Hey, it is Vlad. This is one of those features where I genuinely get excited explaining it, because this is where AI stops being a buzzword and starts saving you real money. Here is the problem Predictive Maintenance solves: right now, your equipment either gets maintained on a fixed schedule (whether it needs it or not) or it runs until it breaks (usually at the worst possible time). An HVAC unit dies on the hottest day of July. A water heater bursts on a Friday evening. A commercial elevator breaks down when the building is full of people. These emergencies are expensive, stressful, and almost always preventable. Predictive Maintenance uses machine learning to analyze your equipment data – age, condition, maintenance history, usage patterns – and tells you what is likely to fail BEFORE it happens. I have seen property managers save thousands of dollars with this feature, because fixing something that is about to break costs a fraction of what it costs to fix something that already broke.
Estimated time: 18 minutes
Before You Begin
- An active Exoserva account with Owner or Admin role – Predictive Maintenance involves AI configuration and cost settings reserved for account owners and administrators. If you are a technician or dispatcher, ask your account owner to set this up or give you Admin access.
- At least 5-10 assets (equipment) added to your account with some maintenance history – the AI needs data to learn from. If you have not added any assets yet, start with the “Equipment and Asset Tracking” guide first. The more detailed maintenance records you have, the better the predictions.
- A plan that includes AI features (Professional or Enterprise) – Predictive Maintenance uses machine learning models that require computing resources. If you are on the Starter plan, you can upgrade from Settings > Billing.
- Some patience for the first few weeks – the AI model gets smarter over time as it learns your specific equipment and patterns. The first predictions might be broad, but after a month of data, they become remarkably accurate.
Step 1: Navigate to Predictive Maintenance Settings
Let us start by finding the Predictive Maintenance settings page. From the left sidebar of your Exoserva dashboard, click “Settings” – this opens the Settings page, which has multiple cards for different configuration areas. Look for the “Predictive Maintenance” card. It might be in the “AI & Automation” section of the Settings page. Click on it to open the Predictive Maintenance page.
When the page opens, you will see a breadcrumb trail at the top showing Settings > Predictive Maintenance. The page header shows the title with an Activity icon, a Status Badge (“Active” in green or “Inactive” in grey), a toggle switch to turn the feature on or off, and a gear button for advanced settings.
If this is your first time here, you will see a disabled state – a welcome screen with a prominent blue “Enable Predictive Maintenance” button. The welcome screen shows three statistics from existing customers: save 30-50% on maintenance costs, prevent 85% of equipment failures, and achieve 92% prediction accuracy. These are real numbers from real businesses using this feature.
Tip: This page has some handy keyboard shortcuts that will save you time as you use it regularly: press Cmd+R (on Mac) or Ctrl+R (on Windows/Linux) to quickly run a new AI analysis, and Cmd+I or Ctrl+I to immediately schedule an inspection. These shortcuts work from anywhere on the Predictive Maintenance page.
From Vlad: I put those statistics on the welcome screen because most people are skeptical about “AI predicting equipment failures” – I was too when we started building this. But after running it on real property portfolios for months, the numbers speak for themselves. The key is giving the system enough data – make sure your assets have solid maintenance histories entered.
Step 2: Enable and Configure the System
Time to turn on the magic. Click the big blue “Enable Predictive Maintenance” button on the welcome screen, or if you are already past the welcome screen, flip the toggle switch in the page header. The system activates immediately and the Status Badge changes from grey “Inactive” to green “Active”. You might notice a brief loading animation as the system initializes.
Since no assets have been analyzed yet, you will see an empty state screen with a blue “Run First Analysis” button (we will use this in the next step) and a 4-step “How It Works” section showing how data flows from your assets through the ML model into actionable predictions.
Before running your first analysis, I strongly recommend configuring the settings. Click the gear button in the page header to open the Settings Slide-Over panel – this is a panel that slides in from the right side of the screen. Here is what each setting does and what I recommend for beginners:
Prediction Horizon – This slider (7 to 365 days) controls how far ahead the AI predicts. Think of it like headlights on your car: longer = more reaction time. Start at 30 days, which gives you a solid buffer to schedule preventive work.
Confidence Threshold – Controls how confident the AI must be before showing a prediction. Like a spam filter: high = fewer but more certain predictions, low = more predictions but some may not pan out. Start at 70%.
Alert Threshold – Dropdown (Low, Medium, High, Critical) setting the minimum severity for notifications. Start with Medium to get alerted about important and critical issues without every minor warning.
Auto-Schedule toggle – Automatically creates work orders from predictions instead of just showing alerts. Leave this OFF for the first month while you learn to trust the predictions, then turn it on once you are comfortable.
Notifications toggles – Enable both email and in-app alerts. You can turn off email later if it feels like too much.
Monitored Asset Types – Multi-select for which equipment types the AI analyzes (HVAC, Plumbing, Electrical, Appliances, Roofing, Structural, Elevators, Fire Safety). Select all that apply. The panel also shows the current ML Model Version, which updates automatically over time.
Warning: Do not set the Prediction Horizon under 14 days. A 7-day horizon sounds good in theory, but in practice you need time to order parts, schedule a technician, and complete the work. Start with 30 days and adjust based on your team’s response time.
Step 3: Run Your First AI Analysis
With the settings configured, it is time for the moment of truth – your first AI analysis. This is where the machine learning model examines all your monitored assets and generates predictions about which ones might need attention. Click the “Run First Analysis” button on the empty state screen (or if you have already run analyses before, click the “Run Analysis” button with the play icon in the toolbar at the top).
When you click the button, a loading indicator appears. Behind the scenes, the system examines every monitored asset, reviews maintenance history, checks equipment age, evaluates condition data, and looks for patterns across the entire Exoserva platform (if similar HVAC units at other businesses failed at the 8-year mark, your 7.5-year-old unit gets flagged).
The analysis typically completes within 10-30 seconds depending on asset count. Once complete, the page transforms into the full dashboard with three tabs: Overview (metrics and predictions), Assets & Health (detailed health status per asset), and Maintenance Schedule (upcoming AI-generated maintenance). If any assets are in critical condition, a red Critical Alert Banner appears at the top showing the count needing immediate attention.
Step 4: Review the Overview Dashboard
The Overview tab is your predictive maintenance command center – the single page where you can see everything that matters at a glance. There is a lot on this page, so let me walk you through each section from top to bottom.
At the very top, you will find a Quick Actions bar with five buttons: Period Selector (buttons for 7D, 30D, 90D, and YTD – “Year To Date” – to choose which time window the data covers), View Critical (jumps straight to the most urgent predictions), Run Analysis (keyboard shortcut Cmd+R or Ctrl+R), Schedule Inspection (Cmd+I or Ctrl+I), and Export Report (Cmd+E or Ctrl+E to download the data).
Below the quick actions, the Hero Savings Card highlights four key metrics: Cost Savings This Month (money saved by catching problems early), Failures Prevented (avoided breakdowns), AI Accuracy (percentage of correct predictions), and Monthly Goal Progress. These are the numbers you share with property owners to justify the investment.
Below the hero card, five KPI Cards show operational detail: Total Assets monitored, Avg Health Score (0-100), Prevented Failures (cumulative), Cost Savings (cumulative dollars), and Prediction Accuracy (confirmed predictions).
Next, four Status Distribution cards break down assets by condition: Healthy, Warning (early signs of issues), Critical (needs immediate attention), and Upcoming Maintenance. If “Critical” is above zero, prioritize those immediately.
Finally, individual Prediction Cards appear for each asset with a detected issue. Each card shows the asset name, a severity badge (Low/Medium/High/Critical), a countdown to predicted failure, a probability bar, a cost comparison (repair cost vs. preventive cost), an expandable AI explanation, and a “Schedule Now” button to create a work order.
Step 5: Monitor Asset Health
The Overview tab gives you the big picture, but sometimes you need to dive deep into individual assets. That is what the “Assets & Health” tab is for. Click on this tab (the second tab at the top of the page) to see every single monitored asset with its detailed health status.
The toolbar at the top offers a search bar (find by asset name or property), Status filter (All/Healthy/Warning/Critical), Type filter (HVAC, Plumbing, Electrical, etc.), and quick filter pills for common combinations. Toggle between Grid View (visual cards) and List View (sortable table) using the view mode buttons.
In Grid View, each AssetHealthCard features a health score ring (0-100 circular gauge: green 70-100, amber 40-69, red 0-39), plus the asset name, property, status badge, last inspection date, and failure probability. In List View, the same data appears in a sortable table – click any column header to sort, which is great for finding the lowest health scores quickly.
Clicking any asset opens the Asset Detail Modal with health factors and “impact dots” (showing each factor’s contribution to the score), AI recommendations (e.g., “Replace filters”), a maintenance history timeline, and a “Schedule Maintenance” button to create a work order directly.
Step 6: Act on Critical Predictions
This is the most important step – where predictions become actions that save you money. When the AI identifies critical predictions, a bright red Critical Alert Banner appears at the top of the Overview tab showing the count of assets needing immediate attention.
Click the “View Critical” button in the Quick Actions bar to filter the dashboard to show ONLY the critical items. This removes all the healthy and warning-level assets from view so you can focus entirely on what matters most. You will see only the Prediction Cards with the highest severity levels.
On each Prediction Card, click the AI explanation toggle to see exactly WHY the AI flagged this equipment. For example: “This 9-year-old Carrier furnace has not had a heat exchanger inspection in 18 months, and similar models show a 73% failure rate at this age without regular inspections.” This transparency helps you evaluate urgency.
Click “Schedule Now” to create a work order pre-filled with asset details, recommended service type, and a suggested completion date based on the predicted failure timeline. Just review, assign a technician, and confirm – the work order flows into your regular job pipeline.
Step 7: Manage Maintenance Schedules
Click the “Maintenance Schedule” tab (the third tab at the top of the page) to see all scheduled maintenance activities in one organized table. Think of this tab as your maintenance calendar in spreadsheet form – it shows everything that is planned, everything that has been completed, and everything that is still waiting for your decision.
The table columns include: Asset (equipment name), Property (location), Date (scheduled), Type (inspection, preventive, or predictive – the AI-triggered ones), Priority (Urgent/High/Medium/Low with color-coded badges), and Status (Completed, Confirmed, Pending, or Skipped).
Entries with the “predictive” badge are AI-generated. For each pending entry, you have three options: Confirm (approve and schedule – also creates a work order if Auto-Schedule is on), Reschedule (change the date), or Skip (decline this maintenance – the AI logs it and adjusts future predictions accordingly).
Important: completed entries automatically feed back into the AI model. When you mark maintenance as completed, the AI updates the health score and adjusts future predictions. The more maintenance you log, the smarter the predictions become.
Step 8: Fine-Tune AI Settings Over Time
Predictive Maintenance is not a “set it and forget it” feature – it gets better as you interact with it and fine-tune the settings based on your experience. After you have been using it for a few weeks, revisit the Settings Slide-Over panel by clicking the gear button in the page header. Here are the adjustments I recommend making at different stages of your journey.
After 2 weeks: Review your predictions. Too many false positives? Raise Confidence Threshold to 75%. Predictions accurate and you want to catch more? Lower it to 65%.
After 1 month: Predictions should be noticeably more accurate. Extend the Prediction Horizon to 45-60 days if you need more lead time for parts and scheduling. Longer horizons mean less certainty, but the tradeoff is worth it for most businesses.
After 3 months: The sweet spot. Lower Confidence Threshold to 55-60% (the AI has learned enough for good lower-confidence predictions). Extend horizon to 60-90 days. Consider turning on Auto-Schedule so predictions automatically become work orders. Adjust Alert Threshold based on notification volume.
Ongoing: Update Monitored Asset Types when you add new equipment categories. The ML Model Version in the settings panel updates automatically as the platform learns from anonymized maintenance outcomes across all customers.
Common Mistakes to Avoid
- Not logging maintenance events – the AI NEEDS service history to learn. Assets with no records produce unreliable predictions. Log every service event, even routine inspections.
- Setting confidence threshold too high (90%+) – this filters out most legitimate warnings. You miss real issues because the AI was “only” 85% sure. Start at 70% and lower gradually as you build trust.
- Ignoring cost comparisons on prediction cards – the “Repair vs. Preventive” numbers are your most powerful tool for justifying proactive maintenance budgets to property owners and management.
- Not monitoring Prediction Accuracy over time – if accuracy drops below 80%, investigate asset data quality: incorrect installation dates, missing maintenance logs, or changed operating conditions are common causes.
- Leaving pending entries for weeks without action – pending entries do not help anyone. Even clicking Skip gives the AI feedback. Set a weekly reminder to review all pending items.
- Turning on Auto-Schedule too early – wait at least one month (ideally three) until predictions are consistently accurate before letting the system automatically create work orders.
What’s Next?
Now that you’ve completed this guide, check out:
Need help? Post in the Tech Support category or contact support@exoserva.com.