Plant Snapshot: The assets and signals you actually care about

• How often do unplanned failures on your critical assets occur in a typical year? Options: Multiple times a month, Monthly, Quarterly, A few times a year, Less than once a year, Unknown
• Which asset classes create the largest unplanned cost for your site? Options: Rotating equipment (pumps, motors, gearboxes), Compressors, Heat exchangers, Process vessels and piping, Electrical assets, Valves and actuators, Other
• Tell me about the most recent failure on your site that cost over $100,000, including the asset, failure mode, and impact.
• Do you currently have vibration, temperature, pressure, current, or other sensors on those assets? Options: Vibration accelerometers, Velocity sensors, Temperature probes, Pressure transmitters, Current transformers, Proximity probes, No, limited sensors, Other
• Which historian or data archive categories hold the longest records for those assets at your plant? Options: Process historian, Edge/plant historian, Cloud telemetry/archive, PLC logs, SCADA historian, Maintenance and work order logs, None/limited

Where maintenance tools fall short for you

• If one reliability metric on your site had to improve this quarter to prove value, which would it be and why? Options: Reduce unplanned downtime, Lower emergency work orders, Reduce maintenance cost per unit, Increase mean time between failures, Increase prediction lead time, Reduce false positive rate
• How many false alerts per month does your maintenance team tolerate before trust erodes? Options: 0-1, 2-5, 6-10, 11-20, More than 20, Unknown
• Who loses credibility internally on your team when an alert turns out to be a false positive? Options: Reliability engineers, Maintenance supervisors, Operations managers, Plant manager, Technicians, Procurement, Other
• Describe your current workflow from anomaly detection to planning a corrective work order.
• If alerts came with clear failure mode context and a recommended work scope, what would stop your planners from acting within 48 hours?

Failures you keep seeing and the data behind them

• Name the recurring failure mode on your assets that surprises you most despite inspections and maintenance.
• Estimate the number of distinct failure events of that type your team has logged in the last 24 months. Options: 0-2, 3-5, 6-10, 11-20, More than 20, Unknown
• Tell me which sensor trends or historian signals on your assets you think showed the earliest signs of that failure.
• When was the last time your team could not find enough labeled failure examples to train a model for an asset? Options: Within last 6 months, 6-12 months ago, More than 12 months ago, Never, Unknown
• Would having access to 10 past failure incidents from your plant with full sensor traces remove the biggest technical barrier to predicting that failure? Options: Yes, No, Maybe

Where integrations and approvals live in your world

• Who signs the go ahead in your organization when a predictive alert would trigger a CMMS work order? Options: Maintenance supervisor, Reliability lead, Operations manager, Plant manager, Production scheduler, Other
• List the CMMS, SCADA, or historian endpoint categories at your site that must be connected for the pilot to be meaningful. Options: CMMS API or connector, CMMS file import, Process historian API, SCADA historian, OPC UA endpoint, Edge gateway stream, Cloud telemetry platform, Other
• Does your CMMS support API based work order creation today, or would the integration rely on file drops or manual entry? Options: API based work order creation, File based import/export, Email driven/manual entry, No integration available, Unknown
• Name the individual roles in your organization that own those system credentials and will approve external access. Options: IT/OT manager, Security officer, Plant manager, Reliability engineer, Maintenance manager, Procurement
• Would a security or vendor onboarding review lasting longer than 6 weeks stop your pilot timeline? Options: Yes, No, Depends on scope, Unsure

What's getting in the way — deal killers and showstoppers

• What single site constraint at your facility, if unresolved, would force you to cancel this project now?
• Estimate how long it typically takes your IT or OT team to provision external read only historian access. Options: Less than 1 week, 1-2 weeks, 3-4 weeks, More than 4 weeks, Unknown
• Describe any internal headcount or skills gaps on your team that would slow model training, integration, or validation.
• List the regulatory, safety, or union approvals at your site that could gate a pilot and who owns them. Options: IT security review, Plant safety review, Union sign off, Environmental compliance, Legal/data privacy review, Other
• Is there a hard budget cap or procurement rule in your organization that would prevent piloting without a purchase order in place? Options: Yes, No, Not sure

Other options on the table and what keeps you with the status quo

• Identify the external providers your team is evaluating for predictive maintenance or whether you are weighing an internal build. Options: Third party vendors, Internal build by your team, Incumbent vendor, Multiple external vendors, No alternatives being evaluated
• Point to the specific features or proof points those alternatives offer that would make you stay with them instead of switching. Options: Prediction accuracy from pilots, Low false positive rate, Clear diagnostic context, Direct CMMS integration, Edge deployment option, Fast onboarding time, Lower price, Reference customers
• Does anyone inside your organization believe these problems are solvable without an outside vendor or partner? Options: Yes, a team proposes internal build, No, external expertise needed, Mixed views across teams, Unsure
• Specify the single measurable improvement your current approach would need to demonstrate for you to keep it as the chosen path. Options: Reduce unplanned downtime by %, Cut emergency work orders by %, Improve detection lead time by days, Reduce maintenance spend by %, Other
• Provide the fastest realistic timeline your organization could use to halt evaluations and proceed if a pilot meets your acceptance metrics. Options: Immediately, Within 1 week, Within 2-4 weeks, 1-3 months, Longer

Are you ready to run a pilot? The technical checklist

• Identify the single integration dependency at your site most likely to delay the pilot beyond 8 weeks. Options: No historian access, No CMMS API, Security onboarding delays, Missing critical sensors, Edge compute constraint, Other
• Do you have a named data owner who can grant historian extracts, sensor mappings, and maintenance logs for the assets in scope? Options: Yes, Partially, some owners, No, Unknown
• Provide the formats and interfaces where your time series data lives, for example raw CSV exports, OPC UA, historian API, or edge gateway streams. Options: CSV exports, OPC UA, Historian API, Edge gateway streams, PLC logs, Cloud telemetry, Other
• Are there edge compute nodes in your environment that can run models, or must everything run in the cloud? Options: Edge nodes available to deploy, Cloud only, Hybrid (some edge, some cloud), Unknown
• Please give the role and contact method for the day to day person in your organization responsible for connectivity and security during the pilot.
• Assuming historian extracts cannot be provided within two weeks, can your team support a parallel approach using raw sensor exports for the pilot? Options: Yes, No, Maybe with limitations, Unsure

When a pilot deserves a yes: metrics that seal the deal

• Specify the minimum average lead time and maximum false positive rate your team requires to accept the pilot. Options: Lead time >= 30 days and FPR <= 2%, Lead time >= 14 days and FPR <= 5%, Lead time >= 7 days and FPR <= 10%, Custom
• Choose whether your team needs diagnostic context on every alert or only on a percentage of high severity alerts. Options: Every alert needs full diagnostic context, Top 10% severity alerts, Top 25% severity alerts, Only high confidence alerts, Other
• Select the pilot duration your team considers statistically meaningful, 4 weeks, 8 weeks, or 12 weeks. Options: 4 weeks, 8 weeks, 12 weeks, Custom
• Give the acceptable false positive rate in the terms your planners use, for example one per 100 assets per month. Options: 1 per 100 assets per month, 1 per 50 assets per month, 1 per 10 assets per month, Zero tolerance, Other
• Report the approvers in your organization who must sign deployment approval and the target time frame for signing if the pilot meets acceptance metrics.

The bottom line: cost levers and measurable impact

• Point to the single financial metric your leadership will use to judge this project, reduced reactive spend, avoided downtime, or increased throughput. Options: Avoided downtime cost, Reduced reactive maintenance spend, Improved throughput or yield, Lower spare parts inventory, Other
• Give your best estimate for the average cost of a critical asset failure on your site in USD. Options: Less than $50,000, $50,000 to $100,000, $100,000 to $250,000, $250,000 to $1,000,000, More than $1,000,000, Unknown
• State the number of high criticality incidents per year on your site for which a predictive win would materially reduce spend. Options: 0-2, 3-5, 6-10, 11-20, More than 20
• Select the benefits your organization would prioritize in any business case, faster mean time to repair, fewer emergency work orders, or reduced spare parts inventory. Options: Faster mean time to repair, Fewer emergency work orders, Reduced spare parts inventory, Lower maintenance headcount, Longer asset life, Other
• Report the budget owner role in your organization and the typical approval lead time to release funds if the pilot meets the business case assumptions. Options: Operations, Finance, Plant manager, Reliability, Procurement, Other

If we agreed today, what's the fastest path to a pilot?

• Assuming a minimal scope pilot, what is the earliest calendar date your team could start given security, data access, and operations constraints? Options: Immediately, Within 1 week, Within 1 to 2 weeks, Within 3 to 4 weeks, More than 4 weeks
• Recommend three assets from your site you would pick to maximize signal and minimize integration effort for a first pilot.
• Please identify the roles from your side to include in the kickoff, covering operations, IT, and maintenance. Options: Operations lead, Reliability engineer, IT/OT lead, Maintenance supervisor, Security officer, Procurement
• Choose whether your team prefers a limited scope pilot focused on prediction quality or a run to failure pilot that also tests end to end CMMS integration. Options: Limited scope: prediction quality, Run to failure: end to end CMMS integration, Hybrid, Undecided
• What would need to happen in your organization in the next two weeks to get the pilot statement of work signed?

• Solution Walkthrough Session

• You confirm the demonstrated model-to-asset mappings would detect the failure mode you described with the lead time you need.
• Confirm the current state and its cost
• Run a sample model run on the provided historian extracts and deliver accuracy and false-positive estimates prior to the pilot kickoff.
• You confirm the alert context and integration pattern supply the diagnostic detail required to create correct CMMS work orders and preserve planner trust.
• Provide 30-90 days of historian data, recent maintenance logs, and a representative failure incident for each selected asset example.
• Map your asset examples to model types
• Demonstrate diagnostic alerts and integration patterns
• Identify 1-3 pilot assets and the primary site owner contacts for each asset, including CMMS endpoint details.
• You agree on the remaining evidence and data access needed to run a pilot that will validate prediction accuracy and false-positive rate.
• Agree on pilot acceptance metrics including required lead time, acceptable false-positive rate, and diagnostic value criteria.
• Validate that this matches your needs

• Solution Walkthrough Session
• Solution Walkthrough Deck
• Solution Brief — Solution Walkthrough

• meeting
• slides
• document

• Ingest Sensor and Historian Data
• Preprocess and Signal-Process Time Series
• Train Equipment-Specific Failure Models
• Deploy Edge Inference Agents
• Cloud Model Hosting and Serving
• Anomaly Detection and Alert Generation
• Failure-Mode Classification and Diagnostics
• Remaining Useful Life Estimation
• CMMS Integration for Work Orders
• Maintenance Record and CMMS Data Ingestion
• Alert Management with Diagnostic Context
• Asset Health Dashboards and Plant Scorecards
• API Connectors for IIoT and SCADA
• Automated Model Retraining and Drift Handling

Ingest Sensor and Historian Data

• Which asset tags and telemetry streams (example: motor_vibration_axis1, bearing_temp, suction_pressure) should we ingest for the pilot assets?
• How many months of historian archives for each pilot asset are available (specify per asset tag if varying)? Options: <3 months, 3-6 months, 6-12 months, 12+ months
• Do any of your sensors or PLC/RTU tags use nonstandard naming conventions that require a mapping file (for example: site_tag -> asset_tag)? Options: Yes, No
• Provide the typical sampling rates for each sensor type involved in the pilot (examples: vibration 12 kHz, temperature 1 Hz, pressure 1 Hz).
• Identify the historian endpoints and access method you can provide (examples: historian archive export, ODBC database snapshot, time-series API) and the estimated timeframe to grant read access. Options: Historian export (SFTP), Time-series API, ODBC/SQL export, Direct historian connector, Other

Preprocess and Signal-Process Time Series

• Which preprocessing steps are required to match your operations data (examples: timezone normalization, gap-filling strategy, unit conversions like psi->bar)?
• How should we handle sensor bursts or intermittent sampling on the pilot assets (options: aggregate to 1-min, flag gaps, request raw segment pull)? Options: Aggregate to fixed window, Flag and skip gaps, Request raw segment pull, Other
• Specify any signal-processing transforms you need preserved for diagnostics (examples: FFT spectra, envelope analysis, cepstrum) for vibration or acoustic sensors. Options: FFT spectra, Envelope analysis, Cepstrum, Time-domain features only, Other
• Are there known calibration offsets or sensor relocations in the historian that require timestamped correction records? Options: Yes, No
• Indicate the acceptable latency for preprocessing batches to be available to model training (examples: hourly, daily, weekly). Options: Near real-time (seconds-minutes), Hourly, Daily, Weekly

Train Equipment-Specific Failure Models

• List the specific failure modes and their historical evidence you want models trained for (examples: bearing outer race brinelling, pump cavitation, heat-exchanger fouling with recorded work orders or root-cause reports).
• How many confirmed failure events (with timestamps and corrective work order) exist in your records per targeted failure mode and asset type? Options: 0, 1-2, 3-5, 6+
• Specify the minimum model performance thresholds you will accept from the pilot for each failure mode (examples: true positive rate >= 75%, false positive rate <= 5%, lead time >= 7 days).
• Who in your organization will approve model features or derived signals (examples: rotating equipment SME, reliability engineer) and provide subject-matter feedback during training iterations?
• Describe any asset-specific constraints for training (examples: variable-speed pump duty cycles, repeated startup transients, seasonal fouling) that must be encoded in the training labels or windows.

Deploy Edge Inference Agents

• Which sites or asset groups require edge inference rather than cloud-only inference (provide asset IDs or plant areas)?
• Specify the edge hardware profile available at the site (examples: industrial PC with 4 CPU cores, 8 GB RAM; gateway with ARM CPU) and whether OT network access allows agent install.
• Do you have existing OT change-control or security policies we must follow for edge installation (examples: maintenance window, network isolation, code signing requirements)? Options: Yes, No
• Indicate the acceptable inference latency at the edge for alerts from a given sensor stream (examples: <1s for protective, <1 min for condition monitoring). Options: <1 second, <1 minute, 1-15 minutes, Near real-time not required
• What evidence will validate a successful edge deployment on site (examples: agent connected to historian, inference logs for 72 hours, sample alert with diagnostic payload)?

Cloud Model Hosting and Serving

• Which cloud tenancy model do you require for model hosting (examples: single-tenant VPC, shared tenancy with logical separation)? Options: Single-tenant VPC, Shared tenancy with separation, On-premises only, Undecided
• Specify the maximum acceptable model response time for an API call that returns a prediction for an asset (example: 200 ms, 1s, 5s). Options: <200 ms, <1 second, <5 seconds, <30 seconds
• Do you require model versioning and rollback controls tied to a CI/CD pipeline or is manual promotion sufficient? Options: Automated CI/CD with versioning, Manual promotion only, No preference yet
• Identify any data residency or encryption requirements for hosted model inputs and outputs (examples: at-rest AES-256, region-restricted storage).
• Are there specific monitoring and audit logs you require for model serving requests (examples: per-request latency, user ID, asset ID, prediction score)? Options: Yes, No

Anomaly Detection and Alert Generation

• Which anomaly thresholds or scoring semantics should be used to trigger an alert for each sensor type (examples: envelope RMS > X, sudden spectral energy increase of Y dB)?
• How should alerts be prioritized and routed (examples: high for immediate planning, medium for next shift, low for trending) and which plant roles receive each priority?
• Do you require a suppression window or cooldown after an alert to avoid repeated notifications for the same condition (example: suppress identical alert for 24 hours)? Options: No suppression, 1 hour, 8 hours, 24 hours, Custom
• Specify the minimum lead time you require from an alert to when maintenance must be scheduled for the asset class (examples: 3 days, 2 weeks). Options: <24 hours, 1-3 days, 3-7 days, >7 days
• Indicate any regulatory or safety conditions under which alerts must escalate immediately (examples: rotating equipment overspeed events, high-pressure venting).

Failure-Mode Classification and Diagnostics

• For each targeted failure mode, list the diagnostic artifacts you expect with the classification (examples: spectral peak at 1x rpm, imbalance severity, fouling factor estimate).
• How should diagnostic confidence be represented in alerts (examples: probability score, top-3 ranked causes, required human review flag)? Options: Probability score, Top-3 ranked causes, Human review required, Other
• Do you have recently completed root-cause analysis reports or teardown photos that can be used to validate classifier outputs? Options: Yes, No
• Specify any domain rules to apply to diagnostics (examples: ignore transient startup spikes, exclude data during known maintenance windows).
• Who on your team will be the diagnostic SME to resolve classifier disagreements and label additional examples during model iteration?

Remaining Useful Life Estimation

• Which asset types require remaining useful life (RUL) estimates rather than binary alerts (examples: heat exchanger fouling, compressor bearing wear)?
• What planning horizon do you require for RUL outputs (examples: weeks, months) and how should uncertainty be expressed (examples: median plus 90% CI)?
• Do you have periodic inspection records or non-destructive test (NDT) results that can be used as RUL ground truth for calibration? Options: Yes, No
• Specify how RUL estimates should feed planning workflows (examples: convert to recommended lead-time for spare parts procurement, create CMMS preventive tasks).
• Are there hard operational thresholds after which RUL is no longer useful (examples: component must be replaced at 80% wear regardless of RUL)? Options: Yes, No

CMMS Integration for Work Orders

• Which CMMS endpoints and methods can you provide for work-order creation (examples: REST API, SOAP API, file drop, database direct insert)? Options: REST API, SOAP API, File drop (CSV), Database insert, Other
• List the CMMS work-order fields required by your planners that must be populated from an alert (examples: asset ID, failure code, priority, estimated labor hours).
• Will new work orders created by the platform follow an approval workflow in your CMMS, and if so what is the typical approval path and SLAs? Options: No approval required, Single approver, Multi-step approval, Undecided
• Specify the acceptance criteria we must meet for successful CMMS integration (examples: work order created with all required fields, attachment of diagnostic PDF, API success response within 5s).
• Who is your CMMS technical contact that will provide sandbox credentials and map asset IDs between historian tags and CMMS asset records?

Maintenance Record and CMMS Data Ingestion

• Which historical maintenance record tables or exports are available for ingestion (examples: work_order_history.csv, labor_hours table, spare_parts usage)?
• How complete are your maintenance records for pilot assets (percentage of events with causal codes and timestamps)? Options: <50%, 50-75%, 75-90%, >90%
• Do maintenance records include post-failure root-cause tags or manufacturer repair codes that can be used to label training events? Options: Yes, No
• Specify any retention or PII constraints for maintenance logs that affect how long we can store and process ingested records.
• Who will provide the mapping between CMMS asset IDs and historian asset tags for join operations during model training?

Alert Management with Diagnostic Context

• Which communication channels should alerts and diagnostic context be delivered to (examples: email to reliability team, SMS for duty engineer, CMMS work order creation)? Options: Email, SMS, CMMS work order, Web dashboard, API webhook
• How much diagnostic detail must accompany each alert for the maintenance planner to act (examples: short summary + 1-page diagnostic, full spectral plots, recommended spare parts list)?
• Do you require templated alert text or structured fields to be inserted into your planning workflows (examples: prefilled failure code, estimated downtime)? Options: Yes, No
• Identify who on your team will be the primary alert recipient for the pilot and who should be CCed for visibility (roles, not names).
• Are there any shift or union rules that influence when alerts may be acted on (examples: no nonessential work after 10pm, permit-to-work requirements)? Options: Yes, No

Asset Health Dashboards and Plant Scorecards

• Which KPIs do you want on the plant scorecard for pilot assets (examples: percent of assets with high risk, mean remaining useful life, monthly false-positive rate)?
• How frequently should dashboards refresh and who needs access (examples: shift leads view live, managers weekly report)? Options: Real-time, Hourly, Daily, Weekly
• Do you require role-based dashboard views (examples: planner view, reliability engineer view, plant manager view) with different data columns? Options: Yes, No
• Specify any export or data-slice capabilities you need from dashboards (examples: CSV export of high-risk assets, PDF diagnostic snapshot).
• Are there legacy reporting formats or scorecards we must reproduce to match internal governance reports? Options: Yes, No

• decision_readiness
• success_criteria
• stakeholders
• gaps
• desired_state
• current_state
• stakeholders
• decision_readiness
• success_criteria
• current_state
• desired_state
• gaps
• stakeholders
• gaps
• current_state
• desired_state
• success_criteria
• decision_readiness
• decision_readiness
• decision_readiness
• decision_readiness
• decision_readiness

• Master Services Agreement (MSA)
• Statement of Work (SOW)
• Subscription Agreement
• Order Form — Commercial Terms
• Service Level Agreement (SLA)
• Data Processing and Access Agreement (DPA)
• Acceptance Criteria & Handoff Annex
• Governance & Escalation Charter
• Change Order Agreement
• Work Product and IP License Exhibit
• Security and Regulatory Compliance Addendum
• Termination and Transition Services Annex

• Go-live Health Check
• First Measurement Review
• Acceptance Gate Review (90-day)
• Quarterly Success Review and Continuous Improvement

• Close or reassign open issues older than 30 days and update the shared tracker with expected resolution dates.
• Schedule any model retraining or threshold adjustments and document the validation protocol to be used before the acceptance gate.
• Restate Pilot Evaluation acceptance criteria and numeric targets
• Each acceptance criterion recorded in Pilot Evaluation marked pass or fail and the buyer's acceptance decision documented.
• Remediation plan with specific resolution items and dates captured for any failed criterion.
• Incumbent system decommissioning status confirmed or a retention/archival plan recorded.
• Publish the acceptance gate decision and criterion-level results to the shared workspace.
• Track and publish remediation items with resolution timelines until all acceptance criteria are satisfied.
• Confirm and document the incumbent system's decommissioning or retention-read-only plan and archive status.
• Production performance trends
• Confirm production prediction performance remains within acceptable bounds versus the numeric targets recorded in Pilot Evaluation or surface required remediations.
• Reduce the count of open issues older than 30 days and lower weekly false-positive alerts.
• Agree a prioritized quarterly improvement backlog with validation windows and deployment dates.
• Publish the quarterly performance dashboard and a short root-cause memo for any regressions.
• Prioritize the model improvement backlog and schedule retraining or A/B test windows.
• Re-confirm success criteria and ownership
• All core integrations (historian, SCADA, CMMS) verified as connected and returning expected data.
• Early operational users have access and have acknowledged at least one platform alert or work order.
• All critical blockers logged with a target resolution window and tracking owner.
• Publish the deployment checklist completion status to the shared channel within 24 hours.
• Provide an export of user accounts, roles, and first-use activity for adoption analysis.
• Resolve critical integration failures and document remediation steps and validation tests.
• Present first-window prediction performance
• Determine whether prediction true positive rate and false positive rate are trending toward the numeric targets recorded in Pilot Evaluation.
• Confirm CMMS work-order automation success rate and whether integration fixes are required prior to the acceptance gate.
• Agree a concrete remediation plan and timeline to reach the Acceptance Gate.
• Deliver the raw event-level dataset and labeling rules used to compute prediction metrics for independent validation.
• Produce an integration error log and proposed fixes for CMMS work-order failures.
• Present outcome data against each acceptance criterion
• Operational KPI review
• Deployment and integration validation
• Review CMMS work-order automation performance
• Document pass/fail per criterion and capture acceptance decision
• Early adoption and usage signals
• Open issues and incident burn-down
• Root-cause diagnosis for metric gaps
• Blockers and open issues triage
• Agree corrective actions and timeline to acceptance gate
• Model refinement and deployment pipeline
• Agree remediation plan for any failed criteria
• Agree immediate remediation actions
• Incumbent system decommissioning and data migration status
• Quarterly action plan and next checkpoints