ENTERPRISE INTELLIGENCE  ·  REV 1.0

The AI model isn't your advantage.
What it learns is.

Transform plant data, workflows and decisions into a proprietary learning layer that compounds over time, while remaining independent of any single model.

FIG. 01: SIX SOURCES, ONE LAYER
ERP MES CMMS SOPs SENSORS OPERATORS LEARNING LAYER
The Enterprise Problem

Your sensors are the eyes.
Nothing is the brain.

A vibration sensor can tell you a bearing is degrading. It can't tell a technician how to fix it, see a defect on the line, read a process drift, or raise the work order. That gap, not a lack of AI, is where most industrial intelligence programs stall.

ERPFinancial & operational systems of record
MESProcess execution, line-level data
CMMSWork orders, maintenance history
SOPs & ManualsInstitutional judgment, undigitised
SensorsVibration, process, energy signal
OperatorsTacit knowledge, shift handovers

SIX SOURCES. ZERO SHARED MEMORY.

01

Connect

SOPs, manuals, deviation records, work orders, process and energy tags, brought into the same picture as the sensor data you already run.

02

Model

An ISA-95 hierarchy, enterprise to asset, described once. Every document and work order hangs off it, inherited by every use case after.

03

Intelligence

Retrieval over your own plant knowledge, vision on your own defects, and cross-signal detection linking vibration with process and energy drift.

04

Act

Work orders, alert triage and setpoint recommendations, gated by human approval on every write. Advisory first, autonomous only when you choose.

Incremental, not a replacement

Your maintenance subscription doesn't change. This is bought with different money (the AI budget), and nothing here is contingent on the maintenance line.

A third of the work is already done

Most industrial AI stalls on data. You already have twelve months of signal history on sixty assets, that's why you start ahead of a greenfield site.

Priced per site, renewed annually

One site, one measured result, no multi-year lock. You decide again every year.

What Stays Yours

An SI leaves you a project. We leave you something you own.

This is the difference between hiring a consultancy and owning a learning layer. Three things sit inside your tenant boundary, not ours, and none of them leave if you switch models or switch vendors.

Plant Model

Your ISA-95 hierarchy

The asset, area, line and site structure built in Release 1 lives in your tenant. It's yours to keep, extend, and reuse, with us or without us.

Defect Library

Your labelled images

Every defect image used to train Vision QC, and every label attached to it, stays in your environment. It's a training asset you keep building.

Test & Eval Data

Your definition of "good"

The evaluations that measure whether an answer, a triage, or a recommendation was actually right are yours, and we work with any model, so you're never stuck if one gets pulled or repriced.

The Reverse Information Paradox

Every prompt teaches someone.
The question is: who keeps the lesson?

PROMPT MODEL MEMORY EVAL
Five Principles Of Ownership

What it means to own your intelligence

Five things every enterprise moving into the AI era has to hold onto, deliberately, not by accident.

C1

Control

Build your own private evals, they define what "good" means inside your organisation. Retain ownership of memory, traces, feedback and institutional context.

C2

Capability

Build proprietary learning environments inside your own tenant boundary, where models learn against real workflows without exposing your knowledge to anyone else.

C3

Choice

Keep the orchestration layer decoupled from any single model. If one is taken away tomorrow, your ability to operate against your own evals stays intact.

C4

Cost

A decoupled orchestration layer lets you combine context, models and tasks in whatever mix is most efficient, without ever trading away quality to get there.

C5 THE RESULT →

Compound

Bring the four together and you get a continuous learning loop, a hill-climbing machine that compounds the value of every AI investment you make.

Model Independence

Every model is a tenant.
None of them is the landlord.

GPT, Claude, Gemini, Llama: any of them can sit behind your orchestration layer. Swap one out and your evaluations, your memory and your workflows don't move. The generalist changes; the veteran capability you've built stays with you.

ORCHESTRATION CORE GPT CLAUDE GEMINI LLAMA
Industrial Proof: WogiAI

Your sensors already see it.
Now the plant can act on it.

WogiAI is the learning layer applied to a real plant floor, sitting on top of the vibration monitoring you already run, closing the gap between a sensor that detects a problem and a technician who knows what to do about it.

LIVE AT 3 MONTHS: CONNECT

Knowledge, handover, triage

+₹2.08 Cr
  • Plant knowledge assistant with citations
  • Shift handover automation
  • PdM alert triage
LIVE AT 6 MONTHS: EXTEND

Vision QC, cross-signal

+₹2.62 Cr
  • Vision QC across the lines
  • Cross-signal anomaly detection
  • Agentic work orders, human-approved
LIVE AT 12 MONTHS: OPTIMISE

Throughput, energy, scheduling

+₹2.36 Cr
  • Setpoint & process optimisation
  • Energy & emissions module
  • Scheduling agent

FIGURES REFLECT A SINGAPORE REFERENCE SITE: ARITHMETIC, NOT A PROMISE. THE LANDING ENGAGEMENT REPLACES EVERY INPUT WITH YOURS.

System Architecture

One model, described once. Inherited by everything after.

PlantSOURCE

The physical assets, lines and sensors your operators already run every shift.

Data SourcesERP · MES · CMMS · SENSORS

Every system of record you already have, connected as-is, nothing ripped out.

Semantic LayerISA-95 HIERARCHY

Enterprise, site, area, line and asset, described once so every layer above inherits it.

Learning LayerPROPRIETARY

The tenant-boundary layer where models learn against your real workflows, not a vendor's.

MemoryTRACES · CONTEXT

Every prompt, decision and outcome kept as institutional memory, not lost between sessions.

EvaluationsYOUR DEFINITION OF GOOD

Tests you write, scoring what "correct" means on your floor, not a generic benchmark.

OrchestrationMODEL-INDEPENDENT

Routes work to whichever model fits, so no single vendor becomes a dependency.

ModelsGPT · CLAUDE · GEMINI · LLAMA

Interchangeable behind the orchestration layer, swapped without touching what you've built.

AgentsADVISORY → AUTONOMOUS

Start as recommendations a human reviews, graduate to autonomous only when you allow it.

Human ApprovalON EVERY WRITE

Nothing touches a system of record without a person signing off first.

↻ APPROVED ACTIONS RETURN TO THE PLANT, THE LOOP CLOSES, THE LEARNING COMPOUNDS

The Rupee Value

Where every rupee comes from

We'd rather show the arithmetic than assert a number. Eight inputs about your site drive every line below: change one, and the whole model moves.

YEAR ONE: ALL IN
₹1.36 Cr
Landing engagement (₹36L) + platform fee (₹1 Cr/site/yr)
ANNUAL BENEFIT AT MONTH 12
₹7.06 Cr
RETURN: 5.2× in year one
One site, 6.3× return across three years
MTTR down 15%
CONNECT
₹1.26 Cr
Technician search time recovered
CONNECT
₹0.45 Cr
Planner productivity
CONNECT
₹0.22 Cr
PdM alert triage
CONNECT
₹0.15 Cr
Vision QC across lines
EXTEND
₹1.92 Cr
Cross-signal anomaly detection
EXTEND
₹0.70 Cr
Throughput up 0.5%
OPTIMISE
₹1.40 Cr
Energy down 6%
OPTIMISE
₹0.96 Cr
How we measure the "before"
  • We capture the baseline (MTTR, scrap rate, energy per unit, the numbers above) during the landing engagement, before Release 1 goes live, and agree it with you in writing.
  • Month 12 gets measured against that written baseline, not against a number either of us picks after the fact. Without this step, the return figure becomes an argument we lose.
What we deliberately left out
  • Downtime your predictive maintenance already prevents isn't counted twice, your maintenance budget bought that when it bought the sensors. We model residual downtime only: 3%, not 6%.
  • Every improvement rate sits at or below the bottom of its published benchmark range.
  • The throughput line assumes you're capacity-constrained. If you're demand-constrained, it's worth overtime avoidance instead, and drops by roughly 60%. That's the first question we'll ask you.
How You Start

Two numbers, and no multi-year lock

This sits on the AI budget, not the maintenance budget. Your existing subscription is untouched, and nothing here depends on renegotiating it.

One-Time: Landing Engagement
₹36 Lakh
Paid upfront, scoped to one site. Replaces every assumption in this document with your data, Release 1 goes live inside it.
Recurring: Platform Fee
₹1 Cr / site / yr
Covers all three releases as they ship. Renews annually, benchmarked against enterprise AI programmes rather than maintenance vendors.
Year One: Total
₹1.36 Cr
Against ₹7.06 Cr of annual benefit at month 12: 5.2× in year one, 6.3× across three years.
WHAT WE'LL NEED FROM YOU
  • 01
    Sensor history12+ months of signal on the instrumented assets, already true.
  • 02
    Document corpusSOPs, manuals, deviation records. PDFs are fine; a filing cabinet is not.
  • 03
    Defect imagesRoughly 200 labelled examples per class, or a 3-week collection window.
  • 04
    CMMS accessRead-only in Release 1, approved write-back in Release 2. Integrates with [list your supported CMMS/historian systems, e.g. SAP PM, Maximo, Fiix].
  • 05
    Process & energy tagsHistorian or PLC data alongside vibration, what makes cross-signal work.
  • 06
    A named AI sponsorCAIO or CDO, budget-holding, in the room. This one is the deal.
Adoption

Tooling only pays off if technicians actually use it. Release 1 includes floor-level training, a shift champion on each crew, and usage tracked alongside the ROI metrics above, so low adoption shows up as a number, not a surprise at renewal.

Security & Access

The questions your IT team asks before this goes further

We're asking for CMMS write access and plant documents, so these get answered before your security review stalls the deal, not during it.

Data residency[confirm hosting region & deployment model]
Model provider access[confirm what raw data, if any, reaches third-party model APIs]
Certifications[list current certifications, e.g. SOC 2, ISO 27001]
CMMS write-backREAD-ONLY IN RELEASE 1 · APPROVAL-GATED IN RELEASE 2
Document access[confirm retention & deletion policy for uploaded SOPs/manuals]

BRACKETED FIELDS NEED CONFIRMED ANSWERS FROM ENGINEERING BEFORE THIS SECTION GOES LIVE.

Every workflow becomes experience.
Every experience becomes intelligence.
That's the asset. Not the model.

Start With Two Hours, Not A Business Case

Own your intelligence before someone else owns the learning.

Bring your eight numbers. We'll rebuild the model on your site, live, in one session. One site, one named AI sponsor, two hours.

Drawn
Learning Layer
Scope
One Site
Sponsor
CAIO / CDO
Duration
2 Hours
Status
Open