Skip to main content
How Can We Help?Contact

Manufacturing, Industrial & Utilities

Keep the line running and the lights on. See the next failure coming.

We join what your ERP, MES, historian, asset, quality and outage systems each know. Maintenance is planned from how a press or a transformer is wearing, defects are caught at the operation that causes them, forecasts are good enough to build to, and a late supplier or an overloaded feeder shows up before anything stops. Machine and grid data is read, never written back to the controls.

Talk through your operations data See results from three manufacturers

Industrial equipment and parts · Consumer goods · Food and beverage · Electronics · Electric, gas and water utilities

A transmission tower and transformer feed a production line: parts ride a conveyor under a press and past a camera station, and leave on a truck; as the data line reaches each station it lights: the transformer running hot, the worn press, the part pulled for a defect, and the re-planned route1234
  1. 1GridFeeder transformer running hot; crew sent before it fails
  2. 2MaintenancePress 3 bearing wear flagged, fixed at the next changeover
  3. 3QualityTwo parts pulled at the camera, not at final inspection
  4. 4DeliveryOutbound route re-planned to hold the customer’s window
Illustrative: from the power feeding a line to the truck leaving it.

Five problems that start upstream of where they show up

Each one surfaces on the plant floor, out on the network or in a customer’s inbox, and each traces back to data one system holds and the next never sees. Each is read the way a five-whys is: from what the floor sees to the root cause.

  1. The same assets fail, again

    What the floor sees

    A press stops every few weeks, or a transformer fails in every heat wave, and maintenance hears about it from production or from customers calling in.

    Why?

    Maintenance runs on the calendar, while the vibration, temperature, load and cycle data that shows wear sits in the historian, unread.

    Root cause

    Asset condition and maintenance history live in two systems nobody has joined.

    Cost to the plantLost hours on the constraint or outage minutes on the feeder, parts flown in, overtime to catch up, and orders that ship late.

    See OP 40: maintenance planned from condition

  2. Scrap found at final inspection

    What the floor sees

    A batch fails at end-of-line inspection, after hours of work have gone into it.

    Why?

    Process settings, material lots and inspection results are recorded in different systems, by different people.

    Root cause

    Nobody can see which settings or lots go with which defects, so drift is caught only once parts fail.

    Cost to the plantScrap and rework, line capacity spent on parts that cannot ship and, once a defect reaches a customer, a corrective action request.

    See OP 30: first-pass yield, counted two ways

  3. A forecast the plant does not build to

    What the floor sees

    Sales, planning and the plant each carry their own number, and the plant builds to the one it trusts.

    Why?

    The forecast is built from last year’s shipments and rules of thumb, not from orders, market signals and lead times.

    Root cause

    No single forecast is accurate enough, or explained well enough, to plan on.

    Cost to the plantToo much stock in some lines, shortages in others, expediting, and missed delivery dates.

    See OP 40: forecasts the plant will build to

  4. Supplier problems found when the truck does not arrive

    What the floor sees

    A late or short delivery stops a line, and purchasing hears about it from the floor.

    Why?

    Lead times, supplier quality and the parts each supplier feeds are spread across the ERP, spreadsheets and email.

    Root cause

    Nobody has mapped which suppliers, and which of their suppliers, each product depends on.

    Cost to the plantLine stoppages, premium freight, and no quick answer when a customer or customs asks where a part came from.

    See OP 50: the rules on origin

  5. Crews sent by the calendar, not by risk

    What the field sees

    Circuits are trimmed and inspected on a fixed cycle, and the next outage still comes from one that was not due.

    Why?

    Outage history, asset age, inspection findings, tree growth and weather sit in the outage system, the GIS, work management and spreadsheets.

    Root cause

    Nobody can rank circuits and assets by the risk of failing before the next cycle.

    Cost to the utilityOutage minutes, contractor crews and overtime after storms, and reliability figures the regulator and customers see.

    See OP 40: the work on the network

Three manufacturers, and what each one measured

Each case set out as an inspection report: the client, what was built, and the results in ruled cells. The figures are the clients’ own; the clients are described, not named.

Report 01

Plan

A global manufacturer

A forecasting engine for sales and demand planning

Inaccurate forecasts left excess inventory in some product lines and shortages in others. CRM, ERP and third-party market data were integrated and cleansed; machine-learning forecasts by product line and region took in seasonality and trend and refreshed as new data arrived, with dashboards for the people who plan. The client reported that the gain let it optimize inventory and eliminate supply shortages.

Results for a global manufacturer
ResultWhat changed
+20%forecast accuracy

Reported by the client.

Case study: Sales forecasting for a global manufacturer

Report 02

Source and deliver

A large food manufacturer and distributor

An AI and analytics layer on the existing planning system

Who buys what, from where and how often was mapped across supermarkets, distribution centers and direct-to-warehouse deliveries. Cleaner demand, lead-time and location signals went into the existing MRP; a route engine sequenced deliveries and consolidated loads within each customer’s delivery window; planners got planned versus actual by customer and location. The MRP was never replaced.

Results for a large food manufacturer and distributor
ResultWhat changed
12%fewer delivery miles per order
18%fewer stockouts at key supermarket partners
9%better plan adherence

Reported by the client.

Case study: Route optimization on an existing planning system

Report 03

Service

A global electronics manufacturer

An omni-channel virtual agent for technical support

The manufacturer has a large installed base of devices, and its customers opened tickets through several channels, many of them for status, entitlement and basic troubleshooting. The agent, in chat and voice, looks up the user and their history across the CRM and the service desk, offers the three most likely fixes, reads some device and service data live from the manufacturer’s own servers, and hands off to the service desk with the full context.

Results for a global electronics manufacturer
ResultWhat changed
37%fewer Tier 1 live-agent contacts
28%higher first-contact resolution
22%faster time-to-answer

Reported by the client, for the issue types the agent covers.

Case study: A virtual agent for technical support

OEE, and seven more numbers each operation counts its own way

When the plant, finance, a customer or a regulator quote different numbers, the dashboard is rarely at fault; the counting rules are. Start with one shift and change how changeovers are counted. Then open any of the seven that follow.

One shift, two OEE numbers

A 480-minute shift on a line with an ideal cycle of one minute a part: 324 parts made, 308 of them good. Nothing about the shift changes below. Only the counting does.

Count changeovers as
  1. Shift480 min
  2. Breaksplanned−30
  3. Changeoversplannedlost−40
  4. Planned production time410450 min
  5. Breakdownslost−50
  6. Slow cycleslost−36
  7. Scraplost−16
  8. Good output308 min

OEE75%68%

308 good minutes ÷ 410450 planned minutes

Availability rate
88%80%
Performance rate
90%
Quality rate
95%

Same shift, same 308 good parts, and seven points of OEE between the two answers. A plant that counts changeovers as planned time can report the higher number while losing exactly the same minutes, and a benchmark against another plant means little until both count the same way.

Agree first: whether changeovers, planned maintenance and trial runs are planned time, and which cycle time counts as ideal: the nameplate rate, the best demonstrated, or the standard in the ERP.

The other seven, opened as spec sheets

Unplanned downtime

Counted as
  1. What counts as downonly breakdowns, or also waiting for material, an operator or a quality sign-off
  2. When the clock stopsat restart, or once the machine is making good parts again
  3. Short stopsstops of a few minutes counted as downtime, or as slow running
Used to decide
Which machines get maintenance attention first, and whether a spare is worth holding.
Worked example
A press stops for 40 minutes with a jammed feeder, then makes 15 minutes of bad parts while it settles: 40 minutes of downtime if the clock stops at restart, 55 if it stops at the first good part.
Agree before reporting
Which causes count, when the clock stops, and where short stops go?
Data from
the machine historian, MES downtime codes, the maintenance system

First-pass yield

Counted as
  1. Reworkparts reworked and then passed counted as good, or only parts right the first time
  2. Where it is takenat each operation and multiplied through the line, or only at final inspection
  3. Scrap by count or costevery scrapped part counted the same, or weighted by the value already added to it
Used to decide
Where quality engineers look first, and what the line can really ship.
Worked example
Three operations each pass 98% of parts first time. Each looks fine at 98%, but only 94% of parts pass all three first time.
Agree before reporting
Is rework good, and is yield taken per operation or across the line?
Data from
the quality system, inspection and SPC records, the MES

On time in full

Counted as
  1. Against which datethe date the customer asked for, or the date the plant promised
  2. By line or by orderan order counts only when every line ships complete, or each line counts on its own
  3. The windowthe exact day, or within a window agreed with the customer
Used to decide
Customer scorecards and penalties, and where to add capacity or stock.
Worked example
100 orders of five lines each: 480 of the 500 lines ship complete and on time, but the 20 misses fall across 12 orders. On time in full is 96% by line and 88% by order.
Agree before reporting
Against which date, by line or by order, and inside what window?
Data from
ERP orders and shipments, the warehouse and transport systems, customer portals

Inventory turns

Counted as
  1. Cost or salescost of goods sold divided by inventory, or sales divided by inventory, which reads higher
  2. Average or year-endagainst the average inventory over the year, or the closing balance
  3. Which inventoryraw materials, work in progress and finished goods together, or finished goods only
Used to decide
How much cash sits in stock, and which lines to hold less of.
Worked example
Cost of goods sold of 60 million on average inventory of 12 million is 5 turns. Use sales of 80 million instead and the same stock reads 6.7.
Agree before reporting
Cost or sales, average or year-end, and which inventory?
Data from
ERP inventory and costing, the general ledger

Forecast accuracy

Counted as
  1. Error or biashow far off the forecast is, or whether it runs consistently high or low
  2. At what levelby item and week, or by product family and month, where misses cancel out
  3. Which forecastthe one made a week out, or the one made at lead time, when the plant had to commit
Used to decide
What gets built, and how much safety stock each line carries.
Worked example
Two items are forecast at 100 each and sell 80 and 120. The family total is exact, so accuracy reads 100%, yet each item missed by 20 units.
Agree before reporting
Error or bias, at what level, and from which forecast?
Data from
the planning system, ERP orders and shipments, the CRM, market data

Standard cost variance

Counted as
  1. Which standardthe standard set at the start of the year, or one updated as material prices move
  2. Price or usagematerial price separated from usage and scrap, or one combined variance
  3. Where it landson the product, the line, or the plant’s overhead
Used to decide
Pricing, make-or-buy decisions, and which line’s costs to work on.
Worked example
A part has a standard cost of 10.00. Material prices rise by 0.60 and scrap adds 0.40 per good part: a variance of 1.00 that is 60% price and 40% usage, or one unexplained line on the monthly report.
Agree before reporting
Which standard, price separated from usage, and charged where?
Data from
ERP costing, the general ledger, MES production counts

Outage minutes (SAIDI)

Counted as
  1. Major event daysstorm days counted in, or set aside by the IEEE 1366 method so one year can be compared with the next
  2. Short interruptionsinterruptions of five minutes or less counted as outages, or kept in a separate count of momentary ones
  3. Planned workoutages for scheduled work counted, or left out because customers were told in advance
Used to decide
Where reliability spending goes, which circuits get work first, and what the regulator reads in the annual report.
Worked example
A utility with 100,000 customers logs 9 million customer-minutes of interruption in a year, 3 million of them on two storm days that qualify as major event days. SAIDI is 90 minutes with the storms and 60 without.
Agree before reporting
Storm days in or out, which interruptions count, and is planned work included?
Data from
the outage management system, the customer information system, meter events, the GIS

Which number starts an argument in your morning meeting? Bring it to a 30-minute call, and we will show you what it takes to count it once and report it the same way at every site.

Bring one disputed number to a call

Plan, source, make, deliver, and out on the network

Laid out the way supply chain teams map their operations, one stage at a time, with a row of its own for utilities. Each piece names the number it moves; where a case study exists, it is linked.

Plan

What to build, and when

  • Forecasts the plant will build to

    MovesForecast error and bias

    Orders, shipments, pipeline and market signals joined, with machine-learning forecasts by product line and region that refresh as data arrives and show why they moved, so sales, planning and the plant work from one number.

    Case study: Sales forecasting for a global manufacturer

  • Plans that match real demand

    MovesPlan adherence

    Cleaner demand, lead-time and location signals fed into the planning system you already run, so the MRP plans against the way customers actually order. Nothing is replaced.

    First step: Data Platform Accelerator

Source

Who supplies it

  • A map of who supplies what

    MovesLine stops for missing parts

    Suppliers, the parts they ship, lead times, quality history and, where you have it, their own suppliers, joined into one map, so a late shipment or a supplier at risk shows which products and orders it touches.

    First step: Data Platform Accelerator

  • Origin records that hold up

    MovesTime to answer an origin question

    Bills of materials, purchase records and supplier declarations kept together by part and lot, so you can show where an input came from when customs, a customer or an auditor asks.

    First step: AI & Data Assessment Sprint

Make

On the line

  • Maintenance planned from machine condition

    MovesUnplanned downtime hours

    Vibration, temperature, current and cycle data from the historian joined to work orders and failure history, so the planner sees which asset is wearing, weeks ahead, and plans the fix into a changeover instead of a breakdown. Data is read from the plant network, never written back to controls.

    First step: AI & Data Assessment Sprint

  • Defects caught at the operation that causes them

    MovesFirst-pass yield and scrap

    Camera inspection at the station and statistical process control on the settings that matter, joined to material lots and inspection results, so drift is flagged while parts are still good and the operator sees the likely reason, not just an alarm.

    First step: AI & Data Assessment Sprint

Deliver

To the customer, and after

  • Routes and loads planned together

    MovesMiles per order and on-time delivery

    Deliveries sequenced, loads consolidated and routes chosen within each customer’s delivery window, with planned versus actual by customer and location.

    Case study: Route optimization on an existing planning system

  • Service that knows the installed base

    MovesTier 1 contacts and first-contact resolution

    A virtual agent for customers and service teams that looks up the user, the equipment and its history, offers the three most likely fixes, and hands off with the full context.

    Case study: A virtual agent for technical support

On the network: electric, gas and water utilities

  • Asset health for transformers, poles and pipes

    MovesFailures and outage minutes

    Inspection findings, load, temperature, dissolved-gas readings, age and failure history joined asset by asset, so replacement and maintenance budgets go to the assets most likely to fail, not the oldest on the list.

    First step: AI & Data Assessment Sprint

  • Vegetation and inspection cycles set by risk

    MovesTree-caused outages per mile

    Outage history, aerial or satellite imagery, growth rates and weather joined circuit by circuit, so crews trim and inspect where the next outage is most likely, with a plan you can defend to the regulator.

    First step: AI & Data Assessment Sprint

  • Load forecasts and meter data that agree

    MovesForecast error at the feeder

    Smart meter data, weather, new connections and rooftop solar joined feeder by feeder, so planners see peak load and reverse flow where they will happen, and customer usage questions are answered from the same data.

    First step: Data Platform Accelerator

Under every stage

  • One data model from the ERP to the machine

    ERP, MES, historian, quality and maintenance data, and for utilities the GIS, outage and meter records, modeled once: along ISA-95 for plants and the Common Information Model for utilities, so every use above draws on the same parts, assets, circuits and orders.

    First step: Data Platform Accelerator. Practice behind it: Data Engineering & Platforms.

  • OT security and an AI register

    What may connect to the plant or grid control network and through which zone, and which AI uses need engineering sign-off, kept on one register with an owner for each.

    First step: AI Governance Quickstart. Practice behind it: AI Security & Cyber Risk.

  • People on the floor and in the field

    Operators, technicians, line crews and planners trained on their own equipment during their own shifts, alerts that give a reason, and supervisors who decide what changes in standard work, so the tools are used after go-live.

    First step: Adoption Review. Practice behind it: AI & Data Literacy.

Your biggest loss is somewhere else, on the line or on the network? Bring the one that costs the most hours to a 30-minute call.

Talk through your biggest loss

The rules depend on what you make or run, and who you serve

A matrix, not a checklist: find the rows that fit your plants, networks, products and customers. Each rule links to its source, with where it stands as of October 2026.

Rules that may apply to the data of a manufacturer, an industrial company or a utility: who each applies to, where it stands and what it asks
RuleApplies if you areWhere it standsWhat it asks of your data
Cybersecurity Maturity Model Certification (CMMC) (opens in a new tab)
  • A Defense Department supplier
  • Holding controlled unclassified information
Partly suspendedSelf-assessments have appeared in new contracts since Nov 10, 2025. Third-party assessments, due from Nov 10, 2026, were suspended on Jul 13, 2026, pending a review.Know where controlled unclassified information lives, analytics and AI included. DFARS 252.204-7012 and NIST SP 800-171 still apply to every system that holds it.
FDA Food Traceability Rule (FSMA section 204) (opens in a new tab)
  • Making, packing or holding foods on the Food Traceability List
Enforcement deferredThe compliance date was Jan 20, 2026; Congress has directed FDA not to enforce the rule before Jul 20, 2028.Key data elements for each critical tracking event, tied to a traceability lot code, and records to FDA within 24 hours of a request, as a sortable spreadsheet during an outbreak or recall.
FDA Quality Management System Regulation (21 CFR 820) (opens in a new tab)
  • A medical device maker
Applies nowEffective Feb 2, 2026. It replaces the Quality System Regulation and incorporates ISO 13485:2016.FDA can now inspect management review, quality audit and supplier audit reports, and ISO 13485 requires software used in the quality system, AI tools included, to be validated for its use.
Uyghur Forced Labor Prevention Act (opens in a new tab)
  • An importer into the United States
Applies nowThe presumption has applied since Jun 21, 2022.Goods made wholly or partly in Xinjiang, or by listed entities, are presumed made with forced labor. Rebutting that takes clear and convincing evidence: supply chain tracing back to raw materials, and complete answers to CBP’s requests.
ITAR: encrypted technical data (22 CFR 120.54) (opens in a new tab)
  • Making defense articles
  • Holding their technical data
Applies nowSince Mar 25, 2020.Unclassified technical data sent or stored with end-to-end encryption, to FIPS 140-2 or equivalent strength, is not an export, provided it is not sent to or stored in a proscribed country and no third party holds the means to decrypt it. It decides where drawings and models may sit, and which AI tools may read them.
NERC CIP standards, now with internal network monitoring (CIP-015-1) (opens in a new tab)
  • A registered entity on the bulk power system
Applies nowMandatory and enforced. FERC approved CIP-015-1 in Order No. 907 (Jun 26, 2025; effective Sep 2, 2025); compliance phases in under NERC’s implementation plan.BES Cyber Systems are categorized by impact and kept inside electronic security perimeters, with access control, logging and now monitoring of the traffic inside them. Analytics or AI that reaches those systems sits inside the same rules.
America’s Water Infrastructure Act, section 2013 (opens in a new tab)
  • A community water system serving more than 3,300 people
Applies nowSecond round under way: assessments certified by Mar 31, 2025, Dec 31, 2025 or Jun 30, 2026, depending on size, and response plans six months later.The risk and resilience assessment covers electronic, computer and automated systems and their security, monitoring practices, and operation and maintenance, and is reviewed and recertified to EPA every five years.
Illinois Biometric Information Privacy Act (opens in a new tab)
  • Running an Illinois site
  • Using fingerprint, hand or face time clocks
Applies nowSince 2008; amended Aug 2, 2024.A public retention and destruction policy, written notice and a written release before any scan, and destruction once the purpose is met or within three years of the last interaction. An electronic signature now counts as a release, and repeated scans of one person by the same method count as one violation.

For orientation only, not legal advice: what applies depends on what you make, where, and for whom, and each rule links to its source. Checked October 14, 2026.

Bill of data: the systems we join

Set out the way a bill of materials is, level by level, from the operations model down to the machine. Machines and grid devices are read through a gateway your OT team approves; nothing writes back to a controller.

Bill of data: the systems we join
LevelComponentWhat it carries
0Operations data model (ISA-95, CIM)One model of parts, lots, assets, circuits, orders and people, shared by every report and every model.
.1ERPOrders, bills of materials, routings, inventory, standard costs and purchasing.
.1MESProduction counts, downtime codes, and genealogy by lot and serial number.
..2HistorianSensor readings by the second: vibration, temperature, current, pressure and cycle times, over OPC UA or MQTT.
...3Machines and controllersRead only, through a gateway in the plant DMZ.
.1Quality systemInspections, SPC charts, nonconformances and corrective actions.
.1Maintenance systemWork orders, failure codes, spare parts and asset history.
.1Product lifecycle recordsDrawings, revisions and specifications, with export markings where they apply.
.1Warehouse and transportStock locations, shipments, loads and routes.
.1Utility network systemsOutage management, the GIS, the SCADA historian and work management, for utilities.
..2Meter dataInterval reads and meter events from smart meters, by meter and feeder.
.1Suppliers and customersEDI orders and forecasts, supplier declarations, certificates and portals.

Put an AI use case through the gate before it reaches the floor or the field

Pick a use case to see what it acts on, where it reaches, the rules to check, what has to be true before it passes, and who would usually own the gate.

Maintenance

Predictive maintenance alerts

  1. What it acts on

    Which machines are serviced, and when.

  2. Where it reaches

    • Could affect worker or product safetySometimes
    • Connects to the plant networkYes
    • Touches export-controlled dataNo
    • Monitors or decides about workersNo
    • Uses customer or supplier dataNo

  3. Rules to check

    • OT security practice, usually IEC 62443: machine data read through a monitored zone, never by writing to controllers.
    • CMMC, if the plant’s network is in scope for defense work.

  4. Exit criteria

    • A read-only connection through the plant DMZ.
    • Alerts checked against past failures before go-live.
    • The maintenance planner decides every work order.

  5. Gate owner (suggested)

    Operations gate

First step: AI & Data Assessment Sprint

Inspection

Camera inspection at the station

  1. What it acts on

    Which parts are pulled from the line.

  2. Where it reaches

    • Could affect worker or product safetySometimes
    • Connects to the plant networkYes
    • Touches export-controlled dataNo
    • Monitors or decides about workersSometimes
    • Uses customer or supplier dataSometimes

  3. Rules to check

    • Customer quality requirements: a change to an approved inspection method may need the customer’s approval, as under PPAP in automotive.
    • For medical devices, the QMSR and ISO 13485: software in the quality system validated for its intended use.

  4. Exit criteria

    • Tested on parts with known defects, rare ones included.
    • Rejects held for a person to confirm until the error rate is agreed.
    • Cameras aimed at the part, not the operator.

  5. Gate owner (suggested)

    Cross-functional gate

First step: AI & Data Assessment Sprint

Process

Recommended machine settings

  1. What it acts on

    How the process runs: temperatures, speeds and pressures.

  2. Where it reaches

    • Could affect worker or product safetyYes
    • Connects to the plant networkYes
    • Touches export-controlled dataSometimes
    • Monitors or decides about workersNo
    • Uses customer or supplier dataNo

  3. Rules to check

    • Your management-of-change procedure, and OSHA process safety management where highly hazardous chemicals are involved.
    • Machine and process safety: AI stays outside safety functions.

  4. Exit criteria

    • Recommendations only; an operator or engineer makes every change.
    • Limits set by engineering that no recommendation can exceed.
    • Never connected to a safety instrumented system.

  5. Gate owner (suggested)

    Cross-functional gate

First step: AI & Data Assessment Sprint

Utilities

Switching and setpoint advice in grid or water operations

  1. What it acts on

    How the network is switched, loaded or pressured.

  2. Where it reaches

    • Could affect worker or product safetyYes
    • Connects to the plant networkYes
    • Touches export-controlled dataNo
    • Monitors or decides about workersNo
    • Uses customer or supplier dataNo

  3. Rules to check

    • NERC CIP, where the tool reaches BES Cyber Systems or the networks around them.
    • America’s Water Infrastructure Act: process control systems are part of a water system’s risk and resilience assessment.

  4. Exit criteria

    • Advice only: an operator issues every switching order and setpoint change.
    • No write access to SCADA or any other control system.
    • Tested in a study or simulation environment before it advises live operations.

  5. Gate owner (suggested)

    Cross-functional gate

First step: AI & Data Assessment Sprint

Utilities

Outage prediction and crew staging before storms

  1. What it acts on

    Where crews, trucks and materials wait before a storm arrives.

  2. Where it reaches

    • Could affect worker or product safetySometimes
    • Connects to the plant networkNo
    • Touches export-controlled dataNo
    • Monitors or decides about workersNo
    • Uses customer or supplier dataNo

  3. Rules to check

    • No AI-specific rule, but the state utility commission can review storm decisions and restoration times after a major event.

  4. Exit criteria

    • Forecasts checked against past storms before they are used.
    • The model advises; operations decides where crews wait.
    • Each staging decision logged with the forecast it used.

  5. Gate owner (suggested)

    Operations gate

First step: AI & Data Assessment Sprint

Planning

Demand forecasting

  1. What it acts on

    What gets built and stocked.

  2. Where it reaches

    • Could affect worker or product safetyNo
    • Connects to the plant networkNo
    • Touches export-controlled dataNo
    • Monitors or decides about workersNo
    • Uses customer or supplier dataSometimes

  3. Rules to check

    • Little beyond your own policies, provided customer forecast data is used only as each customer’s agreement allows.

  4. Exit criteria

    • Error and bias measured against the current forecast before switching.
    • Planners can see why a forecast moved.
    • A planner approves every plan.

  5. Gate owner (suggested)

    Owner check

First step: AI & Data Assessment Sprint. Case study: Sales forecasting for a global manufacturer

Planning

Production scheduling and sequencing

  1. What it acts on

    What runs on which line, and in what order.

  2. Where it reaches

    • Could affect worker or product safetyNo
    • Connects to the plant networkSometimes
    • Touches export-controlled dataNo
    • Monitors or decides about workersSometimes
    • Uses customer or supplier dataNo

  3. Rules to check

    • Labor agreements, if the schedule assigns overtime or shifts to people.

  4. Exit criteria

    • The scheduler can override any sequence.
    • Changeover and maintenance windows respected.
    • Run beside the current schedule for a month before it replaces it.

  5. Gate owner (suggested)

    Operations gate

First step: AI & Data Assessment Sprint

Sourcing

Supplier risk and origin tracing

  1. What it acts on

    Which suppliers you use, and what you tell customs.

  2. Where it reaches

    • Could affect worker or product safetyNo
    • Connects to the plant networkNo
    • Touches export-controlled dataNo
    • Monitors or decides about workersNo
    • Uses customer or supplier dataYes

  3. Rules to check

    • The Uyghur Forced Labor Prevention Act: tracing records that can stand as clear and convincing evidence.
    • Supplier agreements: what their data may be used for.

  4. Exit criteria

    • Every risk score shows its sources.
    • A buyer reviews before any supplier is dropped.
    • Records kept by part and lot.

  5. Gate owner (suggested)

    Operations gate

First step: Data Platform Accelerator

Engineering

An assistant over drawings and specifications

  1. What it acts on

    What engineers reuse in new designs and quotes.

  2. Where it reaches

    • Could affect worker or product safetySometimes
    • Connects to the plant networkNo
    • Touches export-controlled dataYes
    • Monitors or decides about workersNo
    • Uses customer or supplier dataSometimes

  3. Rules to check

    • ITAR and the EAR: export-controlled technical data reaches only authorized people, and sits in the cloud only under the encryption carve-out.
    • CMMC and NIST SP 800-171, where drawings are controlled unclassified information.
    • Customer drawings used only as each customer’s contract allows.

  4. Exit criteria

    • Export-controlled drawings kept out, or walled off, by default.
    • The assistant follows existing permissions in the product records.
    • Every answer cites the drawing and revision it came from.

  5. Gate owner (suggested)

    Cross-functional gate

First step: AI Governance Quickstart

Maintenance

A technician’s assistant over manuals and work instructions

  1. What it acts on

    How a technician diagnoses and fixes a machine.

  2. Where it reaches

    • Could affect worker or product safetyYes
    • Connects to the plant networkNo
    • Touches export-controlled dataSometimes
    • Monitors or decides about workersNo
    • Uses customer or supplier dataNo

  3. Rules to check

    • OSHA lockout/tagout (29 CFR 1910.147): the assistant never stands in for the energy control procedure.

  4. Exit criteria

    • Answers only from current, approved manuals and work instructions.
    • Every answer names the document and revision.
    • Safety steps shown exactly as written, never summarized.

  5. Gate owner (suggested)

    Operations gate

First step: AI & Agent Pilot

Service

A virtual agent for customer support

  1. What it acts on

    What customers are told about their equipment.

  2. Where it reaches

    • Could affect worker or product safetySometimes
    • Connects to the plant networkNo
    • Touches export-controlled dataNo
    • Monitors or decides about workersNo
    • Uses customer or supplier dataYes

  3. Rules to check

    • Customer contracts and privacy terms: which customer and device data the agent may use.
    • Product claims and warranty terms stated as they are written.

  4. Exit criteria

    • Answers from current product, entitlement and warranty data.
    • Hand-off to a person with the full context.
    • Anything safety-related routed straight to a person.

  5. Gate owner (suggested)

    Operations gate

First step: AI & Agent Pilot. Case study: A virtual agent for technical support

Plant HR

Monitoring operators, or biometric time clocks

  1. What it acts on

    How people are measured, scheduled or disciplined.

  2. Where it reaches

    • Could affect worker or product safetySometimes
    • Connects to the plant networkNo
    • Touches export-controlled dataNo
    • Monitors or decides about workersYes
    • Uses customer or supplier dataNo

  3. Rules to check

    • Illinois BIPA, for fingerprint, hand or face scans: written notice, a written release and a public retention policy.
    • Labor agreements: where a union represents the workers, introducing monitoring may need to be bargained.

  4. Exit criteria

    • A purpose agreed before any data is collected.
    • Workers told what is collected and why.
    • No automated discipline: a supervisor decides.

  5. Gate owner (suggested)

    Cross-functional gate

First step: AI Governance Quickstart

An early read, not a ruling. What applies depends on what you make, where, for whom, and how the tool is used; your own change and safety processes decide. The rules in full: OP 50.

Suggested gate owners. More than one may be needed: engineering owns the change to the process, OT security owns the connection, and operations owns how the floor or the field uses it.

  • Cross-functional gate: engineering, safety, OT security and quality, with HR or export compliance where people or controlled data are involved.
  • Operations gate: the plant or operations manager, the process owner and OT security.
  • Owner check: a named owner, access controls and a periodic check.

A use case not on the list, or one already running on the floor or in the field without these safeguards? Thirty minutes is usually enough to decide what it needs before it goes further.

Talk through one use case on the floor

Where to start, by who is asking

Set out as a routing: who each route is for, how long it takes, what it runs as and what it produces. None commits you to the next.

  1. Route A

    Unplanned downtime diagnostic

    For a VP of operations, plant manager or asset manager losing hours to failures

    Time

    Two weeks, one asset group

    Runs as the AI & Data Assessment Sprint

    Output

    One asset group’s condition, maintenance and operating data traced and checked against what a predictive model would need; the gaps on one list with their cost; and a go or no-go on a maintenance pilot, with the downtime or outage minutes it should recover.

    Ask about a two-week sprint

  2. Route B

    Where AI should start in your operations

    For a general manager or owner deciding what to fund first

    Time

    Several hours, one session

    Runs as the Executive AI Workshop

    Output

    Your leadership team in one working session: your AI ideas ranked by value, data readiness, risk and what the floor will adopt, and one first move to fund.

    Ask about the AI workshop

  3. Route C

    Data Maturity Assessment

    For an IT or operations lead who wants a quick read first

    Time

    About six minutes, free

    Output

    Fifteen questions on lineage, quality, ownership, access and platform. It fits between shift meetings; the result is instant and needs no email.

    Score your data foundation

After either of the first two: a first build on one line or one asset group, measured against a number you already track, such as downtime hours, first-pass yield, forecast error or outage minutes, then a decision to scale, change or stop.

AI & Machine Learning · Data Platform Accelerator · AI & Agent Pilot

Asked by operations, IT and quality

Will anything you build connect to our machines, SCADA or controls?

Only to read, and only by the route your OT team approves: usually the historian, or a gateway in the plant or substation DMZ, never directly to a controller. Nothing we build writes back to controls or sits in a safety function. When a recommendation should change a setting or a switching plan, a person makes the change through your normal change process. For utilities, nothing reaches a system inside a NERC CIP electronic security perimeter without your compliance team’s approval.

Do you work with utilities as well as manufacturers?

Yes: electric, gas and water utilities, on asset health, outage and vegetation risk, load and meter data, and the security rules that come with operational systems. The case studies published so far are from manufacturers.

Our ERP and MES are old. Do we need to replace them first?

No. The food manufacturer in OP 20 kept its MRP; the work added cleaner signals and a layer on top. We read from the systems you run, model the data once, and leave your systems of record where they are.

Our software vendors are adding AI. What is left for you to do?

Switch on what fits; vendor features are often the quickest route. We help you decide which to use and bring them under one policy, and build only where a question crosses systems no single vendor sees, such as downtime set against maintenance history and the production plan, or a forecast that uses orders, market data and lead times together.

One plant has good data and three do not. Where do we start?

In the plant with the data, on one line or one asset group, measured against a number that plant already tracks. What works there becomes the template the other plants connect to, and the gaps in the other three are listed with what it would take to close them.

How soon will a predictive maintenance model show results?

The diagnostic takes two weeks and ends with a go or no-go. After that it depends on how often the asset fails: a press that jams monthly proves a model sooner than a pump or a transformer that fails once a year, which is why the diagnostic checks the failure history first.

How do you handle drawings and data that are export-controlled or CUI?

We find and mark them first, keep them out of anything not cleared for them, and build on systems that meet NIST SP 800-171 where controlled unclassified information is involved. Controlled data stays in your environment and under your access controls, and only people your export compliance program authorizes work on it.

Will operators trust it?

When it explains itself and they had a say. We build with the operators and technicians who will use it, show the reason behind every alert, and leave the decision with them. Supervisors decide what changes in standard work, and training happens on the plant’s own equipment, during the shift.

Do you only work with large companies?

No. The work is sized to one plant, one line, one asset group or one question at a time, so it fits a family-owned manufacturer or a municipal utility as well as a global company. Many start with the Executive AI Workshop or the free Data Maturity Assessment.

Pull the cord on the failure that costs you most.

The machine that keeps going down, the transformer that fails in every heat wave, the defect found too late, or the supplier you cannot trace. Thirty minutes with the person who would lead the work, on what to fix first.

Book 30 minutes about your operation

A failure nobody saw coming? Thirty minutes on the machine, the network or the forecast.

Book a call about your operation