Forecasting & Auto-Replenishment · Kava OS Pro
Load the van for tomorrow, not for last Tuesday.
Kava OS turns vending demand forecasting into work a driver can use: an item-level forecast for every device, a pre-kitted pick list for every stop, and a route that only visits machines that need it. Fleets running it see 41% fewer out-of-stocks and 27% fewer service visits.
- out-of-stock events
- −41%out-of-stock events
- service visits per device per month
- −27%service visits per device per month
- per device per month on Pro
- $39per device per month on Pro
Your sales history is lying to you about demand.
When a spiral sits empty from Wednesday to Friday, sales for that item read zero. Demand didn't go to zero. The product did. A moving average learns the stockout as if it were a preference, orders less, and causes the next stockout sooner.
Kava OS forecasts each item at each device, and treats hours spent out of stock as missing demand, not low demand. It accounts for day of week and time of day, site calendars and shift patterns, and the price changes you made, so a markdown doesn't get mistaken for a sudden craving.
- Item-level, device-level forecasts. Cold brew at site 14 isn't cold brew at site 15.
- Stockout hours modeled as censored demand.
- Par levels recalculated from the forecast instead of set once at install.
From forecast to full shelf
Forecast every selection
Each item at each device gets a demand forecast for the window until its next possible visit.
Decide who needs a visit
Devices that will cross below par before the next window go on today's route. A machine that is 70% full on Tuesday can usually wait.
Pre-kit the pick list
Pick lists are built per machine and per spiral, in route order. The warehouse packs one tote per stop, so nobody counts gaps at the machine.
Re-sequence as things change
A device goes offline, a compressor alert fires, a host adds an event: the route is re-sequenced and the driver app updates.
Close the loop
Restock counts recorded at the device feed straight back into the next forecast.
Fewer gaps. Fewer stops.
Fixed weekly routes vs. forecast-driven routes
| What changes | Fixed weekly route | Kava OS forecast-driven |
|---|---|---|
| Which machines get a visit | Every machine on its scheduled day | Machines forecast to drop below par before the next window |
| What's on the van | A best guess plus extra cases | One pre-kitted tote per stop, per spiral |
| Time at each machine | Count, guess, walk back to the van | Swap in the tote, confirm, leave |
| Stockouts between visits | Discovered at the next visit | Predicted and scheduled around |
| Adding a location | Often means adding a route | Fits where visits freed up capacity |
Routing uses live telemetry from Fleet Command. Legacy machines join through Kava Link, which reports real-time sales and inventory.
The things a spreadsheet forgets
Calendars and shifts
Site schedules, holidays, campus terms and shift changes. Finals week at a residence hall is not an ordinary week.
Price effects
Dynamic Pricing changes feed into the model, so a markdown reads as a markdown, not as new demand.
Shelf life
Fresh items are forecast against their sell-by dates, so you stop loading sandwiches that will expire before they sell.
Moving pars
Par levels follow the forecast. The slot that needed 12 in March may need 7 in July.
Van capacity
Pick lists respect what actually fits on the van, and the route respects how long each stop really takes.
Feedback
Every restock is a correction. The forecast learns from what was found on the shelf, not just what was sold.
Students buy breakfast at 1 a.m. during finals. The cooler restocks itself before the morning rush.
How many of tomorrow's stops are visits to machines that are mostly full?
Say a driver makes 25 stops a day, and 7 of them are to machines that didn't need a visit yet. That's more than a quarter of the day spent driving product in a circle. The 27% cut in service visits doesn't come from driving faster. It comes from not going. Pull one route's last month of visits and mark every stop where the driver loaded less than a tote.
Forecasting connects to the rest of Kava OS
Forecasting questions
How long before the forecasts are useful?
Forecasts start from the sales and DEX history you already have, where it's available, and sharpen as live telemetry accumulates. Fresh-food sites and sites with strong weekly patterns tend to settle fastest because the signal repeats.
Does it work on my older machines?
Yes. Kava Link L2 connects to the MDB bus and DEX port and reports real-time sales and inventory per selection. That is all the forecast needs. The median install is 22 minutes.
Do I have to pre-kit to get value?
No. Pick lists help at the machine too. But the 27% reduction in service visits comes from pre-kitting and dynamic routing together: skip the machines that don't need you, and spend less time at the ones that do.
Can drivers and managers override the route?
Always. Managers can lock stops, drivers can flag access problems, and changes flow back into tomorrow's plan. With Kava Agent, route changes can run as Suggest, Approve or Auto.
Which plan includes it?
Forecasting & Auto-Replenishment, pre-kitting and dynamic routing are part of Pro at $39 per device per month, or $33.15 with annual billing. You can try Pro free for 30 days on up to 10 devices.
Where to go next
Run one route on forecasts and compare.
Start a 30-day Pro trial on up to 10 devices and put a forecast-driven route next to your current schedule. Count the stops.