Legacy vending reports a single useful number: how much money came out of the coin box. Everything else is guesswork dressed up as a report. AI-enabled smart coolers are the opposite - they produce so much data that the danger flips from "not enough information" to "drowning in dashboards." The operators who win are the ones who know which metrics to watch and which to ignore.
This playbook cuts the list down to the KPIs that drive operator decisions: what to stock, what to cut, when to restock, where to raise prices, and when a location is quietly dying. It is written for active operators running at least one AI smart cooler from XMAI or HaHa.
1. Why Analytics Matter More on AI Coolers
A legacy glass-front forces product decisions to be blunt. You have 40 selections, you pick them once a quarter, and you rarely know which ones are losing money. AI coolers have no such excuse - every transaction is timestamped, every SKU tracked per facing, every door-open recorded.
That changes the operator's job. The margin on an AI cooler is not won at the initial stocking. It is won through weekly iteration on the planogram. Analytics is the engine of that iteration.
Core principle: On a legacy machine, stocking is an art. On an AI cooler, stocking is a weekly experiment with measurable feedback. Operators who treat it as an experiment outperform operators who treat it as a habit.
2. The Core Operator KPIs
Start with these six. They apply to every AI cooler in every vertical, and they answer the questions operators actually need answered.
| KPI | What it tells you | Watch cadence |
|---|---|---|
| Velocity per SKU | Units sold per SKU per week. Identifies movers and dead weight. | Weekly |
| Margin per facing | Gross margin $ earned by each shelf position, not each product. | Weekly |
| Days of supply | How many days of inventory remain at current velocity. Drives restocking. | Daily |
| Stockout hours | Hours per week a top SKU was sold out. Direct revenue loss. | Weekly |
| Revenue per door-open | Avg $ per customer interaction. A soft ceiling on basket size. | Weekly |
| Door-open to purchase ratio | Customers who open but do not buy. Above 15% means something is wrong. | Weekly |
Velocity per SKU
Track units per week per SKU, not per month. Monthly numbers smooth over the signal. A product selling 20 units the first week and 4 the fourth week is dying, but a monthly average of 12 hides that. Weekly granularity catches the trend early.
Margin per facing
This is the metric most legacy operators never computed. The right question is not "how much did this product earn?" but "how much did this shelf position earn?" A slow-moving $4 protein bar at 50% margin often beats a fast-moving $1 candy bar at 25% margin - but only if you measure per facing, not per unit.
Days of supply
Every SKU should have a target days-of-supply range. For most captive-audience locations, 5 to 10 days of supply is the sweet spot. Below 5 you are risking stockouts. Above 10 you are overstocking and tying up working capital.
Stockout hours
Every hour a top SKU is sold out is revenue that cannot be recovered. Aggregate this weekly per machine - operators consistently underestimate how much they lose to preventable stockouts.
Revenue per door-open
Useful because it decouples traffic from basket size. A location may show lower total revenue while actually growing basket size - that is a sign the audience is stabilizing around a regular, higher-intent customer base.
Door-open to purchase ratio
Most AI smart coolers will show you how often a customer opened the door without completing a purchase. An abandonment rate above 15% means pricing, selection, or payment friction is pushing customers away.
3. AI-Only Metrics (Impossible on Legacy)
These next metrics are exclusive to AI-enabled smart coolers. They do not exist on a legacy glass-front, and they are where the real insight lives.
Pick-and-return events
When a customer picks up a product, inspects it, and puts it back, the cameras register the event. High pick-and-return on a specific SKU is a signal: interest is there, but the purchase decision is getting stopped. Price? Ingredient label? Packaging? Dig in.
Dwell time per shelf
How long customers spend looking at each shelf before selecting. Short dwell plus high sales equals a destination SKU. Long dwell plus low sales equals confusion - that shelf needs simplification.
Cross-shelf patterns
AI coolers can surface which products are bought together. Customers who buy a protein drink often grab a protein bar - that pairing should be shelved adjacent.
Learning-refresh events
Every time the AI needs to re-learn a product, it is logged. Frequent re-learns mean packaging variability from your supplier or inconsistent facing by your restocker. Both are fixable once surfaced.
4. Time-of-Day and Day-of-Week Patterns
Time-based analytics is where AI coolers pay back their cost most clearly. Every transaction is timestamped, which lets you see patterns legacy operators never could.
- Morning vs afternoon shift. Office locations spike at 10am and 2pm. Hospitals spike at shift changes. Gyms spike at 6am and 5pm. Your planogram should be biased toward morning products up front in a morning-heavy location.
- Weekday vs weekend. Residential locations often flip product mixes between weekdays (office-style snacks and energy drinks) and weekends (indulgence, family-size).
- Payday and pay-cycle effects. In some verticals - construction sites, warehouses - sales spike visibly on the Friday of each pay cycle. Stock up the day before.
- Event spikes. Hotels, stadiums, entertainment venues have predictable event calendars. Feed the calendar into your restocking cadence.
Most AI dashboards include built-in time-of-day and day-of-week views. Use them at least monthly when you review each location's performance.
5. Dashboards to Watch Daily, Weekly, Monthly
Metric discipline matters. Do not check every number every day - you will miss the signal.
Daily (2 minutes)
- Any machine alerts: temperature, connectivity, payment reader faults.
- Days of supply at each machine - any SKU under 3 days goes on tomorrow's restock list.
- Yesterday's revenue by machine vs rolling 7-day average. Any machine down 30%+ deserves a look.
Weekly (30 minutes)
- Velocity-per-SKU ranking at each machine. Bottom three SKUs are candidates to cut.
- Margin-per-facing ranking. Reassign facings to the top producers.
- Door-open to purchase ratio. Anything above 15% gets diagnosed.
- Stockout hour totals. Any top-10 SKU with stockouts needs a planogram facing increase.
Monthly (2 hours)
- Full time-of-day and day-of-week review per machine.
- Cross-shelf pairing analysis - re-layout shelves if a pattern is strong.
- Location-by-location revenue trend across 90 days. Any location flat or declining gets a contract check-in.
6. Alert Thresholds That Actually Work
Dashboards are noisy. Alerts cut through. Configure the following automated notifications - most AI platforms support these directly.
- Temperature out of range for more than 15 minutes. This is the food-safety alert. Must never be ignored.
- Connectivity loss for more than 30 minutes. No connection means no sales logging and no payment processing. Investigate immediately.
- Any top-5 SKU at zero inventory. Priority restock trigger.
- Revenue down more than 40% vs rolling 7-day average. Something is wrong: payment reader issue, door-lock malfunction, or a serious audience shift.
- Pick-and-return event rate above 20% on any SKU. Price or product problem - worth a manual review.
Resist alert fatigue: Do not configure alerts on every metric. Five actionable alerts that you actually respond to beat fifty that you ignore. If you miss an alert two weeks in a row because it did not drive action, delete it.
7. Frequently Asked Questions
Do AI smart coolers provide these metrics out of the box?
Most of them, yes. Velocity, days of supply, stockouts, door-open ratios, and time-of-day breakdowns are standard across modern AI dashboards. Margin-per-facing often requires entering your cost-of-goods into the planogram, which is a one-time setup task.
Can I export the raw data?
Most platforms support CSV or API export. Operators running 10+ machines typically pull data into a lightweight spreadsheet or BI tool for cross-machine comparison.
How often should I adjust my planogram based on analytics?
Weekly for new installations (first 8 weeks), then every 2 to 4 weeks once a location stabilizes. Over-adjusting confuses the AI and fragments sales data.
What is a reasonable door-open to purchase ratio?
85%+ purchase rate is the healthy benchmark for a well-merchandised cooler with rational pricing. Below 80% signals selection, price, or payment-flow friction.
Do I need a full-time analyst to manage all this?
No. A single operator running 1 to 10 machines can handle this in under an hour per week once the rhythm is established. Fleets above 25 machines benefit from a part-time category manager or VendAiMart's managed services.
Want Help Turning Analytics into Revenue?
VendAiMart helps active operators set up the right KPIs, alerts, and review cadences for AI-enabled smart cooler fleets. From planogram optimization to multi-machine dashboards, we build the reporting layer operators actually use.