More vehicles
Same events, less time
- Days
- Fall in proportion
- Cost
- More devices to fit and follow
Evaluation · Free planner
Tracker and camera pilot sizing: how many vehicles and how many days?
A pilot with too few events proves nothing. Enter the event you count, your fleet and your calendar. The planner gives the vehicles and days that collect enough events, and the math is on the page.
- Runs in your browser
- Exact Poisson interval
- Formulas shown below
The event you count
Any event your tracker or camera logs, such as harsh braking or speeding. It appears in the report.
Rate is given per
Use the vehicle-day if you know events per working day. Use 1,000 km if your report gives events per distance.
per day
Typical value 0.8. Replace it with the average from your own tracker reports. Count only days when the vehicle worked.
per 1,000 km
Typical value 8. Replace it with your own figure.
km
Typical value 100. The planner multiplies it by the rate per kilometer.
×
Typical value 1.5. Use 1 if every vehicle behaves alike. To measure it, divide the variance of the weekly counts per vehicle by their mean.
What you want from the pilot
The goal
Measure the rate gives you a trustworthy baseline. Detect a change tests whether a program, such as driver feedback, cuts the event.
How sure you want to be. 95% is a typical choice.
%
The true rate should fall within this share of the measured rate. 20% is a typical choice.
%
The smallest fall in the event rate that matters to you. 25% is a typical choice.
If the cut is real, how often the test should see it. 80% is a typical choice.
Collect it in the pilot doubles the length. Choose History if you already have tracker data for the same vehicles over the same length.
Your fleet and calendar
vehicles
Vehicles you can fit and follow in the pilot.
groups
Vehicle types, shifts, routes or sites that behave differently. Enter 1 if there is only one.
vehicles
Typical value 2, so one odd vehicle does not decide the result.
days
Typical value 6. Days when the vehicle normally works.
%
Typical value 85%. Allows for workshop days, leave and time offline.
days
Typical value 14 days, so the period covers two full weeks.
days
Typical value 30 days. A before-and-after plan counts both periods.
How this is calculated
Measure: events = φ × N_eff, where N_eff is the smallest count with an exact interval of ± accuracy
Change: vehicle-days = φ × (zα + zβ)² × (λ0 + λ1) ÷ (λ0 − λ1)²
Days = vehicle-days ÷ (vehicles × availability × days per week ÷ 7)
The full method is below.
Uses the event rate, goal and calendar from the first tab. Enter the vehicles and days you plan to run.
vehicles
The vehicles that will report during the pilot.
days
Calendar days. A before-and-after plan has two periods of this length.
Pilot sizeIndicative
Plan checkIndicative
Take the sizing to your pilot plan
Report with settings, results and method.
Pilot sizing sheet
Get the sizing sheet for your pilot plan
- The event, rate and goal you entered
- Vehicles and days that collect enough events
- Table of options by number of vehicles
- What changes if your rate guess is wrong
- Method and next steps
When the plan does not fit
Three ways to shorten a pilot
The number of vehicle-days is fixed by the event and the accuracy you want. You can move only the three things below.
More days
Same vehicles, longer run
- Vehicles
- Fall in proportion
- Risk
- Season, routes or drivers may change
A more frequent event
Fewer vehicle-days for the same accuracy
- Vehicle-days
- Fall as the rate rises
- Check
- The event must matter to your decision
What does the pilot sizing planner tell you?
A telematics pilot answers a question: how often does an event happen, or does a program cut it? A pilot with a handful of events cannot answer either. This planner turns your question into numbers. It finds how many events you need, how many vehicle-days that takes, and how many vehicles and days fit your fleet and calendar.
Two goals are built in. Measure the rate gives you a baseline you can trust, for example harsh braking per vehicle per day. Detect a change tests whether the rate falls after you start a program such as driver feedback. A second tab checks a plan you already have.
Why events, not vehicles, decide a pilot
- Precision comes from the count. Ten events give a rate that can be half or double the truth. A hundred events narrow it to about one fifth either way.
- A rare event needs many vehicle-days. At 0.05 events per vehicle per day, 153 events take 3,060 vehicle-days.
- Vehicles are not all alike. Routes, shifts and drivers differ. The spread factor adds events to allow for it, and the group minimum makes sure each type is in the pilot.
- Time has a floor. Even a large fleet needs at least a full weekly cycle, so the shortest period stays at 14 days unless you change it.
Inputs you need
| Input | Value |
|---|---|
| Events per vehicle per active day | 0.8typical |
| Spread between vehicles | 1.5×typical |
| Confidence | 95%typical |
| Accuracy of the rate | ± 20%typical |
| Cut to detect | 25%typical |
| Chance of seeing it (power) | 80%typical |
| Share of active days with data | 85%typical |
| Shortest period | 14daystypical |
| Vehicles per group, at least | 2typical |
Typical values for a starting point. They are not Tracom device specifications. Your own tracker reports replace them.
| Input | Where to find it | Tip |
|---|---|---|
| Event rate | Tracker or camera event report | Take at least two weeks of recent data and divide by active vehicle-days. |
| Spread factor | Weekly counts per vehicle | Variance divided by mean. Leave 1.5 until you have data. |
| Groups | Your fleet list | One group per vehicle type, shift or site that behaves differently. |
| Availability | Workshop and leave records | Include days a device is offline. |
How the pilot is sized
Events of one kind that happen at random with a steady average rate follow a Poisson distribution. The planner uses that standard model and widens it with the spread factor φ, so the variance is φ times the mean. This is the quasi-Poisson approach. The effective count is the events divided by φ.
Measure the rate
Interval for a count n at confidence c (exact, Garwood):
lower = Gamma quantile (α/2, n) upper = Gamma quantile (1 − α/2, n + 1)
Smallest N_eff whose half-width (upper − lower) ÷ 2 is at most the accuracy × N_eff
Events = φ × N_eff Vehicle-days = events ÷ rate
The normal shortcut (z ÷ accuracy)² gives 96 events for ± 20% at 95%. The exact interval needs 102, because a Poisson interval is not symmetric.
Detect a change
λ1 = λ0 × (1 − cut)
Vehicle-days per period = φ × (zα + zβ)² × (λ0 + λ1) ÷ (λ0 − λ1)²
zα: two-sided normal value for the confidence zβ: value for the power
Power of a plan = Φ( (λ0 − λ1) × √vehicle-days ÷ √(φ × (λ0 + λ1)) − zα )
Both periods have the same exposure. The formula is the standard normal approximation for two Poisson rates. It is rough when a period has fewer than about 20 events, and the planner warns you.
Vehicles and days
Active vehicle-days per vehicle per calendar day = availability × days per week ÷ 7
Days per period = vehicle-days ÷ (vehicles × that figure), never below the shortest period
Recommended vehicles = the smallest number that finishes inside your longest pilot,
and is at least groups × vehicles per group
The verdict is Ready to plan when your pool is at least 20% larger than the recommended number, so vehicles that drop out do not break the plan. It is Possible with little spare below that, and Not possible when the pool or the days are too small.
Worked example: harsh braking in a mixed fleet
- Inputs: 0.8 events per vehicle per day, spread 1.5, ± 20% at 95%, 2 groups of at least 2 vehicles, 6 days a week, 85% availability, up to 30 days, 40 vehicles in the pool.
- Events: N_eff = 102, so events = 1.5 × 102 = 153.
- Vehicle-days: 153 ÷ 0.8 = 192 active vehicle-days.
- Vehicles: Each vehicle gives 0.85 × 6 ÷ 7 = 0.729 active days per calendar day. For 30 days: 192 ÷ (30 × 0.729) = 8.8, so 9 vehicles.
- Verdict: A pool of 40 is far above 1.2 × 9, so the plan is ready.
- Change instead: To see a 25% cut (0.8 to 0.6) at 95% and 80% power with spread 1: 7.849 × 1.4 ÷ 0.04 = 275 vehicle-days per period, so 26 vehicles for two periods of 15 days.
The planner opens with other example values, so the numbers on screen will not match this example until you enter it.
How to read the results
- Lamp and word. Ready to plan, Possible with little spare, or Not possible. The sentence under it says why.
- Table by vehicles. Shows how long the pilot lasts with more or fewer vehicles, and which options fit your longest pilot.
- If your rate guess is wrong. The same plan at half and at double your rate. Half the rate doubles the vehicle-days.
- Check my plan. Shows the accuracy or the chance of seeing your cut that a plan gives, and the smallest cut it can see.
Common mistakes when sizing a pilot
- Counting vehicles, not events. Ten vehicles for a week may hold 40 events or 4. The count decides the accuracy.
- Using days when the vehicle did not work. Rate per active day only. A vehicle in the workshop adds no events.
- Comparing different seasons. Heat, holidays and Ramadan change driving. Keep both periods alike or add a control group.
- Choosing a rare event. A collision rate needs years of fleet data. Use a frequent event as a stand-in for the pilot.
- Treating the result as a guarantee. It is an estimate from a guess of the rate. Run it again after the first week.
Assumptions and limits
The planner assumes that events are independent and that the average rate is steady during the pilot. It assumes equal exposure in the two periods of a before-and-after plan, and no other change between them. Real fleets break these rules to some degree. The spread factor, the availability and the 20% spare in the verdict are there to allow for it. All defaults are typical values you set, not measured results and not Tracom device specifications. The result is indicative. Use it to plan, then check it against the first week of real data.
Related Tracom solution
Plan the pilot with Tracom
Tracom designs its devices around second-by-second data. Ask which events your device model records. Tell us your vehicles, groups and the event you want to measure, and we will help you plan the pilot.
- Second-by-second data. Records how the vehicle moves in fine detail.
- Driver behavior monitoring with voice alerts. Ask which events your device model offers.
- SCMS portal. Manages device settings and configuration files in one place.
- Advanced configuration files. Ask how to keep the same thresholds in both periods of a before-and-after test.
How many vehicles does a tracker pilot need?
It depends on how often the event you measure happens. The planner counts events, not vehicles. It finds the events needed for the accuracy you choose, then the vehicle-days that collect them. A frequent event needs few vehicles. A rare one needs many vehicles or a long pilot.
Why does the planner use a Poisson count?
Events that happen at random with a steady average rate follow a Poisson distribution. It is the standard model for counts such as harsh-braking events. The spread factor widens it for real fleets, where vehicles differ.
What is the spread factor?
It is the variance of the event counts divided by their mean. At 1 all vehicles behave alike. Above 1 the counts vary more, so you need more events. Replace the typical 1.5 with a figure from your own weekly counts per vehicle.
Can I test a driver feedback program with it?
Yes. Choose Detect a change. The planner assumes equal exposure before and after and no other change, such as season or routes. A control group of vehicles without feedback makes the test stronger.
Is the result a guarantee?
No. It is an estimate based on a guess of the event rate. If the real rate is half your guess, you need twice the vehicle-days. Run the planner again after the first week of real data.
Does Tracom set or approve the size of my pilot?
No. The planner is a neutral statistical estimate from the numbers you enter. Tracom does not set pilot sizes or approve results. Tracom makes tracking and video devices.
Plan your pilot
Start a pilot that can settle the question
Send your vehicle count, groups and the event you want to measure. We will go through the sizing with you.
Related tools and templates