Start: A Quick Plant Snapshot

• Which best describes your role and primary focus at the plant? Options: Reliability / R&D Engineer, Maintenance Director / Manager, Operations / Plant Manager, Digital Manufacturing / IIoT Lead, IT/OT Architect, Other (please specify)
• What type of facility and processes are we talking about? Options: Refinery / Petrochemical, Power / Utility, Mining / Minerals, Food & Beverage / Process, Pharma / Chemical, Metals / Heavy Industry, Other
• How many critical assets (rotating machinery, heat exchangers, compressors, large pumps) do you consider in-scope for predictive maintenance today? Options: 1–10, 11–50, 51–200, 200+
• Roughly how often does an unplanned failure in a critical asset cost your site over $100k (choose best fit)? Options: Multiple times per year, Once per year, Every few years, Rarely / Never
• Who will be our primary day-to-day contact for this initiative (name / title)?

If a Machine Could Talk, Would You Trust What It Says?

• When your monitoring systems raise an alert today, how confident are you that it indicates a real, actionable degradation? Options: Very confident, Somewhat confident, Often uncertain, Usually not confident
• Tell us about the last time an automated alert led to either a false alarm or a missed failure—what happened and how did your team react?
• How tolerant would your maintenance planners be of false positives during a pilot (e.g., % of alerts that can be false before trust breaks)? Options: <1%, 1–5%, 6–10%, 11–25%, >25%
• What specific diagnostic detail would make an alert believable and actionable for your planners (e.g., likely failure mode, estimated RUL, recommended inspection steps)? Options: Likely failure mode, Estimated remaining life (RUL), Severity / confidence score, Suggested inspection checklist, Parts likely required, Other
• How long would it take to rebuild trust after a string of false alarms? Please describe past experiences if possible.

Where It Actually Hurts — A Few Stories About Real Failures

• Describe a high-impact failure that still bothers you—what failed, what was the root cause, and why does it stick in your memory?
• How often do failures like that occur for similar assets at your sites? Options: Weekly, Monthly, Quarterly, Annually, Less often
• What are the typical business consequences when a critical asset fails (select all that apply)? Options: Production loss / revenue impact, Safety incident / near miss, Regulatory compliance issue, Expedited parts / repair cost increase, Customer service impact, Other
• When those failures occur, who bears the cost or explains it to leadership—and how does that conversation usually go?
• If you could prevent just one recurrent failure across the plant, which would it be and why?

What’s Working With Your Data — And What’s Not

• You told us you have historians and sensors—how confident are you that the right signals exist to detect the failures you care about? Options: Very confident, Somewhat confident, Doubtful, Unsure
• Which of the following signals are available and consistently recorded for the assets in scope? Options: Vibration (accel/velocity), Temperature, Pressure, Current / Power, Flow, Control/SCADA tags, Maintenance logs / CMMS history, Not sure / need to check
• How complete and labeled is your incident history (i.e., records of failures with timestamps and root cause)? Options: Well labeled and timestamped, Partial labels / inconsistent, Mostly unlabeled / free-text notes, No usable incident history
• What data quality issues have you repeatedly experienced (missing tags, timestamp drift, sampling inconsistency, sensor calibration, etc.)?
• Who owns the data access and where does it live (on-prem historian, cloud lake, hybrid)? Options: On-prem historian (PI/OSI/etc.), Cloud data lake, Edge storage, Hybrid, Unsure
• If we asked for a sample dataset for a quick modeling exercise, how long would it take to provide it and who would we coordinate with? Options: <1 week, 1–2 weeks, 2–4 weeks, Longer / need approvals

Who Actually Decides — And Who Gets the 2AM Call?

• When it comes to adopting predictive maintenance, who are the decision-makers and what are their top concerns?
• Which teams must be engaged/approve for a pilot to proceed (select all that apply)? Options: Reliability / RCM, Maintenance planning, Operations, Enterprise IT, OT / SCADA team, Safety / Compliance, Procurement / Legal
• Who will be the operational owner responsible for triaging and acting on alerts during the pilot?
• How do you prefer to define success for the pilot — technical thresholds, business KPIs, or both? Options: Technical only (accuracy/FP/RUL), Business KPIs only (downtime/costs), Both technical and business
• What timeline are your stakeholders expecting for a pilot decision (weeks, months)? Options: Immediately / weeks, 1–3 months, 3–6 months, Longer / undecided

If We Could Prove Value in Weeks, What Would That Look Like?

• What measurable outcomes would make you say a pilot was unquestionably successful?
• Which performance targets matter most to you for models (pick up to three)? Options: Prediction accuracy, False positive rate, Lead time before failure, RUL granularity, Diagnostic clarity / failure mode classification, Operational integration reliability
• For RUL or advance warnings, what minimum lead time would allow planning and spare parts staging (choose all that apply)? Options: <24 hours, 1–7 days, 1–4 weeks, >1 month
• What operational acceptance tests would you run to validate the predictions in a live environment (examples: controlled inspections, non-disruptive tests, staged shutdown)?
• What KPIs will you track with leadership to justify moving from pilot to production (select up to three)? Options: Reduction in unplanned downtime, Maintenance cost savings, MTTR improvement, Spare parts inventory reduction, Work-order accuracy, Operator adoption rate

Integration & Workflow Reality Check — The Work-Order Moment

• You may have CMMS integration plans—how many manual steps are still required today from detection to scheduled work-order? Options: Fully manual, Mostly manual with some automation, Partially automated, Fully automated
• Which CMMS and IIoT/SCADA platforms do we need to integrate with for automatic work-order creation or alerts? Options: Maximo, SAP PM, Infor EAM, Oracle EAM, Custom CMMS, PI/OSI/AVEVA, Rockwell/Schneider/Siemens SCADA, Other
• Describe the ideal work-order payload and handoff from analytics to planners—what fields and context must be included for someone to act?
• What approvals or human checkpoints must exist before a generated work-order becomes scheduled or dispatched? Options: Supervisor approval, Planner review, Technician sign-off, Automatic dispatch, Other
• How would you like diagnostics presented to maintenance (e.g., prioritized root-cause likely, confidence band, suggested inspection steps, visual trend)? Options: Short text summary + confidence, Detailed signal trend visuals, Step-by-step inspection checklist, Parts list suggestion, All of the above

Training Models — Data Access, Security, and Speed

• How comfortable are you with an external team training models on your production data under an agreed security posture? Options: Very comfortable, Comfortable with NDA/controls, Hesitant / needs approvals, Not comfortable
• Which deployment model do you prefer for training and inference? Options: Cloud training + cloud inference, Cloud training + edge inference, On-prem training + edge inference, Edge-only (air-gapped)
• What security or compliance requirements must we satisfy before accessing data (e.g., network zoning, VPN, IOT gateway, SOC2, legal approvals)?
• Do you have labeled failure examples we can use for supervised training, and if so, how many incidents per failure mode are available? Options: Many (>50), Some (10–50), Few (1–9), None / unlabeled
• Who will be the owner responsible for model validation and approving model promotion to production?
• How quickly do you expect an initial proof-of-value model (non-production) to be available for review? Options: <2 weeks, 2–4 weeks, 1–2 months, Longer

What Could Stop This From Succeeding (Honest Risks)

• If this pilot failed, what do you think the real reason would be—technology, people, process, budget, or something else? Options: Technology / data quality, Lack of stakeholder buy-in, Operational workflow mismatch, Budget / commercial terms, Integration complexity, Other
• Which internal resources are likely to be the bottleneck during the pilot (e.g., data owners, OT engineers, reliability team, operators)? Options: Data / historian admins, OT / SCADA engineers, Maintenance planners, Operations supervisors, IT security
• What internal objections or concerns have you heard about predictive solutions in the past, and how were they voiced?
• What would help you mitigate those risks—specific governance, assurances, sprint cadence, or pilot guardrails?
• Who needs to be visibly successful internally to keep funding and support for a production rollout?

Commitment & Next Steps — Small Bets, Clear Signals

• What pilot duration feels reasonable to demonstrate meaningful value (and why)? Options: 2–4 weeks, 1–3 months, 3–6 months, 6+ months
• Which assets and failure modes would you want included in an initial pilot (list names or asset types)?
• What minimal SLAs or acceptance thresholds must be met to consider the pilot successful (choose up to three)? Options: Accuracy threshold, Max false positive rate, Minimum lead time before failure, Integration reliability (work-order success), Operator adoption rate
• Who will sign off on pilot completion and who will approve moving to a commercial engagement?
• What practical next step would make it easy for your team to say yes to a pilot (e.g., NDA, proof-of-concept agreement, single-asset trial)? Options: NDA and data exchange, Single-asset free trial, Paid pilot with defined KPIs, Technical kickoff workshop, Other
• Is there anything else we should know before proposing a pilot scope or a readiness checklist?

Start: Help Us See Your Plant

• Which site, plant area, and asset groups should we focus on for this discovery (pick the best fit)? Options: Refinery / Process units, Power generation / Turbines, Compressor stations, Pumps & rotating equipment, Heat exchangers / Boilers, Mining / Mineral processing, Other (describe below)
• What is the primary point of contact for reliability and who will coordinate data access?
• What timeframe would you ideally like to see results or a working pilot (business expectation)? Options: 4–6 weeks, 2–3 months, 3–6 months, 6+ months, Undecided
• Who are the stakeholders we should assume will be involved (names or roles: reliability, ops, IT, procurement)?
• Is there a compliance, audit, or corporate standard we must align to during discovery (e.g., ISA, NERC, company policy)? Options: Yes — will provide details, No formal standard, Not sure / need to check

Why do failures still blindside you?

• When unplanned failures happen, what consequences do you feel most acutely (beyond repair cost)? Options: Production loss, Safety risk, Regulatory exposure, Contract penalties, Customer dissatisfaction, Reputational damage, Other
• How often in the past 12 months did a failure occur with little or no warning? Options: Weekly, Monthly, Quarterly, A few times a year, Rarely
• Tell us about one recent failure that felt avoidable—what happened and why do you think it wasn't predicted?
• How does the maintenance team usually discover and escalate an impending failure today (e.g., alarms, operator rounds, condition monitoring, customer complaint)? Options: SCADA alarms, Operator rounds/visual, Vibration route-based CM, Automated analytics, Customer / downstream feedback, Other
• How confident are you that earlier detection could have reduced impact on that example incident? Options: Almost certain, Likely, Possibly, Unclear

Where's the data hiding (and what it's not telling us)?

• If your historians could speak, which gaps would they complain about first? Options: Missing signal types, Inconsistent sampling, Short retention, Poor naming/metadata, No labeled incidents, Access restrictions, Other
• Which historian(s) and data stores are in use at the site? Options: OSI PI, Wonderware Historian, AVEVA/GE Historian, PLC local logs, Cloud data lake, SQL/Time-series DB, Other (specify)
• Estimate the percent of your critical assets that have continuous time-series coverage (vibration/temperature/pressure/current etc.). Options: 0–20%, 21–40%, 41–60%, 61–80%, 81–100%
• What is your typical data retention window for raw sensor data? Options: <1 month, 1–3 months, 3–12 months, 1–3 years, 3+ years
• Do you have incident-labeled datasets or failure logs that can be matched to time-series signals? Options: Yes — organized and timestamped, Partially — some events labeled, No — mostly free-text reports, Not sure
• If partially or fully labeled, roughly how many failure events do you have per failure mode for a representative asset class? Options: 0–5, 6–20, 21–100, 100+

Is your instrumentation telling the full story?

• Which sensor families are installed on the assets we're discussing? Options: Vibration/accelerometers, Temperature probes, Pressure transducers, Motor current sensors, Flow meters, Speed/Tachometers, Acoustic sensors, Other
• How healthy are those sensors—do you have automated sensor health or calibration records? Options: Automated health checks & calibration logs, Periodic manual calibration, Sporadic / ad-hoc checks, No formal sensor health program
• What percent of telemetry is typically missing or flagged as bad on a daily basis for these assets? Options: <1%, 1–5%, 6–15%, 16–30%, >30%
• When a sensor fails or drifts, what is the maintenance workflow to repair or replace it?
• Are there any sensors you’ve intentionally avoided using for analytics because of noise, sampling, or access issues? Which and why?

How do work orders really get created—and do they ever complete the feedback loop?

• Who is responsible for converting a condition alert into a work order and is that person empowered to act? Options: Maintenance planner, Reliability engineer, Shift supervisor, Operations manager, Other
• Which CMMS (if any) do you use and how are new work orders typically created? Options: SAP PM, IBM Maximo, Infor EAM, Oracle eAM, Custom CMMS, No CMMS / spreadsheets, Other
• Are work orders created manually, via scripted integrations (API/DB), email triggers, or automatically from existing analytics? Options: Manual entry, API/REST integration, Database export/import, Email-to-CMMS, Automated from analytics, Other
• How long does it usually take from identifying an issue to scheduling and completing the work (typical SLA/lead time)? Options: <24 hours, 1–3 days, 3–7 days, 1–4 weeks, Longer
• Do completed work orders get linked back to timestamps/events in the historian so we can label outcomes? Options: Yes — routinely, Sometimes — inconsistent, No — not linked, Not sure
• What would increase your teams' willingness to act on a predictive alert (e.g., diagnostic confidence, estimated RUL, cost savings projection)? Options: Clear root-cause diagnosis, Estimated RUL with uncertainty, Work instructions attached, Cost/impact estimate, Operator training, Other

If a pilot had to win your trust, what would it have to prove?

• What minimum prediction performance would make you consider changing workflows (choose the most important metric)? Options: High precision (low false positives), High recall (few missed failures), Accurate RUL windows, Actionable diagnostics, Time-to-failure lead time
• What false-positive tolerance would erode trust for your maintenance team? Options: <5%, 5–10%, 11–20%, >20%, Unsure—depends on context
• Which failure modes or asset classes should be in scope for the pilot to be meaningful? Options: Rolling element bearings, Gearboxes, Seals & packing, Heat exchanger fouling, Pump cavitation, Motor winding degradation, Other
• How many assets or units would you consider a representative pilot (to show statistically meaningful results)? Options: 1–5 assets, 6–20 assets, 21–50 assets, 50+ assets, Depends on failure frequency
• What duration do you think is necessary to validate performance (lead time, seasonality, failure arrival)? Options: 4 weeks, 8–12 weeks, 3–6 months, 6+ months
• What operational acceptance tests would convince you the solution is ready to deploy (examples: live alert accuracy, correct work-order creation, operator usefulness)?

What integrations, security, or policies could stop this before it starts?

• Which deployment model is preferred or required at your site? Options: Cloud-hosted, On-premise / edge device, Hybrid (edge processing + cloud), Undecided
• Do you have network policies that restrict outbound connections, require jump boxes, or mandate air-gapped segments? Options: Open outbound allowed, VPN / jump box required, Strictly air-gapped, Varies by zone
• What authentication and identity controls are required for integrations (SSO/SAML, client certs, API keys)? Options: SSO / SAML, Client certificates, OAuth/API keys, Local accounts only, Other
• Are there procurement, legal, or data residency constraints we should know about (e.g., no cloud storage of raw sensor data)? Options: Yes — will provide policy, No constraints, Unsure / need to check
• Who in your IT/OT organization will own firewall and integration approvals?
• What formats or interfaces will we need to support for data access (PI connectors, OPC-UA, REST APIs, CSV exports, direct DB read)? Options: PI connector, OPC-UA, REST API, CSV/flat file, Direct DB access, Other

Who signs, who operates, and who will actually change behavior?

• If predictions become reliable, which role will own the final decision to act on a predictive work order? Options: Maintenance planner, Reliability engineer, Operations supervisor, Plant manager, Cross-functional committee
• Which roles will need training and what level of detail do they prefer (high-level dashboards, diagnostic runbooks, hands-on workshops)? Options: High-level dashboards, Diagnostic runbooks, Hands-on operator workshops, Maintenance planner walkthroughs, IT/Integration training
• Who will be responsible for verifying alerts and closing the loop back into our data (labeling true/false positives)? Options: Maintenance techs, Reliability engineers, Operations, Automated CMMS status sync, Other
• Is there an internal champion or sponsor who will help prioritize integration work and remove roadblocks? If so, who?
• How do you prefer ongoing collaboration during a pilot—weekly checkpoints, bi-weekly demos, or on-demand support? Options: Weekly checkpoints, Bi-weekly demos, Monthly summaries, Ad-hoc/on-demand

Quick intro — who’s joining the mission?

• Which roles from your team will be directly involved in getting models ready for pilot (select all that apply)? Options: Reliability / RCM Engineer, Maintenance Manager / Planner, Operations Supervisor, Data Engineer / IT, OT / Controls Engineer, Plant Manager, Vendor / OEM Representative, Other
• Who is the single technical owner we should coordinate with for data access and test environments?
• Who has final approval authority to run a pilot and push a model into production? Options: Plant Manager, Maintenance Director, Reliability Lead, IT/OT Governance Board, Corporate Reliability, Other
• How much weekly availability can the core team commit to for rollout coordination and troubleshooting in the next 60 days? Options: >10 hours/week, 5–10 hours/week, 1–5 hours/week, Minimal / As needed
• Is there an existing internal or external vendor we must coordinate with for access, and who are they? Options: No, we handle internally, Yes — OT/controls vendor, Yes — historian/IIoT vendor, Yes — CMMS vendor, Other (please specify)

If the data could talk, what story would it start with?

• If your sensor and historian data had to prove one thing about asset health, what would you want it to show first? Options: Early degradation trend, Imminent failure signature, Intermittent anomaly patterns, RUL estimates, Other
• Which data sources are available and how are they accessed (select all that apply)? Options: PI/OSI PI AF, Honeywell/PlantScape historian, Azure/Cloud data lake, Edge gateways / MQTT, SCADA historian, CSV exports/manual extracts, Other
• What is the typical sampling frequency for your key signals (vibration, temp, pressure, current)? Options: Sub-second / high-frequency, 1–10 seconds, 10–60 seconds, 1–5 minutes, Irregular / event-based
• Do you have a single canonical timestamp and asset identifier across systems, or will we need to map multiple naming conventions? Options: Single canonical schema, Multiple systems — mapping required, No clear schema — needs design
• Describe any known data quality issues (gaps, clock drift, sensor noise, duplicated tags) and how often they occur.

Where might the project unexpectedly hit a wall?

• Which of these security or network constraints are likely to slow or block model training and testing? Options: No external access from OT, Strict firewall / no VPN, Encrypted historian with limited APIs, No data export policy, Change control windows, None of the above / open
• How do you currently separate OT and IT environments, and who owns cross-domain approvals? Options: Strict OT-only, IT approval needed, OT/IT collaborative team, IT-managed with OT oversight, No formal separation
• Are service accounts, certificates, or credentials required for read/export access — and are they already provisioned? Options: Yes — provisioned, Yes — need provisioning, No credentials required, Unsure
• Have you previously blocked third-party analytics on security grounds? Tell us about that incident and resolution.
• What change-control or maintenance windows will restrict when we can extract data or run live acceptance tests? Options: Daily windows, Weekly planned outages, Monthly maintenance window, Only during emergency/major outages, No constraints known

The truth about labeled incidents (and near-misses) — how real is it?

• Roughly how many confirmed failure events for the target asset types do you have in your records over the last 3 years? Options: 0, 1–5, 6–20, 21–50, 50+
• Do failure records include clear root-cause labels, timestamps, and attachments (repairs, photos, operator notes)? Options: Yes — full RCA and attachments, Partially — timestamps but limited RCA, Mostly qualitative notes only, No, labels are missing
• How consistent is sensor coverage during historical incidents (were the key signals recording throughout the failure progression)? Options: Consistently recorded, Partial coverage, Rarely recorded, Unknown
• Are near-misses or preventive actions logged in the same system as failures, and can we access them for model training? Options: Yes — consolidated, No — separate system, Not logged reliably
• Who on your team currently owns labeling and root-cause confirmation, and how long does a typical label review take?

Edge, cloud, or hybrid — where does your future live?

• Which deployment target do you prefer for pilot inference and why (edge device, local on-prem, cloud)? Options: Edge (on-device), On-prem server, Cloud (public), Hybrid
• What bandwidth or latency constraints exist between plant OT networks and your corporate/cloud environment? Options: High bandwidth/low latency, Moderate, Low bandwidth/high latency, No connectivity
• Is there an approved edge hardware or gateway standard at the site (vendor/model), or will we propose one? Options: Existing approved hardware, No standard — vendor selection needed, We'll use cloud-only for pilot
• Do you have on-prem compute resources (VMs / GPU) available for model training or will training run in the cloud? Options: On-prem available, Cloud-only, Hybrid / depends on model
• Are there corporate policies restricting data leaves (e.g., PII/PII-like, export controls) that would affect where we train or store models? Options: Yes — restrictions apply, No major restrictions, Unsure — need to check

What will actually make operators act — metrics that matter beyond accuracy

• What minimum prediction performance would your team accept to act on an alert (select all that apply)? Options: Precision / low false positive rate priority, High recall / catch most failures, Balanced F1, Specific RUL lead-time target, Diagnostic confidence > threshold
• Which of these operational KPIs will determine pilot success for you? Options: Reduction in unplanned downtime, Work-order accuracy & fix rate, False positive rate, Lead time before failure, Maintenance hours saved, Operator trust adoption
• How much lead time (e.g., days/weeks) do you need from a prediction to reasonably plan a corrective action? Options: Hours, 1–3 days, 4–14 days, 2+ weeks
• Would you require diagnostic context with each alert (probable failure mode, affected component, recommended actions)? Options: Yes — required, Nice to have, Not necessary
• How should alerts be delivered and actioned (select all that apply)? Options: CMMS work-order creation, Operator HMI/SCADA alerts, Email / Pager, Mobile app, Dedicated reliability dashboard, Other

What could make this stall — and who will stop it?

• If you had to name one single blocker that would most likely kill the pilot, what would it be? Options: Lack of labeled incidents, Security / network policy, Insufficient team time, Budget constraints, Vendor coordination issues, Other
• Which stakeholders must sign off before we can begin data extraction and training?
• What is your target timeline for having a model-ready dataset and the first training run? Options: <2 weeks, 2–4 weeks, 1–2 months, 2+ months
• Are there planned plant changes (equipment upgrades, sensor rollouts, process changes) that could impact the pilot during the next 3 months? Options: Yes — scheduled, Maybe — under discussion, No
• Who will be the internal champion ensuring cross-team follow-through, and how will they be empowered?

Let’s put a date on readiness — quick commitments

• If we asked for read-only access to a representative dataset within 30 days, how confident are you that can be provided? Options: Very confident, Somewhat confident, Unlikely, Impossible
• Which of these artifacts can you provide within two weeks to accelerate readiness (select all that apply)? Options: Data schema / tag list, Sample historical exports, Failure reports / RCAs, Network diagrams & access contacts, Security/IT contact & ACLs, None available
• Who will create the ticket or approval we need to begin extracting data (name and role)?
• What is the preferred method for sharing credentials or secure access information with our onboarding team? Options: Secure vault (e.g., CyberArk), Encrypted email, On-site handoff, VPN token exchange, Other
• Is a non-disclosure agreement or site visit approval required before we can proceed, and is it in place? Options: NDA in place, NDA required — needs signing, Site visit approval pending, Not required

Quick intro — who are we teaming up with?

• Please tell us your role and the team you represent (e.g., Reliability Engineer — Rotating Equipment, Maintenance Director — Refining, OT Lead).
• Which types of assets are you most focused on for predictive capability right now? Options: Centrifugal pumps, Compressors, Motors/drive trains, Heat exchangers, Turbines, Valves/actuators, Other
• Roughly how many instrumented critical assets (the ones where an unplanned failure costs >$100k) are in scope at your site or region? Options: 1–10, 11–50, 51–200, 200+
• What would success look like for you in a single sentence (free response)?
• Who will be our day-to-day contact for technical data questions and who is the final decision-maker for pilots?

What keeps you up at 2 AM about equipment reliability?

• If your current approach to asset failures had to defend itself in front of your leadership, what would be the hardest criticism to answer? Options: Too many surprises, High reactive spend, Low confidence in diagnostics, Integration gaps with CMMS, Inadequate data/history, Other
• How often do unplanned critical failures occur that meaningfully disrupt operations? Options: Weekly, Monthly, Quarterly, Yearly, Rarely/never
• Tell us about a recent failure you wish you'd seen coming — what happened, and what consequences did it cause (cost, safety, downtime)?
• How does your team currently decide which failures or assets 'deserve' predictive focus — by cost, frequency, safety, or other criteria? Options: Cost of failure, Frequency of failure, Safety/criticality, Regulatory impact, Ease of instrumentation, Other
• When a prediction or alert has been wrong in the past, how did that affect trust and the subsequent workflow?

Where the data lives — and the real barriers to using it

• How would you describe the top three systems where relevant signals and records live (e.g., PI/OSI, Azure Data Lake, local historian, CMMS, PLC/edge)?
• Which of the following best describes your historian/SCADA coverage for the assets in scope? Options: Full coverage (all critical signals), Partial coverage (most assets have key sensors), Sparse coverage (few sensors), No historian — only PLC/local logs
• What percent of incidents have labeled failure events / maintenance outcomes in your records (useful for supervised training)? Options: >80%, 50–80%, 20–50%, <20%, Unknown
• Are there consistent data quality issues we should know about (gaps, clock drift, unit mislabeling, frequency mismatch)? Please list the top two.
• Who currently controls access to the raw historian and CMMS data, and how long does it take to get read access for a vendor project? Options: IT/OT team (days), IT/OT team (weeks), Local engineering (days), Local engineering (weeks), Policy blocks external access

When alarms become background noise — how alerts actually behave

• Why do most current alerts or thresholds fail to create confident maintenance actions in your organization? Options: Too many false positives, Lack of diagnostic context, Poor timing (too late/too early), Alert routing issues, No clear owner
• What false positive rate would destroy trust on your team (i.e., how many incorrect alerts per month is unacceptable)? Options: >50%, 20–50%, 10–20%, 5–10%, <5%
• When an alert is raised today, what is the typical workflow from receipt to action (who evaluates, what steps, how long until a work order is created)?
• Do your current alerts include diagnostic guidance (likely root cause, affected components, confidence level)? Options: Yes — actionable diagnostics, Partial — some clues, No — only alarms
• How tightly does your CMMS link to OT workflows today: automatic work-order creation, manual entry, or no integration? Options: Automatic integration, Semi-automated (manual confirmation), Manual only, No CMMS integration yet

If predictions were trusted — what would change for your operation?

• Imagine predictions arrive with enough lead time and accuracy — which outcome matters most to you? Options: Reduced emergency repairs, Lower spare parts inventory, Longer mean time between failures, Improved production uptime, Safer operations, Other
• What quantitative targets would you set for a pilot to consider it successful (choose all that apply and provide numbers in the next question)? Options: % reduction in unplanned downtime, % reduction in false positives, Median lead time to failure (days), RUL accuracy target, Number of correct work orders generated
• Please provide the numeric targets you’d like to see for the items you selected (e.g., 30% downtime reduction; FP <10% monthly; lead time ≥14 days).
• How will you validate that a diagnostic is 'useful' operationally — who signs off and what evidence do they need?
• Beyond technical metrics, what behavioral change would signal success (e.g., planners act earlier, techs schedule inspections differently)?

Who's on the team and who will stand behind the results?

• Which stakeholders must be convinced before a pilot can move to production (pick all that apply)? Options: Reliability/R&M, Maintenance planners, Operations/shift leads, IT/OT, Safety, Procurement/Finance, Plant leadership
• What are the top three questions each stakeholder group will ask before approving production deployment?
• How much calendar time can your core team commit to a pilot (data access, weekly reviews, acceptance testing)? Options: <4 hours/week, 4–8 hours/week, 8–16 hours/week, >16 hours/week
• Who will own model performance in production—Reliability, OT, or a joint team—and who will own CMMS change management? Options: Reliability, OT, Joint team, Other
• Are there procurement or legal constraints (e.g., data residency, vendor onboarding timelines) that typically add time to projects? Options: Yes — significant, Somewhat — moderate, No — minimal

Pilot to production — what would make the handoff painless?

• What scope do you see as realistic for an initial pilot (number of assets, failure modes, lines/units)? Options: 1–3 assets, 4–10 assets, 10–25 assets, Facility-wide
• Which integrations are required during pilot to prove end-to-end value (select all that apply)? Options: CMMS work-order creation, Operator dashboards/alerts, SCADA/PI tags ingestion, Edge deployment for local scoring, Cloud model training, ERP for parts/workflows
• What acceptance tests would you require before concluding the pilot (examples: live-data accuracy, diagnostic triage trials, successful auto work-orders)?
• What SLA expectations do you have for an initial production rollout (accuracy targets, FP caps, latency for alerts)? Options: High stringency (strict targets), Moderate (balanced), Flexible (iterative tuning)
• What budget range and approval path exists for converting a successful pilot to production? Options: Already budgeted, Requires CAPEX approval, Requires OPEX reallocation, No budget yet — need business case

Your worries, the smallest convincing win, and next steps

• What is your single biggest objection or fear about adopting predictive models in your environment? Options: False positives erode trust, Models won't generalize, Data/privacy concerns, Integration will break workflows, Lack of internal skills
• What would be the smallest practical outcome that would convince your team this approach is worth scaling (a micro-proof)?
• How do you prefer to experience a pilot: hands-off demo, collaborative co-development, or embedded support with onsite/remote training? Options: Hands-off demo, Collaborative co-development, Embedded support & training
• Are there any regulatory, safety, or union rules that would affect how we run tests or schedule inspections prompted by our predictions? Options: Yes — significant constraints, Some constraints, No relevant constraints, Unsure — need to check
• If we propose a clear first step within the next two weeks (data access checklist or a short discovery call), are you ready to commit to it? Options: Yes — ready now, Yes — but need scheduling, Not yet — internal blockers, Unsure