healthcare
- Label & Graphic Materials (LGM)
- SG · Solutions Group
- Intelligent Labels (RFID)
- Vestcom · Embelex · atma.io
Business hierarchy drives KPI rollups · agent scope · cost-center mapping
▲
▲
02Lakehouse
+ ingestion
+ ingestion
Ingest
Oracle GoldenGate (CDC)
Databricks Auto Loader
Kafka / Confluent
Fivetran (SaaS)
OPC-UA → MQTT
Medallion (Delta Lake)
🥉 Bronze
raw, append → 🥈 Silver
cleansed, joined → 🥇 Gold
KPI / mart
raw, append → 🥈 Silver
cleansed, joined → 🥇 Gold
KPI / mart
Govern & compute
Unity Catalog
Photon
DLT
MLflow
Vector search
▲
01Sources
systems of record
systems of record
◯ Oracle stack core ERP / SCM / HCM
Fusion ERP Cloud
Fusion SCM (RFID-linked)
Fusion HCM
Fusion EPM
Manufacturing Cloud
Procurement Cloud
OCI · Exadata
Autonomous DB
GoldenGate
◧ OT / Plant floor EBITDA pillar
AVEVA Historian
Rockwell PLC
SCADA · OPC-UA
MES
Energy meters
◨ Product & commercial
atma.io (RFID / DPP)
Salesforce CRM
Coupa
Workday-adjacent
PIM / PLM
◌ External signals
USPTO / EPO
DDGS news
Commodity / freight feeds
EU DPP / ESPR
Brand sell-through
◍ Partner forecast feed external · private
Azure Databricks (brand-side)
Oracle ADW (autonomous DW)
98M+ records
6 quarters
Weekly refresh
Worked example — how the stack solves "why did Mentor Line 7 scrap spike yesterday?"
1L1 Source · AVEVA Historian tag
L7.OVEN.TEMP + Oracle Manufacturing downtime code DT-204→
2L2 Lakehouse · Bronze ingest via GoldenGate + Auto Loader → Silver joins recipe v3.2 → Gold scrap-by-shift mart
→
3L3 Metadata twin · Technical: lineage + freshness OK · Semantic: "scrap %" = ((reject_kg / produced_kg) × 100), owner = COO
→
4L4 Ontology · Line → Plant → Recipe → Product walk; recipe v3.2 also runs on Lines 4, 9
→
5L5 LLM · No PII/IP in prompt → Gemini. Recipe-tuning prompt with raw formulation → router flips to vLLM
→
6L6 Agent · Yield Optimizer simulates oven-temp / tension setpoint shift, drafts one-click fix
→
7L7 Use case · Shift supervisor approves · scrap recovers in 2 shifts · ~$180k/mo avoided
Section 9 · Commercials
Pricing model — predictable license, pass-through infra, optional accelerator
Three clean line items. The platform license is the only recurring SCIKIQ fee. Infrastructure is consumed and billed directly by the client's cloud. Use-case delivery is an optional CapEx accelerator at go-live.
Recurring · OpEx
Platform License
SCIKIQ agent registry, control plane & standard support
- Term · 3-year or 5-year
- Includes · standard support, version upgrades, security patches
- Scope drivers · # use cases · # sources · # legal entities
- Billing · annual, in advance
Locks unit economics; supports multi-year CFO planning.
Pass-through · OpEx
Infrastructure
Cloud compute, storage, LLM inference — paid directly by client
- Owned by · client's cloud account (Azure / AWS / GCP)
- Covers · compute, storage, network, model inference, observability
- Why direct · no markup, full FinOps visibility, tenant isolation
- Optimization · SCIKIQ tunes cache, batch & model routing to keep cost down
Variable cost scales with usage — fully transparent on the client's bill.
One-time · CapEx
Setup & Use-Case Delivery
Infra setup + optional initial agent build-out
- Infra setup · landing zone, networking, identity, observability wiring
- Optional · initial use-case delivery (one or more agents from the registry)
- Booked as · CapEx, amortizable over license term
- Outcome · production-ready agents at go-live, not slideware
Sized per scope of go-live use cases — fixed-fee, milestone-based.
What every license includes — the SCIKIQ 4 Cs
C
Connect
Plug into any source — ERP, MES, CRM, data lake, files, APIs, feeds.
C
Curate
Build the ontology, KB and taxonomy that ground every agent in real facts.
C
Control
Observability, governance & security guardrails around every agent and user.
C
Consume
Use the agents — via personas, copilots, APIs and embedded experiences.
Licensing models — pick the shape that fits the org
Per Department
License scoped to a single function — e.g. Finance, Manufacturing, or R&D.
- Lower entry point · fastest TTV
- Sized by use cases & sources in scope
- Upgrades to enterprise without re-platforming
Best for: pilot land & expand within one P&L owner.
Most common
Enterprise-Wide
One license covering the full enterprise — all pillars, all departments.
- Unlimited departments under a single legal entity
- Tiered by # use cases & # sources configured
- Cross-pillar agent reuse (Revenue × Data × EBITDA)
Best for: CXO-sponsored, multi-pillar agentic transformation.
Per Legal Entity (LE)
One license per LE — for groups with many subsidiaries, JVs, or regions.
- Clean ring-fenced billing per LE
- Separate data residency & consent boundaries
- Volume tier across LEs at the group level
Best for: holding companies, multi-region groups, regulated entities.
What sizes the license
Use cases
#
agents activated from the registry — by pillar
×
Sources configured
#
systems wired in (ERP, MES, CRM, data plane, feeds)
×
Legal entities
#
if licensed per LE — otherwise enterprise-wide
=
License tier
$
3-yr or 5-yr fixed
Total cost stack at a glance
| Line item | Type | Who pays | Cadence | Includes |
|---|---|---|---|---|
| SCIKIQ Platform License | OpEx | Client → SCIKIQ | 3-yr / 5-yr · annual | Registry, control plane, standard support, upgrades |
| Cloud Infrastructure | Pass-through | Client → Cloud provider | Monthly · usage-based | Compute, storage, LLM inference, observability |
| One-time Setup | CapEx | Client → SCIKIQ | One-time · milestone | Infra setup; optional initial use-case delivery |
No hidden infra markup. No per-seat creep. License scales by what creates value — use cases, sources, entities.
TL;DR
A report tells you what happened.
An agentic system tells you what to do about it — grounded in the same facts you would cite yourself.
The ontology, KB and taxonomy are what make that grounding possible. Without them, the AI is a guessing chatbot. With them, it's a colleague.
An agentic system tells you what to do about it — grounded in the same facts you would cite yourself.
The ontology, KB and taxonomy are what make that grounding possible. Without them, the AI is a guessing chatbot. With them, it's a colleague.