Downtime and bad parts cost more than most plants see

8.3%

Of planned production time lost to downtime in US discrete manufacturing, about $245B a year1

NIST, Annual Report on the U.S. Manufacturing Economy, 2024

$150K

Top-end cost of one hour of unplanned downtime for a small or mid-sized manufacturer2

Siemens, True Cost of Downtime 2024

85%

Of defective parts caught by trained inspectors, who also rejected 35% of good ones3

Sandia National Laboratories, 2015

70%

Of manufacturers still enter production data by hand4

NAM Manufacturing Leadership Council, 2024

OEE with the reason behind every loss

Availability from real stop times, performance from counted cycles, quality from rejects seen at the outfeed. Each of the six big losses gets its own time, cause and clip.

Availability

80%

Time the machine was scheduled but not running.

BreakdownsSetup and adjustment

Why it stopped →

Performance

85%

Running, but slower than it should.

Small stopsReduced speed

Shift timeline →

Quality

96%

Parts made that weren't good.

Start-up rejectsProduction rejects

Counted and checked →

Molding cell B, shift 165%OEE, with every lost minute explained

Example data. OEE = availability × performance × quality.

A sensor knows a machine stopped. The camera knows why.

Clamps, cycle sensors and stack lights show the same red for a tool change, an empty parts cart or nobody at the press. The reason waits for someone to enter it, often hours later. The camera sees the cause and saves the clip.

What a sensor records

P7 · 09:30 to 10:30 · example data

09:3010:0010:30

State: Stopped, 10:02 to 10:16
Duration: 14 min
Reason: Not entered
Evidence: None

Terminal · P7 · qualify stop 10:02

MaterialChangeoverMaintenanceNo operatorOther

Someone picks a reason at the end of the shift. Often it's "Other".

A stopped injection molding press with an amber stack light, an empty parts cart beside it and nobody at the station, while a forklift brings a box of resin pellets down the aisle — CAM-05 · Molding cell B · P710:14

P7 · Waiting on material · 14 min · clip saved10:02–10:16

Parts cart emptyNothing staged for the next run
No operator at the pressStation empty since 10:03
Resin delivery arrivingForklift at the press at 10:14

Reasons without the tabletClassified from the scene. Operators confirm instead of typing from memory.

Evidence, not a codeEvery stop, changeover and reject keeps its clip.

Small stops and slow cyclesThe performance losses nobody writes down, timestamped by machine and shift.

01 · AVAILABILITY AND PERFORMANCE

Every minute of the shift, in states that mean something on your floor

Every minute lands in an OEE loss you can act on. You choose each machine's states, and we build them from what the camera sees: the gate, the mold, the operator, the parts cart, the technician. Below, half a shift on the presses above.

Molding cell B · shift 1Press states from 06:00 to 10:35 · example data

Running 78%

Select a segment to see what the camera saw.

Presses, lines, cells, ovens, cranes. If a person can see its state, so can the camera.

Each machine type gets its own states and stop reasons.

Which of your machines should we look at first?Send a few hours of footage. We'll show you its states and OEE losses.

Book a free feasibility study

02 · SETUP AND ADJUSTMENT LOSS

The minutes inside every changeover, step by step

Changeover is an availability loss. A sensor or app logs when it starts and ends. The camera shows what happened in between: which minutes were waiting, which were fetching and which were the work.

A setter guides a steel mold hanging from an overhead crane into an injection molding press, with a tooling cart of clamps and hoses beside the machine — CAM-02 · Molding cell B · P2Tue 14:31

Also measuredDie changeTool changeFixture swapLabel and film roll changeFlavor or product changeNozzle changeColor change and purge

P2 · mold swap, job 4471 to 4480Tuesday, shift 2 · example data

37 min

What shorter changeovers are worth on one machine

Use your own numbers. Illustrative, not a promise.

Changeovers a day Minutes each today Minutes each after

45 minBack each production day on this machine

188 hA year, at 250 production days

Minutes saved × changeovers a day × 250 days. Freed time becomes output only if the machine is the constraint and there's demand.

Timed the same way every shiftLast good part to first good part, both visible on camera.

Work that belongs before the stopFetching tooling, waiting for a setter, hunting for clamps.

Your best changeover as the standardTrain from the fastest crew's clip for the same job.

03 · QUALITY LOSS

Count every good part and every reject, then check the parts themselves

Most monitoring tools count cycles and leave scrap to a button press. A camera over the outfeed counts good parts and rejects for the quality side of OEE, checks each part for visible defects, and shows what changed when rejects rise.

The outfeed of an injection molding press: parts drop onto a short conveyor under a camera mounted on a bracket, into a blue bin for good parts and a red bin for rejects — CAM-P4-OUT · P4 outfeed10:35

Rows of molded plastic parts on a conveyor seen from directly above; two parts have ragged flash along an edge and one is missing a corner — CAM-P4-OUT · frame 10:31:073 flagged · 72 passed

P4 · shift 1 so far06:00 to 10:35 · example data

Rejects rising

1,184Good parts

23Rejects

98.1%Quality rate

Rejects per hour10:00 hour: 10

What changed before the riseEvents on P4 from the same cameras · example data

Quality in your OEE, not typed inGood and reject counts by press, job and shift.

Every part, not a sampleShort shots, flash, missing features, wrong color.

Traceable to the cycleThe clip of the cycle that made a returned part.

Part checks need a camera over the outfeed. Reject counts can come from a camera that already sees the bins.

04 · MACHINE HEALTH

Older machines with no network port still report

US manufacturing equipment averages 9 years old by value, and many machines are far older.5 Legacy equipment is manufacturers' top smart-factory roadblock.6 Those machines still have gauges, sight glasses and stack lights. A camera can read them.

The side of an older hydraulic press: two analog pressure gauges, an oil sight glass, a stack light and a drip of oil on the floor below a fitting — CAM-09 · Press shop · hydraulic press H207:45

Hydraulic press H2No controller connection · example data

2 to check

Also visibleCoolant levelChip buildupConveyor slipSmoke or mistGuards open while runningHMI alarm screenProof a PM was doneTime to respond and repair

Keep the signals you have. Add what they can't see.

Sensors and controllers count cycles well. They can't explain a stop, break down a changeover or check a part. Agnify fuses their signals with what the camera sees.

What you need	Retrofit sensorsCurrent clamps, cycle sensors, buttons	Controller or PLC connectionWhere the machine has one	Agnify cameras + signal fusionExisting cameras, plus one where needed
Run, stop and cycle counts	YesCheap and very reliable	YesExact, from the controller	YesFrom the scene, or from your signal where one exists
Why it stopped	Operator inputA button or a reason picked on a terminal	PartlyAlarms and program states; not an empty cart or a missing operator	YesNo operator, no material, tool change, jam, maintenance, with the clip
Inside a changeover	Start and endLogged in an app	Start and endFrom program changes	Each stepWaiting, fetching, mold out, mold in, first good part
Part quality	Operator inputScrap entered by hand	If wiredReject counts from the machine	YesCounted and checked at the outfeed
Older machines with no network	YesA sensor on each machine	NoNeeds a controller to connect to	YesStack light, gauges, motion, parts
Inside an enclosed machine	YesNo line of sight needed	Yes	With fusionCamera on the door, light and outfeed; signal for the inside
Around the machine	No	No	YesMaterial staged, carts, forklifts, operator at the station
Hardware on each machine	A sensor and a battery per signal	A connection and gateway per machine	None where a camera already sees it; one camera can cover several machines

Signal fusion. Where a machine has a PLC output, stack-light tap or cycle sensor, Agnify uses it for exact counts and timing, and the camera for the reason, the changeover steps and the part check.

Start with the cameras you have and add one where a machine needs it

No sensor on every machine and no controller integration to start. We connect to your existing feeds, add a camera where a view is missing, and bring in machine signals where they help.

STEP 1

Connect the views

RTSP streams, your NVR or recorded footage, mapped to the machines each camera sees.

STEP 2

You define the states

Your team decides which states and stop reasons matter. Agnify's engineers program them and test on your footage before go-live.

STEP 3

Run the floor on it

State timelines by machine and shift, alerts with the frame, reports, and events through the API.

Cloud by default

Nothing to host. Results in minutes for dashboards, shift reports and trend alerts.

Real time on site

Alerts the moment a machine stops, from a server next to your cameras.

Signals in, events out

Machine signals in. States, counts and events out to your MES, BI or spreadsheets.

See your machines' states before you commit

A free feasibility study on one cell or line, with an agreed end date.

Book a free feasibility study

You send a few hours of footage, or camera accessPick the machines you most want to understand.
We come back with a written resultWhich states and reasons we can see, how reliably, and what the first timeline would show.
If it's useful, we scope a pilot togetherA few machines, live, judged against criteria agreed up front.

Questions plant teams ask first

We're looking at sensor-based monitoring. Why cameras?

Sensors count cycles well, and Agnify can use their signal. They can't tell you why a machine stopped, what happened in a changeover or whether the parts are good. The camera can, and keeps the clip.

Do we need new cameras?

Often not for machine states: a camera that sees the machines and their stack lights is usually enough. For part checks we suggest a camera over the outfeed. The feasibility study checks angles and resolution.

Some of our machines are fully enclosed.

We watch what's visible (door, stack light, outfeed, operator, material) and fuse a machine signal, such as a PLC output or cycle sensor, for the inside.

How do you handle privacy and our workforce?

Reports are by machine, role and shift, never by person. No facial recognition, whole-person blur on request, short retention, and video can stay on a server in your building. We help you brief your teams and any works council or union.

Can it alert us the moment a machine stops?

Yes, from a server on site. Dashboards and reports run in the cloud within minutes. Agnify isn't a certified safety system and doesn't replace machine guarding.

How do we know it's accurate on our machines?

We measure it on your footage, against stops, changeovers and rejects your team checks by hand, before anything scales.

What does it take from our IT team?

Camera streams or an NVR export, and a network path out. A server on site if you need real time or video must stay in the building.

Tell us about your machines

Tell us what you run and what you want to know. We'll reply within one business day.

Free, time bound and no commitment
Works from a few hours of footage or a camera stream
Or email info@agnify.ai

Know what every machine is doing, and why