Field-Service & Predictive-Maintenance AI Agent

What this agent does

Ranks the queue, does not call the failure

Sensor channels, alarm history and work orders, ingested from the condition-monitoring, CMMS and asset sources on file.

Units, sample rates and asset identifiers, normalised so one asset reads the same in each source.

Reason

A ranked queue of assets, ordered by signal strength in the instrumented channels.

Your existing PM regime and consequence classification, applied exactly as written.

Each flag bound to the channel, window and threshold it came from, and what the data cannot see.

Decide

Failure modes with no instrumented precursor, marked as invisible here.

Dispatch proposals held for the named reliability engineer.

Out

The signal, the flag, the engineer's decision and the outcome, kept against the asset record.

Execute write actions only inside the approval boundaries agreed during implementation.

→ Product statement

The agent ranks the work; a reliability engineer decides what is dispatched, and the employer stays accountable for the work itself.

Example workflow

One asset, signal to dispatch

AgentHuman

1Condition signal receivedVibration, thermal, motor-current, telemetry feed or CMMS alarm

2Evidence assembledChannel history, thresholds, work orders and the asset's existing PM regime

3Work rankedRank, channel evidence, lead-time window and confidence

4Controls appliedBase-rate checks, coverage checks, consequence classification and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is dispatched at any of them — the agent is ranking, and the engineer's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold

High confidence

Goes to the reliability engineer to approve.

Low confidence

Adds a condition-monitoring read first.

Engineer approval

The flag is held with its channels, its coverage gaps and the confidence.

Approve · Amend · Send to condition monitoring

Approved — released to dispatch▼

6Field-service system updatedOnly where write access and approval policy allow it

7Outcome evaluatedFlag precision, lead time achieved, wasted trips and failures nobody flagged

Wasted trips

A trip that found nothing is counted against the flag that sent it.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required8 items

Dispatching anyone to work on energised equipment.

Applying, verifying or clearing a lockout or tagout.

Changing a maintenance regime or an inspection interval.

Standing an asset down, or returning it to service.

Automation boundaryAgent acts unaided

✓Rank assets by the signal strength in the instrumented channels.

✓Assemble the evidence behind a flag, source by source.

✓Brief a technician from the documented procedure.

✓Report flag precision and the lead time actually achieved on outcomes.

Any write happens inside the boundaries agreed at implementation, never ahead of approval.

Issuing or justifying an energised-work permit.

Deciding that a person is qualified for a task.

Setting arc-flash boundaries or PPE category.

Changes to thresholds, ranking or approval rules.

Example output

One flag on one asset, annotated

Everything the agent ranks is attached to the signal it was drawn from.

Field-service output · single assetIllustrative example

Asset

Flagged condition

Signal trend

Source channel

Confidence

Coverage

Line-side drive unit

Bearing-band energy rising across consecutive sampling windows

Rising, sustained

Vibration channel on file

63%

Bearing wear only

As receivedRead from the channel history and the asset register — nothing on this side is scored by the agent.

Signals used Vibration channel history Motor-current trend Work-order history

Why this rankIt ranks above other assets on signal alone, and an engineer still decides.

ActionApproveAmendSend to condition monitoring

What the score decidesBelow the threshold the flag gets a condition-monitoring read before it reaches the engineer.

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value landsValue 01 – 04

Every monitored assetFrom the instrumented channels

03Ranking

Rank on the signal

Draw on the channel history, the work orders and the site's existing PM regime.

01Approved path

Rank the work, do not clear it

The queue arrives ordered, so the first visit of the day is the best-evidenced one.

02Human review

Name what the sensors cannot see

Not age-related is not the same as predictable — failure modes with no measurable precursor are named as gaps, not ranked low.

04Build an evidence trail

The signal, the window it covers and the technician who acted on it stay on the asset record.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

Condition monitoringSKF · Emerson AMS
Fluke · National Instruments

Asset and maintenanceIBM Maximo · SAP PM
Infor EAM · Fiix

Field serviceServiceMax · IFS
Salesforce Field Service

Agent

Field service & predictive maintenance

Reads the channels
Ranks the assets
Holds for approval

Industrial telemetryOPC UA · MQTT
Historian exports

Observability & evaluationOpenTelemetry · Langfuse
Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and a dispatch

The layers sit one inside the next. What none of them catches is in the map below.

L6 · Outermost — last line of defenceInward → L1 · closest to the model

L6Rollback / safe modePull ranking back to raw condition reporting when evaluation or production signals degrade.Roll back

L5Version monitoringTrack model, threshold and channel-mapping changes, and any learning after release.Track

L4TraceabilityRecord the channels, the window, the rank, the decision and the outcome.Record

L3Engineer approvalHold flags for the named engineer; it governs dispatch, not whether the flag was right.Gate

L2Policy guardrailsTest flags against configured coverage, consequence and threshold rules; a failure returns the flag.Restrict

L1Confidence thresholdsRoute low-confidence flags to a condition-monitoring read before the engineer sees them.Require review

Model coreFlag produced — rank, channel evidence, coverage gaps and confidence

L1 – L2Test whether a flag may stand

L3Puts the dispatch in an engineer's hands

L4 – L5Hold the signal the flag was drawn from

L6Falls back to raw condition reporting when signals degrade

How Nestack evaluates it

Accuracy is the wrong measure here — evaluate precision and lead time.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the queue the field team works

Depth of coverage ▼

E1Final-output evaluationOf the assets the agent flagged, how many actually failed?

E2Step-level evaluationDid the agent use the right channels, thresholds and asset configuration?

E3Tool evaluationDid it read the correct asset and the correct channel?

E4Confidence calibrationDo low-confidence flags actually produce more wasted trips?

E5Slice evaluationHow does precision change across specific asset classes?

E6Business outcomeHow many dispatches were raised that a technician found nothing to do on?

Floor — the outcome the plant answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, placed at the stage each one originates.

Agent lifecycleDirection of processing →

01 · Retrieval1 mode

WK-03

Stale channel data

Signal read from a sensor that has drifted or failed.

Stage gathersChannel history, thresholds, work orders and registers

02 · Reasoning2 modes

WK-04

Base-rate collapse

A rank is scored at a base rate the fleet does not have.

WK-06

Regime-shift misread

A changed duty cycle is read as degradation.

Stage proposesRank, coverage gaps, lead-time window and confidence

03 · Tool / write2 modes

WK-02

Uninstrumented failure mode

A failure with no precursor is ranked as healthy.

WK-05

Repeat flag on one asset

An asset is flagged again before the last trip closes.

Stage writesOnly where write access and approval policy allow it

04 · Output1 mode

WK-01

Wasted dispatch

A technician arrives and finds nothing to do.

Stage returnsThe ranked queue the engineer and field team see

05 · Change / Version1 mode

WK-07

Silent threshold drift

A model or threshold change widens what the agent flags.

Stage tracksModel, thresholds, channel mapping and regime config

Sev-1 · work dispatched outside the boundary Sev-2 · a real failure is ranked as healthy Sev-3 · channel degrades, flag routes to review

See the complete Electronics failure-mode catalogue →

Affected slices

Precision follows the base rate, not the model

Base rates differ by asset class, and precision follows the base rate more closely than it follows the model. Nestack reports the wasted-dispatch rate by class, not only in total.

Slice performance — reported separately, not only in aggregateIllustrative example

Slice	Failure rate	Lift	Status
Assets with sparse history	8.0%	3.2×	Review
Newly commissioned assets	5.8%	2.3×	Review
Power electronics	3.5%	1.4×	Watch
Rotating machinery	1.8%	0.7×	Normal

Bar: wasted-dispatch-rate lift vs. rotating-machinery baseline · scale 0–4.0× · tick marks the 2.0× review threshold 2 of 4 slices over threshold

Evidence-linked improvement

A cycle ends in a case, not a cause

The cycle ends in a regression case, not in a meeting about what happened. That suite is what the next flag raised on an asset is measured against.

Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn

01Detect

Wasted-dispatch rate rises in an asset class.

02Diagnose

Not the model first — an engineer reads the channels behind the flags until the cause narrows to one.

03Improve

Every change is versioned against the flags that exposed it.

04Verify

A failing case holds the release back.

05Learn

The case joins the suite for good, and the flag thresholds are revisited.

Learn → DetectThe return edge. The next flag is measured against a longer suite than this one.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, channels, ranking workflow, evaluation, integration, then production validation and handover.

Workstream Week 1Week 2Week 3Week 4Week 5Week 6

01Maintenance workflow discovery and boundary definition.

02Condition and CMMS source assessment.

03Existing PM regime, consequence and threshold mapping.

04Channel ingestion and normalisation.

05Ranking logic and evidence binding.

06Confidence scoring and flag routing.

07Engineer approval workflow.

08Field-service-system integration.

09Precision and lead-time cases.

10Guardrails and dispatch controls.

11Asset-trail instrumentation.

12Deployment, documentation and Agent Care handover.

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tier PilotOne site, one asset class ProductionProduction field-service systems AdvancedMultiple sites / regions

Introduced at Pilot

Ranking to your channels and regime✓✓✓

Engineer approval✓✓✓

Flag-precision baseline✓✓✓

Introduced at Production

Reporting by asset class—✓✓

Approval workflow in your systems—✓✓

Approved write-back—✓✓

Field-service-system integration—✓✓

Introduced at Advanced

Multi-regime and multi-standard rules——✓

Multi-stage engineering approvals——✓

High asset volume——✓

Multi-site dispatch controls——✓

Build price From $5,000 From $8,000 Custom quote

Final build priceConfirmed after discovery based on integrations, workflow complexity, asset volume, approval controls and deployment requirements.

Separate from buildBuild pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

You bringWe build with it

01Your instrumented channels and asset register → Channel ingestion and asset mappingWeek 1

02Representative failure and work-order history → Ranking baseline, threshold extraction and evidence bindingWeek 2

03Your existing PM regime and consequence classes → PM-regime, consequence and threshold-rule mappingWeek 1

04Access to relevant APIs, feeds or exports → Condition, CMMS and field-service assessment, then integration setupWeek 2

05Dispatches that should not have been raised → Precision cases and the evaluation suiteWeek 4

06What must reach a qualified person before work starts → Confidence scoring, flag routing, guardrails and approval controlsWeek 3

07Named reliability engineers to approve flags → Engineer approval workflow, then pilot and production validationWeeks 5–6

Nothing else is required Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

The bands follow the real work, which is why evaluation and pilot share the fifth week.

Phase W1W2W3W4W5W6

Discovery W1

Build W2 – W3

Evaluate W4 – W5

Pilot & Launch W5 – W6

Week focus W1Maintenance workflow discovery, regime mapping and the automation boundary W2Channel integration and the ranking baseline W3Ranking workflow, confidence logic and approval controls W4Evaluation suite, coverage-gap checks and failure-mode testing W5Field-service integration, pilot assets and targeted corrections W6One maintenance cycle monitored under the reliability team, then handover

Reading the bandA bar covers the weeks its work is named in, and nothing else. The week 5 overlap is real, not padding.

At the end of W6The cycle closes validation and Agent Care assumes monitoring.

DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Electronics AI agent

Build a field-service agent around the maintenance regime you already run.

Show us your channels, your PM regime and who approves work. Rank it wrong and you either burn technician days on healthy machines or hear about the failure from the floor.

Discuss this agent →

Nestack Agents · Field service & predictive maintenanceAGT-EL-08 · Agent Care available after launch