Roughly 70% of manufacturing predictive maintenance pilots never scale beyond the first dozen assets. Not because the technology fails — because the pilot was never designed to prove anything. No scope boundary, no baseline, no success metric, no path to production. This guide walks the 90-day pilot that gets to a credible go/no-go decision, with every reading, every alert and every validated catch held against the asset. OXMAINT AI is the AI-powered CMMS and maintenance management software built to run the pilot and the scale-up on one platform — so the proof travels directly into production.
PREDICTIVE MAINTENANCE PILOT · MANUFACTURING IMPLEMENTATION · 2026
Predictive Maintenance Pilot: The 90-Day Manufacturing Implementation Guide
A pilot that proves nothing wastes 90 days and a quarter's budget. One that is scoped tight, grounded in a baseline, measured against clear metrics and validated against actual failures travels straight into production. The OXMAINT AI maintenance management software runs the whole 90 days on one platform — asset register, condition data, alerts, work orders, closeout evidence — so the result is a decision, not a debate.
1Plan
2Scope
3Instrument
4Baseline
5Measure
6Validate
~70%
of manufacturing PdM pilots never scale — the technology works; the design doesn't
5-15
tier-A assets is the right pilot size — small enough to control, big enough to prove
90 days
end-to-end: plan, instrument, baseline, measure, validate, decide to scale
1
validated catch (predicted failure confirmed by physical finding) outweighs ten dashboard alerts
The 90-Day Pilot Timeline — Week by Week
A good pilot is a scheduled research project, not an open-ended rollout. The 12 weeks below carry a clear objective and exit criterion per phase. OXMAINT AI holds the plan as live milestones, so a weekly stand-up is a glance at the board, not a status rebuild — schedule a walkthrough of the 90-day pilot plan in OXMAINT AI.
Phase
W1W2W3W4W5W6W7W8W9W10W11W12
PlanCharter, team, budget, success metrics
ScopePick 5-15 tier-A assets & failure modes
InstrumentFit sensors; route data into platform
BaselineLearn normal operating bands per asset
MeasureAlerts active; track catches, misses, FPs
Validate & DecideGo / No-Go review & scale plan
What Happens in Each Phase
Each of the six phases has one job and one deliverable. Skip a phase — or paper over it — and the final go/no-go has no evidence to lean on. The OXMAINT AI software holds the deliverable per phase against the pilot charter, so progress is auditable instead of asserted.
01
Plan
Pilot charter signed by maintenance, operations and finance. Named sponsor, named lead, pre-agreed budget and success metrics. Without this, the project is a hobby.
Deliverable: signed charter + metric targets
02
Scope
5-15 tier-A assets picked using the asset scorecard. Failure modes to target are named explicitly — pump bearing wear, motor winding, gearbox tooth, fan imbalance.
Deliverable: pilot asset list + target failure modes
03
Instrument
Right sensor on the right mounting point for each failure mode. Data path to OXMAINT AI confirmed; vendor portal left as drill-down only.
Deliverable: live sensor feeds per asset
04
Baseline
Each asset runs under its real duty cycle for two to three weeks so OXMAINT AI learns the normal operating band per load point. Alerts stay off during this window.
Deliverable: baselines + threshold policy
05
Measure
Alerts active for four weeks. Every alert tracked as catch, miss or false positive against the physical finding. Work orders raised from alerts and closed on evidence.
Deliverable: catch/miss/FP ledger per asset
06
Validate & Scale
Metrics compared against the charter targets. Go / Hold / No-Go called on evidence. If Go, scale plan ties next 50-200 assets to the same workflow.
Deliverable: go/no-go memo + scale roadmap
A PdM Pilot Is Research — and Research Needs a Hypothesis.
The pilots that scale are the ones that could have failed and been declared a failure. The OXMAINT AI maintenance management software is the single platform where every reading, every alert, every work order and every physical finding lives — so the final answer is honest.
Picking the Right 5-15 Assets
The single biggest reason pilots fail is the wrong assets — too many, too routine, or chosen because they already had sensors. The right pilot asset is tier-A critical, has a known or expected failure mode in the next 90 days, and has a measurable baseline. Score each candidate on the four axes below; the top-ranked 5-15 go into the pilot.
Scoring axis
Weight
What a strong candidate looks like
Production impact
High
Asset stops a line or a cell when it fails; no quick redundancy to swap in
Failure history
High
Documented repeat failures of a specific, measurable mode in the last 24 months
Sensor fit
Med
A standard sensor family (vibration, temp, current, oil) catches the target failure mode cleanly
Operating stability
Med
Runs regularly under a stable enough duty cycle to baseline inside the pilot window
The Five Success Metrics That Decide Go / No-Go
Pilot success is not a feeling — it is five metrics measured against targets agreed in the charter. OXMAINT AI logs each catch, miss and false positive against the asset so these numbers roll up automatically. Scoring 4 of 5 on target is a confident go; 2 of 5 is a no-go whatever the demo looked like.
01Catch RateTarget ≥ 70%
Of real failures that developed in pilot assets, how many did the system catch early? Measured against physical finding, not alert count.
02Lead TimeTarget ≥ 7 days
Average days from alert to the failure the alert predicted. A 2-day lead on a pump bearing is good; a 2-hour lead is useless to planning.
03False-Positive RateTarget ≤ 20%
Alerts that led to a wasted inspection with no finding. Low FP is what keeps ops acting on the next alert instead of ignoring it.
04PM-to-Corrective ShiftTarget: PM share up
Movement in planned-vs-unplanned work mix across pilot assets over the measurement window. A real shift here is the business case.
05Avoided DowntimeTarget: > break-even
Estimated downtime avoided by acting on confirmed catches, costed against the pilot's sensor and platform spend. The real ROI test.
The Pilot Readiness Checklist
Before week 1, these items need to be true. Each one looks small; collectively they are the difference between a pilot that produces evidence and one that produces slides. OXMAINT AI holds the checklist against the pilot charter, with every box timestamped as it closes.
The Week-12 Go / Hold / No-Go Gate
A pilot ends in one of three honest answers — and the right one is dictated by the scorecard against the charter targets, not by how good the dashboards look. OXMAINT AI holds the metric roll-up next to the asset-level evidence, so the gate is a 60-minute meeting with one slide, not a quarter-long argument.
GO · SCALE
4 or 5 of 5 metrics on target
Lock the sensor policy, confirm scale budget, start the next 50-200 assets on the same workflow. Keep the pilot assets running under the same measurement.
HOLD · EXTEND
3 of 5 on target
Extend the pilot by 4-8 weeks on the specific metric that is short. Common fixes: refine baselines, tighten FP thresholds, add missing sensors.
NO-GO · STOP
2 or fewer on target
The pilot is honest evidence that this scope, this sensor family or this failure mode is not the right fit here. Document the learning; retire the sensors or redeploy.
Six Reasons Most Pilots Stall
Nearly every stalled pilot sits in one of six patterns. OXMAINT AI holds the pilot workflow against each so the common failure modes are caught before week 12.
TOO BIG
50+ assets in month one. Noise overwhelms the team; nothing is baselined properly; no clean evidence per asset.
WRONG ASSETS
Pilot runs on the equipment that was already instrumented, not the equipment that would prove the business case.
NO BASELINE
Alerts switched on at vendor defaults from day one. Every alert is a false positive. Operations stops listening by week 3.
NO METRICS
"Does it feel like it's working?" replaces catch rate, lead time and FP rate. The go/no-go becomes a popularity vote.
DATA ISLAND
Readings live in a vendor portal. No CMMS link. No work orders raised. A catch happens and nobody acts on it.
NO SCALE PLAN
Pilot passes; scope beyond the first 15 assets was never budgeted. Six months later the pilot is a case study, not a program.
How the OXMAINT AI Software Runs the Pilot
The charter, scorecard, baselines, alert workflow, catch/miss/FP ledger and go/no-go roll-up are all features the OXMAINT AI maintenance management software was built to run — and the scale-up uses the same platform, no re-implementation — spin up a free account and set up your pilot in OXMAINT AI.
Pilot Charter & Milestones
Charter, success metrics, phase milestones and week-12 gate held in one plan the whole team sees — progress by evidence, not slides.
Asset Scorecard
Scoring axes and weights built into the asset register — the top-ranked candidates for pilot inclusion surface automatically.
Per-Asset Baseline Learning
OXMAINT AI learns normal operating bands per load point before alerts switch on — thresholds are measured against reality, not vendor defaults.
Alert-to-Work-Order Loop
Every threshold breach opens a prioritized defect, drafts the work order with context, routes to the right crew — no catches lost in a portal.
Catch / Miss / FP Ledger
Every alert tagged as catch, miss or false positive on closure with physical-finding evidence — the single source of truth for the gate.
Scale-Up On the Same Platform
Pilot workflow carries straight into the next 50-200 assets — no re-implementation, no second system, no evidence rebuild.
Frequently Asked Questions
How many assets should a PdM pilot actually cover?
5-15 tier-A critical assets — small enough that the team can diagnose every alert, big enough to produce statistically meaningful catch and false-positive numbers across 60 days of measurement. OXMAINT AI holds the pilot asset list and the scoring behind each inclusion.
Start free and set up your pilot asset scorecard in OXMAINT AI.
Why does the baseline phase matter so much?
Alerts switched on at vendor defaults produce false positives from day one, and operations stops acting on them within three weeks. Two to three weeks of baseline learning per asset is the difference between an alert the team trusts and background noise they tune out.
What's the single most important success metric?
The catch rate — the share of real failures the system predicted ahead of time, measured against physical finding. Lead time and FP rate matter, but catch rate is the fundamental test of whether the pilot proved anything.
Can we run the pilot without buying any new sensors?
Sometimes — if existing instrumentation covers the target failure modes on tier-A assets. More often you need one or two new sensors per asset. The pilot sizing of 5-15 assets keeps that spend contained.
What if the pilot comes back as a no-go?
That's a successful pilot — honest evidence that this scope, sensor family or failure mode is not the right fit here. The learning is worth more than six months of a hopeful rollout. Document, redeploy the sensors, and refine the scope for the next attempt.
Design the Pilot So the Answer Is Honest.
Run the whole 90 days on the OXMAINT AI maintenance management software — charter, scorecard, baselines, alerts, work orders, catch/miss ledger, go/no-go gate — and scale straight through on the same platform. The pilot that proves something is the pilot that could have failed.
.png){kind=tracking}
