Managing a fleet usually feels manageable right up until the signals stop lining up. Dispatch sees late vehicles, finance sees rising fuel spend, maintenance sees more unscheduled work, and customer service sees missed ETAs. Everyone has data, but very little of it connects.
That's why the best conversations about fleet management software features shouldn't start and end with a map view. GPS matters, but it's only useful when it feeds decisions about routing, maintenance, safety, compliance, and cost control. The software that delivers the most value is the software that turns movement data into operational action.
A lot of teams start by shopping for a dashboard. They should be shopping for an operating system. That means choosing features that help supervisors act in real time, while also feeding a data platform that supports historical analysis, forecasting, and cross-functional reporting. If you're comparing options, this overview of NZ fleet management systems is a useful market scan, but the more important question is how each feature affects business outcomes after rollout.
The list below focuses on ten capabilities that matter in practice. Some are foundational. Some are differentiators. All of them work better when they're integrated into a broader architecture instead of deployed as isolated modules.
1. Real-Time GPS Tracking and Location Intelligence
A dispatcher sees three vehicles on the map, all marked active, but only one can still make the next service window. The difference is not the dot on the screen. It is whether the platform can turn location signals into ETA logic, dispatch decisions, and usable historical data.
GPS tracking is mature technology. For large fleets especially, it is standard operating infrastructure rather than a differentiator, as noted earlier from GoCodes fleet adoption data. The central question during selection is whether the software stops at visibility or feeds the systems that run the business.
That distinction matters quickly. A delivery fleet needs live ETAs that customer service can trust. A field service team needs nearest-vehicle assignment that accounts for traffic and job status, not just distance. A construction operator needs proof that a truck, trailer, or crew arrived on site and how long it stayed there. Raw coordinates do not solve those problems by themselves.
What actually works
The strongest implementations connect location data to operating rules, event logic, and downstream analytics:
- Use geofencing for repetitive events: Depots, customer sites, and yards are the right places to automate arrivals, departures, and dwell-time alerts. Faberwork's work on geofencing in fleet management shows how location events can replace manual status updates and give operations teams cleaner timestamps.
- Store historical telemetry in a warehouse: Sending trip, stop, and dwell data into Snowflake or a similar platform gives analysts a usable record for route adherence, recurring delays, idle patterns, and SLA performance across regions and customers.
- Set ping rates by business need: Last-mile fleets often need tighter intervals than long-haul operations. Higher frequency improves visibility, but it also raises device costs, data volume, battery draw, and noise in the event stream if no one defines what should trigger action.
- Connect GPS to maintenance context: Location and utilization data become more valuable when they help explain wear patterns, idle-heavy duty cycles, and service demand. That is one reason teams often pair tracking with programs such as Forge Reliability's predictive maintenance.
A practical test is simple. If supervisors still call drivers for updates that should already be visible in the system, the GPS module is not integrated into dispatch and service workflows well enough.
From an architecture standpoint, this feature should also produce data your enterprise stack can use later. Location streams belong in the same environment as work orders, fuel records, and maintenance history. Once that data lands in Snowflake, teams can measure repeat late arrivals by customer, compare route plans against actual execution, and separate one-off delays from structural problems. That is where GPS starts producing measurable business value instead of another dashboard.
2. Predictive Maintenance and Vehicle Health Monitoring
A preventable breakdown usually starts as a data problem long before it becomes a roadside problem. A fault code appears, engine hours cross a threshold, or battery performance slips over several weeks, but no one connects those signals to a work order soon enough. The result is missed service windows, disrupted routes, overtime in the shop, and harder customer conversations.
Predictive maintenance matters because it changes how fleets decide when to act. Calendar-based service still has a place, but it misses the reality of mixed duty cycles. Two vehicles with the same age can have very different wear profiles if one spends its week in stop-and-go urban traffic and the other runs steady highway miles.
The better systems combine telematics, diagnostic trouble codes, odometer readings, engine hours, historical repairs, warranty data, and parts consumption. That combination supports decisions that are operational, not theoretical. Should a vehicle come in now, wait until the next planned stop, or stay in service because the alert is low confidence?
Where predictive maintenance pays off
The first gains usually come from high-cost failure categories and assets with clear usage patterns. Engine, transmission, braking, cooling, and battery systems are common starting points because one unplanned failure in those areas can sideline a vehicle for days.
A practical rollout usually includes these steps:
- Prioritize failure modes by business impact: Start with issues that create the most downtime, towing expense, missed deliveries, or safety exposure.
- Use actual duty-cycle data: Mileage alone is a weak proxy for wear in fleets with heavy idling, PTO usage, harsh braking, or short-trip operation.
- Tie alerts to shop capacity: A prediction has little value if the maintenance team cannot schedule labor, bay time, and parts around it.
- Measure alert quality: Track false positives and missed failures. If alerts create too much noise, technicians and fleet managers will ignore them.
I usually tell clients to treat maintenance models as decision support, not automation that runs on its own. A good platform helps maintenance managers make better calls faster. It does not replace technician judgment, service bulletins, or inspection discipline.
One useful outside capability in this area is Forge Reliability's predictive maintenance, especially for teams that need more disciplined reliability workflows around failure analysis and condition monitoring.
The implementation mistake I see most often is isolation. Maintenance data sits in one application, telematics in another, fuel transactions somewhere else, and warranty claims in email threads or spreadsheets. That setup limits analysis to basic reminders and fault-code monitoring.
Better results come from putting vehicle health data into the same enterprise environment as route history, fuel records, driver events, and work orders. In Snowflake or a similar platform, analysts can test whether repeated brake work is concentrated on specific route types, whether idle-heavy vehicles produce more battery claims, or whether a maintenance vendor is resolving recurring failures on the first visit. That is how predictive maintenance shifts from a shop tool to a business system with measurable impact on uptime, parts planning, and total operating cost.
3. Route Optimization and Dispatch Management
Route optimization looks simple in demos. Drop stops on a map, press a button, and get a cleaner route. Real operations are messier. Drivers have shift limits, customers have delivery windows, vehicles have capacity limits, and the day starts changing as soon as the first stop slips.
The software should help dispatch make trade-offs quickly, not chase mathematical perfection. A practical route engine balances traffic, stop sequence, service time, geography, and operating constraints. It should also recalculate without making the plan so unstable that drivers stop following it.
Common use cases
A few scenarios show where routing features create immediate value:
- Last-mile delivery: Dispatch teams need to absorb late orders and still preserve reasonable ETAs.
- Field service: Technician assignment depends on skill match as much as distance.
- Cold chain operations: A route may look efficient on paper but fail if dwell time threatens temperature control.
The trade-off many managers overlook is driver acceptance. If the platform constantly pushes routes that ignore local knowledge, experienced drivers create their own workarounds. That's not stubbornness. It's a signal that the model hasn't accounted for reality.
For that reason, historical data matters as much as live traffic. Store route plans, actual arrivals, exceptions, and customer outcomes in Snowflake, then review where the algorithm repeatedly underperforms. That feedback loop is what turns route planning from a static feature into a learning system.
A route optimizer should also serve dispatch, finance, and customer support at the same time. If only dispatch can interpret what the system is doing, the feature won't scale beyond operations.
4. Driver Behavior Monitoring and Safety Analytics
Safety tools can reduce risk, but they can also create internal friction if they're rolled out badly. Drivers don't want a black box that only exists to punish them. Fleet leaders need software that supports coaching, documentation, and incident review without turning every shift into a surveillance argument.
That's where behavior monitoring has matured. The stronger platforms combine telematics events, mobile signals, and video context to flag harsh braking, speeding, distraction, seat belt issues, and fatigue-related patterns. The point isn't to collect events. The point is to determine which behaviors predict real risk and which ones just create alert fatigue.
Coaching beats raw scoring
A safety score alone doesn't change much. What changes behavior is a workflow:
- Immediate in-cab feedback: Drivers can correct certain behaviors in the moment.
- Manager review with context: Video and trip history matter more than a red flag count.
- Targeted retraining: Different drivers need different interventions.
The market is moving toward more advanced analytics here. The global fleet management market is projected to expand at a 15.3% CAGR from 2026 to 2030, with some forecasts indicating growth to $128.83 billion by 2033 at a 14.20% CAGR, according to Data Bridge Market Research on the global fleet management market. A major reason is that behavior scoring, accident prediction, and fuel anomaly detection are becoming part of the core platform, not add-ons.
Before adopting video-heavy monitoring, it helps to show teams what good implementation looks like in practice:
Better safety programs reward improvement as visibly as they document exceptions.
The best fleets use Snowflake or a similar platform to segment drivers into coaching cohorts, compare behavior trends by route or region, and separate one-off mistakes from repeat patterns that need intervention.
5. Fuel Management and Cost Optimization
A fleet director reviews the monthly fuel report, sees spend up 9%, and still cannot tell what changed. Was it route mix, more idle time, a few underperforming vehicles, fuel theft, or simple price movement? If the software cannot separate those causes, it produces noise instead of action.
Fuel management works when the platform turns raw consumption data into operational decisions. That starts with normalization. Compare vehicles within the same class, route type, load pattern, and duty cycle. A box truck running dense urban stops should not be scored against a highway tractor or a lightly loaded service van.
That discipline matters because fuel savings often look obvious until teams try to assign accountability. A spike in usage might come from driver habits. It might also come from a failing injector, underinflated tires, extra PTO use, or bad fuel card controls. Good software helps operations, maintenance, and finance see the same event through different lenses.
Where savings usually show up first
The fastest gains usually come from a short list of controllable issues:
- Excess idling: Common in delivery, field service, utilities, and cold-weather operations.
- Off-route miles and poor stop sequencing: Small deviations add up quickly across a large fleet.
- Mechanical inefficiency: Fuel outliers often expose maintenance issues before a breakdown happens.
- Fuel card exceptions: Purchases that do not match vehicle location, tank capacity, or assigned driver need review.
- Spec mismatch: The wrong vehicle for the job can lock in avoidable fuel cost for years.
The implementation mistake I see most often is trying to attack every fuel issue at once. Fleets get better results by picking two or three levers, setting a baseline, and measuring weekly. If idle time is the biggest problem, start there. If card misuse is the actual source of loss, a driver coaching campaign will not fix it.
This is also where enterprise data architecture starts to matter. Telematics data on its own shows movement and engine behavior. Fuel card data shows transactions. Maintenance systems show service history. Snowflake or a similar platform lets teams join those records at the vehicle, driver, and trip level so they can separate price effects from consumption problems and identify which interventions reduced cost.
That changes the quality of decision-making. Instead of saying fuel spend increased, teams can say spend increased because urban idle hours rose in one region, two vehicle models are trending below expected MPG, and card exceptions are concentrated at a small number of locations.
Finance benefits too. A good fuel management process creates an audit trail for cost changes, supports accrual accuracy, and makes budget variance easier to explain. It also helps with capital planning, because persistent fuel underperformance can point to replacement candidates or a need to revisit vehicle spec.
6. Compliance and Regulatory Reporting Automation
Compliance features rarely win software demos, but they matter when an audit lands or a preventable violation disrupts operations. For regulated fleets, automation reduces the number of tasks that depend on memory, manual paperwork, or last-minute scrambling.
The best systems handle hours-of-service records, inspections, maintenance logs, certification tracking, and renewal reminders in the same environment. That keeps dispatch, safety, and back-office teams from reconciling separate systems when deadlines are already tight.
Build rules into operations, not just reports
A lot of fleets make compliance software reactive. They use it to document what already happened. Better fleets use it to prevent bad decisions upstream.
- Dispatch-aware HOS controls: Don't assign work that creates a violation later in the day.
- Expiration alerts with buffer time: Licenses, registrations, and inspections need enough lead time for action.
- Audit-ready storage: Pulling records quickly matters almost as much as having them.
This becomes easier when compliance data lives in a structured warehouse instead of scattered PDFs, emails, and app exports. Snowflake can help centralize inspection history, training records, and asset status so teams can retrieve the right document without slowing operations.
Compliance automation also helps with consistency across mixed workforces. Full-time drivers, part-time crews, contractors, and owner-operators often follow similar processes with slightly different rules. Software should support those variations without forcing your team into awkward workarounds.
The biggest implementation mistake is treating compliance as a safety-team-only feature. Dispatch, maintenance, and HR all create or prevent exceptions. If they can't see the same operational status, the automation layer stays shallow.
7. Mobile Driver Applications and Digital Workflows
If drivers hate the mobile app, the rest of the platform suffers. Routes get ignored, proof-of-delivery gets delayed, exceptions get logged late, and dispatch falls back to calls and texts. This is one of the fleet management software features that looks secondary in procurement and becomes central after launch.
The app has to fit field reality. Drivers need fast navigation, delivery details, signatures, photos, status updates, and issue reporting in a form they can use with one hand, in poor light, under time pressure. A stripped-down desktop interface usually fails here.
What good mobile workflow design looks like
Strong driver apps usually share a few traits:
- Offline support: Core actions should still work in low-connectivity areas.
- Fast exception handling: Late customer, inaccessible site, damaged package, wrong inventory. These shouldn't require a call every time.
- Minimal taps for common tasks: Proof-of-delivery should be fast, not buried in menus.
- Tight backend integration: Customer records, inventory, and maintenance alerts should flow into the same experience where relevant.
Customer service gains from this as much as drivers do. A real-time delivery status with photos or signatures reduces disputes and gives support teams something more useful than “the driver says it was delivered.”
One subtle benefit is data quality. Paper forms and delayed updates introduce ambiguity. A mobile workflow captures time, location, and supporting evidence at the moment work happens. That makes the data more defensible and much more usable in downstream analytics.
The trade-off is training. Even a good app needs a clear rollout plan, especially in mixed fleets where device familiarity varies widely. Adoption rises when supervisors coach the workflow in the field instead of relying on a slide deck.
8. Asset and Equipment Tracking Beyond Vehicles
A fleet platform gets more valuable when it tracks what the vehicle is carrying or towing, not just the vehicle itself. Trailers, containers, generators, compressors, tools, and specialized equipment often disappear from view once they leave a yard. That creates avoidable waste.
For some operations, non-powered assets are the bigger problem. A logistics company may know exactly where the tractor is but still lose time locating the right trailer. A construction team may know which truck reached the site but not whether the high-value equipment assigned to the project is there.
Where expanded tracking helps most
The clearest use cases tend to be:
- High-value equipment: Useful for theft recovery and utilization tracking.
- Shared assets across sites: Helps avoid duplicate rentals or unnecessary purchases.
- Maintenance-dependent equipment: Location-aware service scheduling reduces missed intervals.
- Specialized containers or trailers: Better for compliance and availability management.
This feature is especially effective when paired with geofencing. Entering or leaving a depot, crossing into a restricted zone, or arriving at a maintenance location can trigger status changes automatically. That removes manual scanning steps that often break down under workload.
Asset tracking also changes capital planning. Once teams can see actual utilization patterns, they often discover equipment that's underused, misplaced between sites, or over-relied on by one division while another holds idle inventory. That's a finance conversation as much as an operations one.
The most important implementation choice is scope. Start with assets that are expensive, operationally critical, or historically hard to locate. Trying to tag everything at once usually creates overhead before it creates clarity.
9. Integration, Data Platform Architecture, and Advanced Analytics
A fleet team often reaches the same frustrating point after rollout. GPS is live, maintenance alerts are firing, drivers are using the app, and dispatch can see the map. Yet a finance lead still cannot tie route changes to margin, safety still works from a separate scorecard, and operations analysts spend hours stitching exports together before every review. That is usually not a feature problem. It is an architecture problem.
The difference between a usable fleet system and a strategic one is how well it connects operational data to the rest of the business. Telematics on its own improves visibility. Telematics combined with work orders, fuel transactions, payroll, ERP data, and customer commitments improves decisions. Research summarized by Intangles on telematics and data intelligence reflects the same pattern. Many fleets collect large volumes of data but still lack a unified model for real-time analysis.
Why architecture matters more than another dashboard
Enterprises usually get better results when fleet data flows into a shared platform such as Snowflake instead of staying inside one vendor application. That setup lets teams join trip events with invoices, maintenance spend, labor hours, service levels, and customer outcomes. It also makes analytics reusable across departments instead of trapped in fleet-specific reports.
The design choices that matter most are usually straightforward:
- Use event-driven integration where it fits the process: Exception handling, geofence events, and urgent maintenance alerts lose value if they wait for batch syncs.
- Define shared entities early: Vehicle, driver, trailer, asset, trip, order, and work order need stable IDs across systems before reporting logic scales.
- Separate operational views from analytical models: Dispatch needs current-state alerts. Finance and strategy teams need curated historical data for trend and variance analysis.
- Check data quality at ingestion: Time zone errors, duplicate vehicles, and mismatched driver IDs will undermine every KPI that follows.
- Design for data ownership: Fleet operations, IT, finance, and safety should each know which system is the source of truth for the fields they control.
There is a trade-off here. A direct point-to-point integration can get a pilot running quickly, but it often becomes expensive to maintain once the fleet adds a TMS, ERP, data warehouse, or new telematics provider. A shared data platform takes more planning up front, but it reduces rework later and gives analytics teams a stable base for forecasting, cost attribution, and machine learning.
For teams building custom reporting and operational models, this approach is closely aligned with Faberwork's article on enhancing logistics with Python data analytics. The point is not to add data science for its own sake. The point is to connect fleet events to measurable business outcomes.
Unified fleet data turns separate software features into one operating model.
Done well, this section of the stack changes more than fleet reporting. Finance gets cleaner cost allocation by route, customer, or service line. Operations leaders can compare planned versus actual execution without manual reconciliation. Technology teams get a platform they can extend, whether the next use case is exception prediction, warranty recovery analysis, or more accurate service ETA models.
10. Autonomous and Electric Vehicle Fleet Integration
A fleet team rolls out 20 electric vans, then realizes the routing engine ignores charger availability, the finance team cannot separate charging cost by route, and operations has no reliable view of battery health by shift. The vehicles are in service, but the operating model is not. That is the fundamental integration challenge.
Electric and autonomous vehicles belong in the fleet stack only when the software can treat them as operational assets, not special projects. For EVs, that usually means predictable mileage, repeatable routes, planned dwell time, and access to charging where vehicles already stop. For autonomous deployments, the practical use cases are narrower. Yard operations, fixed campus routes, and geofenced pilots are far easier to control than open-road service.
The implementation question is less about whether the vehicles are advanced and more about whether your data model is ready for them. Battery state of charge, charger sessions, energy consumption, remote diagnostics, software updates, and safety events need to land in the same platform as dispatch, maintenance, work orders, and cost data. In a Snowflake-centered architecture, that matters because operations, finance, and analytics teams can work from one record of vehicle performance instead of stitching together reports from charging portals, OEM dashboards, and telematics tools.
What deserves scrutiny before rollout:
- Charging-aware dispatch: Routes need charging constraints, charger availability, and dwell time built into planning logic.
- Battery and energy telemetry: State of charge, degradation trends, and charging behavior should sit beside mileage, utilization, and maintenance history.
- Cost model by vehicle and route: Finance needs a clear view of vehicle cost, energy cost, maintenance, incentives, and replacement timing.
- Pilot design: Short, repeatable duty cycles with stable schedules produce clean operational data and faster decisions.
- Autonomy controls: Geofencing, remote intervention, event logging, and incident review workflows need to be defined before a pilot starts.
One common failure pattern is treating EV adoption as a sustainability reporting feature. That misses the operational reality. If a planner cannot see whether a vehicle can complete the afternoon route after an unplanned reassignment, or if finance cannot compare energy spend against the ICE routes being replaced, the program will struggle to scale.
The better approach is to extend the existing fleet data model. Vehicle type changes. The management discipline does not. Teams that handle this well connect OEM data, charging data, telematics, and enterprise reporting early, then use that foundation to answer practical questions: which routes fit EV duty cycles, where charging creates bottlenecks, how battery health affects replacement timing, and whether autonomous pilots are reducing labor, delay, or safety exposure in controlled environments.
That is the difference between adding new vehicle technology and building a fleet operation that can use it.
Top 10 Fleet Management Features Comparison
CapabilityImplementation Complexity 🔄Resource Requirements ⚡Expected Outcomes 📊Ideal Use CasesKey Advantages ⭐Real-Time GPS Tracking and Location IntelligenceModerate, device deployment & connectivityGPS devices, low-latency streaming, mapping, SnowflakeLive visibility, faster responses, proof-of-deliveryLast-mile delivery, field service, construction, ride-shareImmediate operational visibility; data foundation for analyticsPredictive Maintenance and Vehicle Health MonitoringHigh, ML models + sensor integrationOBD‑II devices, historical maintenance data, data science30–50% fewer unplanned repairs; longer asset lifeFleets with high maintenance costs, heavy equipmentProactive servicing; lower downtime and repair costsRoute Optimization and Dispatch ManagementHigh, complex algorithms + real‑time updatesCompute resources, traffic APIs, order system integration15–30% fuel/mileage reduction; higher utilizationE‑commerce fulfillment, meal delivery, field service dispatchReduced costs and faster deliveries via optimal routingDriver Behavior Monitoring and Safety AnalyticsHigh, telematics + in‑cabin video + privacy controlsTelematics sensors, cameras, storage, coaching workflows15–35% fewer accidents; lower insurance and liabilityLong‑haul trucking, delivery fleets, school transportImproved safety, objective coaching, risk reductionFuel Management and Cost OptimizationModerate, analytics + integrationsFuel card integration, vehicle sensors, analytics10–20% fuel cost savings; detect fraud/inefficiencyTransport companies, utilities, construction fleetsQuick ROI from idle reduction and driver coachingCompliance and Regulatory Reporting AutomationModerate–High, jurisdictional rule engineHOS integrations, payroll, inspection capture, audit trailsFewer violations; dramatic audit time reductionLong‑haul carriers, multi‑jurisdiction logisticsAutomated compliance, reduced fines and audit effortMobile Driver Applications and Digital WorkflowsModerate, app dev + device managementMobile apps, devices, connectivity, backend APIsEliminate paper; real‑time POD and exception handlingLast‑mile delivery, field service, pharma deliveryStreamlined workflows; better customer visibilityAsset and Equipment Tracking Beyond VehiclesModerate, tag deployment & integrationsGPS tags, inventory integration, maintenance links5–15% reduction in loss; 10–25% better utilizationConstruction, rentals, container logisticsAsset recovery, utilization insight, shrinkage reductionIntegration, Data Platform Architecture, and Advanced AnalyticsVery High, enterprise data engineeringSnowflake/warehouse, ETL/ELT, APIs, connectors, BI toolsUnified data, real‑time KPIs, enable ML and forecastingEnterprises needing cross‑system visibility and analyticsEliminates silos; enables advanced analytics and AI projectsAutonomous and Electric Vehicle Fleet IntegrationVery High, emerging tech + infraEVs/AVs, charging infrastructure, specialized maintenanceUp to 70–80% fuel-cost reduction (long term); sustainability gainsUrban last‑mile pilots, transit agencies, controlled routesFuture‑proofing fleets; major long‑term operating savings
Building Your Future-Ready Fleet Stack
The best fleet management software features don't create value in isolation. GPS without dispatch rules becomes a map. Maintenance without diagnostics becomes a calendar. Safety monitoring without coaching becomes a scorecard nobody trusts. Integration is what turns these capabilities into a system that improves daily decisions.
That's the practical shift many enterprises need to make. Instead of asking which vendor has the longest feature list, ask which capabilities will reduce uncertainty in your operation. Ask which modules can feed a shared data platform. Ask which workflows your teams will use under pressure, on a busy route, during an outage, or in the middle of an audit.
A sensible rollout usually starts with the operational core. That means GPS tracking, routing, mobile workflows, and maintenance visibility. Once those are stable, the next gains come from connecting them. Fuel analysis becomes more useful when tied to route and driver data. Safety programs get sharper when behavior events connect to claims, coaching history, and maintenance records. Compliance gets easier when dispatch and document storage reflect the same status in real time.
Snowflake matters here because fleet data is naturally cross-functional. Vehicle telemetry, trip history, driver behavior, work orders, customer records, and cost data don't belong in separate analytical worlds if the business wants measurable improvement. A shared platform lets technology leaders support real-time monitoring and longer-horizon analysis at the same time. It also gives data teams room to build forecasting, anomaly detection, and automation on top of a cleaner foundation.
That architecture-first approach also makes future changes easier. New asset classes, more sensors, EV charging data, partner feeds, and AI models all become easier to absorb when the fleet stack already has strong integration patterns and clean entity models. Without that foundation, every new feature adds another silo.
For CTOs, CIOs, and operations leaders, the right decision involves more than buying software with the broadest brochure. It is about building a connected operating environment that helps teams act faster, document better, and improve continuously. That may involve internal development, vendor tooling, or a consulting partner. Faberwork LLC is one option for organizations that need Snowflake-centered data solutions, custom software, and integration work around logistics and fleet operations.
The outcome to aim for is straightforward. One fleet view. Shared operational truth. Better decisions at dispatch speed, with enough historical depth to keep improving after the first rollout.
