Phase 1 of 6
Scoping & Truck-Roll ROI Constraints
Define the portfolio, equipment envelope, dispatch economics, and safety-regulatory perimeter that will govern every predictive-maintenance modeling decision.
0/10
Phase Progress
Required Recommended Optional Open-Source Proprietary Trinidy
Portfolio & Asset Scope
Identify tower portfolio in scope for predictive maintenance
Why This Matters
Site class drives both the structural inspection regime under TIA-222-H and ANSI/ASSP A10.48 and the failure signature of the equipment fleet on that structure. A guyed lattice tower in a severe wind zone has an entirely different failure physics profile than a rooftop small cell, and a generator-primary remote site has dominant failure modes (fuel contamination, battery aging) that a grid-tied urban monopole never exhibits. One-sizing a model across site classes is the most common accuracy killer in tower predictive maintenance.
Note prompts — click to add
+ Have we segmented our portfolio by site class, grid-dependency, and climate zone before modeling?+ Which site classes represent 80% of our annual truck-roll spend and should be prioritized first?+ Do we own the structural inspection cycle under TIA-222-H, or does a third party (landlord / tower co.)?
Required
Confirm which site classes your scoring model must cover — structural envelope and RF load differ materially.
Select all that apply
Macro lattice / guyed towers (TIA-222-H Class II/III)
Macro monopoles (urban and highway)
Rooftop sites (building-mounted)
Small cells / DAS nodes
Owned towers (American Tower / Crown Castle / SBA model)
Co-located carrier POPs on third-party structures
Remote / off-grid sites (generator-primary)
Distributed generation / hybrid solar-diesel sites
required
✓ saved
Inventory equipment classes for component-level scoring
Why This Matters
Failure probability scores are only actionable when attached to a repairable component — "site at risk" does not generate a work order, "generator battery bank at risk" does. The component inventory also drives which telemetry streams matter: an RRU-focused model will leak signal if generator run-hours are the dominant predictor in your climate. Component scope is a first-order modeling decision, not a downstream detail.
Note prompts — click to add
+ Does our model output attach to a specific repairable component and CMMS asset ID, or only to a site?+ Which components drive the most emergency (unplanned) truck rolls in our current ops data?+ Do we have complete asset hierarchy in CMMS (Maximo / ServiceNow FSM / Sitetracker), or are components tracked informally?
Required
Predictive maintenance is component-scored, not site-scored — confirm which equipment classes are in scope.
Select all that apply
DC power plant (rectifiers, controllers)
Battery strings (VRLA / lithium)
Diesel / natural gas generators
ATS / transfer switches
HVAC and free-air cooling
Radios (RRU / AAU) and BBU / baseband
Microwave / fiber backhaul transport
Tower lighting (FAA 47 CFR Part 17)
Physical security, door, and alarm sensors
Structural components (guy tension, anchor, coax runs)
required
✓ saved
Establish truck-roll cost baseline per dispatch type
Why This Matters
Industry reference for a US macro site puts routine truck-roll cost in the $500–$1,500 range including labor, fuel, and parts. Emergency dispatch with climber crews, crane, or parts-by-air can exceed this by an order of magnitude, and the 30–50% truck-roll reduction claim is only defensible when the cost basis is stratified — cutting ten $500 routine rolls is not the same ROI as cutting two $10,000 emergency rolls.
Note prompts — click to add
+ Do we track truck-roll cost by urgency class in our finance system, or is it a blended average?+ What percentage of our dispatches are emergency vs. routine, and is the ratio trending?+ Who owns the truck-roll P&L line and will they sign the ROI thesis this model targets?
Required
Quantify the cost of a truck roll by urgency — the economic lever the model must beat.
Single choice
US macro site baseline: $500–$1,500 per routine dispatch
$1,500–$3,000 (emergency / off-hours / parts-by-air)
$3,000–$10,000 (remote / tower-climber / crane required)
> $10,000 (helicopter / severe-weather emergency)
Mixed — tiered by site class and urgency
Not yet measured at the dispatch-type level
required
✓ saved
Define acceptable unnecessary-dispatch (false positive) rate
Why This Matters
Mature tower predictive-maintenance systems routinely achieve a sub-5% unnecessary-dispatch rate — above roughly 10% the field-ops team loses trust in the model and reverts to their own rules of thumb, which destroys the ROI faster than any accuracy problem. The acceptable FP rate is really a workflow-trust number, not a pure statistical one, and it has to be negotiated with the regional ops managers before the threshold is tuned.
Note prompts — click to add
+ Have we agreed the FP ceiling with regional field-ops managers, not just with data science?+ What is the current FP rate on our incumbent rules-based alerting (thermal thresholds, generator run-hour alerts)?+ Do we measure FP rate by component class, or only in aggregate?
Required
Select your tolerance for dispatches where the model predicted failure but the component was healthy.
Single choice
< 5% false-positive dispatch rate (aggressive precision target)
5% – 10% (mature industry benchmark)
10% – 20% (early deployment / expanding coverage)
> 20% (exploratory — rules-based baseline)
Not yet measured
required
✓ saved
Define SLA-breach prevention target
Why This Matters
Carrier SLA-breach penalties run from a few thousand dollars per event on consumer-tier contracts to tens of thousands on enterprise and public-safety contracts, and the FCC NORS outage-reporting thresholds (47 CFR Part 4) create an additional regulatory exposure when an outage crosses the reporting floor. Framing the model around SLA-prevention catch rate forces the conversation onto which failures matter — a generator start-failure during a grid event is not the same class of miss as a fan bearing alarm.
Note prompts — click to add
+ Have we mapped our customer SLA tiers to the failure modes that most often cause breach?+ Which of our sites host FCC Part 4 reportable services and therefore carry regulatory outage exposure?+ Do we separately track SLA-prevention catch rate from overall failure prediction accuracy?
Required
Specify the share of SLA-impacting failures the model must catch before service impact.
Single choice
Prevent > 70% of SLA-breach-causing failures
Prevent > 85% (carrier-tier mature benchmark)
Prevent > 95% (critical-infrastructure / FirstNet-class)
Not yet formalized as a model KPI
required
✓ saved
Specify deployment topology for inference
Why This Matters
The predictive-maintenance function must remain operational during exactly the conditions that justify it — severe weather, grid events, backhaul degradation, generator cutovers. A cloud-routed model that loses its inference path during an ice storm is a model that fails precisely when it should have generated its highest-value alerts. Edge-or-hybrid is therefore an architectural commitment, not an optimization.
Note prompts — click to add
+ What is our inference behavior during a backhaul outage — does scoring continue, degrade, or stop?+ Have we measured how often our sites lose backhaul in a given year, and does that match our model availability target?+ Is the inference runtime co-located with the telemetry collector, or is there a network hop between them?
Required
Select the physical/logical target for the component-level scoring runtime.
Single choice
Edge inference at each site (on-node, no backhaul dependency)
Regional edge hub (aggregated across a cluster of sites)
Private cloud / carrier data center
Public cloud (AWS / Azure / GCP)
Hybrid: edge inference + cloud training
requirededgetrinidy
TrinidyCloud-hosted maintenance AI creates a dependency on external connectivity for a function that most matters during network stress — exactly when backhaul is degraded. Trinidy runs the full inference stack on-site so scoring continues through transport outages.
✓ saved
Inventory safety and regulatory standards that gate dispatch
Required
Confirm the safety and regulatory standards that will constrain when and how the model can recommend dispatch.
Select all that apply
OSHA 29 CFR 1910 (general industry)
OSHA 29 CFR 1926 (construction — climbs and rigging)
ANSI/ASSP A10.48 (telecommunication structures)
TIA-222-H (structural standards for communication structures)
FCC 47 CFR 1.1307 / 2.1093 (RF exposure limits)
FCC 47 CFR Part 17 (tower lighting and marking)
FCC 47 CFR Part 4 (NORS outage reporting)
NEPA environmental compliance
State and local utility crossing / right-of-way
required
✓ saved
Confirm data residency and sovereignty constraints
Required
Map telemetry and maintenance data to jurisdictional constraints before architecture is finalized.
Select all that apply
US critical-infrastructure (CISA sector guidance)
FirstNet / public-safety data residency
EU GDPR — data must remain in EU
UK GDPR — UK residency required
Canada PIPEDA / provincial equivalents
Brazil LGPD
Host-country telecom regulator residency (varies)
No formal residency constraint today
requiredtrinidy
TrinidySensor telemetry is often treated as "non-sensitive" and streamed to cloud vendors — until a national-security review reclassifies location and power-signature data as infrastructure-sensitive. Trinidy keeps raw telemetry on-site; only scored alerts and aggregated features leave the perimeter.
✓ saved
Ownership, CMMS & Workflow Scope
Identify CMMS / field-service platform of record
Why This Matters
Failure probability scores only convert to ROI when they flow into the work-order queue the field-ops team already uses — writing a parallel dashboard is the most common deployment failure in predictive maintenance programs. Maximo, ServiceNow FSM, Sitetracker, and Uptake/GE Digital APM each have different integration patterns and different asset-hierarchy models; the integration effort is real and should be sized before modeling starts.
Note prompts — click to add
+ Does our CMMS have a stable asset ID we can join model outputs to, or will we need to build that mapping?+ Who owns the CMMS integration effort — IT, ops, or the ML team?+ Can the CMMS accept score-driven work-order creation via API, or will a human dispatcher be in the loop?
Required
The CMMS determines the asset hierarchy, work-order format, and dispatch economics the model must target.
Single choice
IBM Maximo
ServiceNow FSM
Sitetracker
Uptake APM
GE Digital APM
American Tower / Crown Castle / SBA proprietary OSS
In-house / home-grown CMMS
None — work orders managed via spreadsheets / email
required
✓ saved
Clarify tower-owner vs. carrier responsibility split
Why This Matters
On a co-location site the tower owner typically owns structural, grounding, lighting (FCC Part 17), and shelter systems while the carrier owns radios, antennas, and backhaul — and power / HVAC / security are often shared with contract-defined responsibility. Your model cannot score what you have neither telemetry for nor dispatch authority over, and misaligned scope creates friction with the counterparty the first time a score triggers a dispatch on their equipment.
Note prompts — click to add
+ Have we documented the responsibility split per site class, not just per corporate contract?+ Which components would require counterparty coordination to dispatch on, and how fast is that handoff today?+ Do we need data-sharing agreements with co-location carriers to access the telemetry our model needs?
Required
Confirm which components the model covers under your operating agreement — this drives data access and dispatch authority.
Select all that apply
Tower-owner: structure, ground, lighting, shelter (American Tower / Crown Castle / SBA pattern)
Carrier: radios, backhaul, antennas
Shared: power, HVAC, physical security
Managed-services contract (tower owner operates for carrier)
Single-entity owned and operated
required
✓ saved