What if the AI flags the wrong driver?

Every flag shows the underlying signals, so the manager can sanity-check before acting. False positives drop fast as the model learns the carrier's specific quit patterns — typically from ~15% in week 1 to <5% by week 8.

Will drivers be told they're on a retention list?

No. The list is internal to HR / fleet management. The conversation with the driver is framed as a normal check-in, not a warning. Drivers respond best when they don't feel measured.

How does this work with owner-operators?

O/O settlements have additional dimensions (lease payments, equipment expenses, escrow draws) that are useful retention signals. The AI uses them. O/O turnover has higher exit cost (lost gross + lost truck) so the math is even more favorable than for company drivers.

What about drivers who are simply having a bad week?

The model uses 4-week rolling baselines, not single-week deltas. A single bad week doesn't move the score much. The flag fires on patterns that persist 3+ weeks — which is much closer to a real quit signal than week-to-week variance.

Does this replace exit interviews?

It complements them. The AI catches quit-risk drivers early; exit interviews still happen for drivers who do leave, and the data feeds back into the model. The whole loop gets smarter over time.

Centrix's AI Reads Every Driver's Settlement Every Week — Catching Retention Risks Before the Notice

Q: Will drivers be told they're on a retention list?

No. The list is internal to HR / fleet management. The conversation with the driver is framed as a normal check-in, not a warning. Drivers respond best when they don't feel measured.

Q: How does this work with owner-operators?

O/O settlements have additional dimensions (lease payments, equipment expenses, escrow draws) that are useful retention signals. The AI uses them. O/O turnover has higher exit cost (lost gross + lost truck) so the math is even more favorable than for company drivers.

Q: What about drivers who are simply having a bad week?

The model uses 4-week rolling baselines, not single-week deltas. A single bad week doesn't move the score much. The flag fires on patterns that persist 3+ weeks — which is much closer to a real quit signal than week-to-week variance.

Q: Does this replace exit interviews?

It complements them. The AI catches quit-risk drivers early; exit interviews still happen for drivers who do leave, and the data feeds back into the model. The whole loop gets smarter over time.

The settlement is a safety mirror nobody looks in

A driver's weekly settlement contains more retention signal than any quarterly survey. Mileage. Pay. Deductions. Disputed line items. The pace of complaints to dispatch. The accuracy of the driver's predicted vs actual home time. The pattern of which dispatcher booked which loads.

Most carriers process 50-100 settlements every week and nobody actually reads them. The accounting clerk verifies the numbers and runs payroll. The fleet manager approves the totals. The settlement gets emailed to the driver. Then everyone moves on. Whatever the settlement was telling you about the driver's headspace gets filed, never read.

Centrix's AI reads every settlement, every week, and connects it to the broader pattern of the driver's recent operational history. The output isn't a chart. It's a per-driver, per-problem report the HR / fleet team acts on the same week.

What the AI is looking for

Each weekly settlement gets cross-referenced against:

Current-period earnings vs the driver's prior 4-period average — a

drop is a quitting signal, especially if the driver's mileage was comparable

Mileage trend vs cohort median — drivers tracking 20%+ below their

own cohort for 3+ weeks running are quitting in 60 days

Deduction pattern — unexpected deductions (advances not previously

taken, escrow draws, equipment damage charges) often correlate with driver financial stress

Disputed line items — a driver who's disputing the same line item

every week is a driver who feels mistreated

Lane mix — drivers consistently getting low-RPM lanes from one

specific dispatcher have a specific human conflict to address

Home time variance — actual vs requested home time, especially when

variance is widening

Communication frequency — Telegram bot interactions, voice calls,

email exchanges. Drivers who stop talking are usually deciding.

The AI runs on a per-tenant ML model trained on the carrier's historical quit data, so it learns the carrier's specific quitting patterns rather than relying on generic indicators.

The per-driver, per-problem report

The AI's output is a structured weekly brief delivered to the HR / fleet manager:

> Driver: James Smith (employee #4729, 18 months tenure) > Quit-risk score: 71/100 (was 58 last week) > > Signal 1 — Earnings down 22% vs 4-week average > - Current period: $1,420; prior 4-week avg: $1,820 > - Mileage down 14% (1,890 vs 2,200 baseline) > - Driver took 1 unscheduled day off this week > > Signal 2 — Lane mix shift > - Last 3 weeks dominated by sub-$2.10 RPM lanes > - All from dispatcher Jones (3rd-week pattern) > - Compare: prior dispatcher (Davis) was $2.35 avg > > Signal 3 — Communication drop > - Telegram bot interactions down 60% vs 4-week average > - No outbound calls to dispatch in 9 days (was 3-5/week) > > Recommended action: Manager check-in this week. Suspected cause: > dispatcher reassignment from Davis to Jones triggered lane mix change. > Conversation angle: ask about the dispatcher relationship and the lane > shift; don't lead with the earnings number.

The brief surfaces the why behind the score, not just the score. The manager can have a specific conversation, not a vague "how's it going?"

Why this works (and surveys don't)

Three reasons settlement-based AI catches retention risks better than quarterly engagement surveys:

1. Settlement data is unbiased. A driver who's on the edge but doesn't want to admit it answers the survey "fine, thanks." Their settlement tells the truth.

2. The signal is ambient. The driver doesn't have to do anything special. They just keep working. The AI picks up the pattern from data they're generating anyway.

3. The cadence is right. Weekly is fast enough to catch the curve in the quit-window (when intervention works) and slow enough to filter single-week noise. Monthly would miss it; daily would over-fire.

The conversation, not the warning

The temptation when you have data is to send a warning. Centrix deliberately doesn't. Warnings make drivers feel measured, not heard. The recommended action is always a phone call from someone the driver knows — usually their dispatcher or fleet manager.

Centrix supplies the conversation framework:

*"Hey, your miles have been a little light the last couple weeks.

Anything going on?"*

Listen. Common answers: home-time gap, equipment frustration, pay

confusion, dispatcher friction, personal stress.

The action is small and immediate: re-route to a lane closer to home,

swap trucks, walk through the next settlement, transfer dispatcher, refer to the carrier's EAP.

Carriers running this loop consistently report 30-50% retention recovery on flagged drivers — meaning of the drivers Centrix flags as quit-risk, half get re-engaged through the conversation. The other half quit anyway, but typically with notice and a referral instead of ghosting.

What the math looks like

A 100-driver fleet at industry-baseline 95% turnover replaces 95 drivers/year × $14K = $1.33M/year on the hiring treadmill.

Identify the top quit-risk decile weekly: 8-10 drivers
Conversations recover 30-50% = 3-5 drivers/week
36-60 retained drivers/year × $14K = **$504K-$840K of replacement

cost avoided**

Plus the qualitative wins: stable dispatcher-driver relationships, fewer disrupted lanes, less recruiting pressure, more experienced drivers in the truck.

Where to start

If you're 50+ drivers and don't have a structured weekly retention process:

Connect Alvys settlements + Samsara + Ensilog. The AI needs the

integrated view.

Run the weekly brief in shadow mode for 30 days. The manager sees the

recommendations but isn't expected to act on every one — calibrate signal vs noise.

Pick the top 5 drivers per week for actual conversations. Build the

muscle. By month two, the conversations are routine and the recovery rate stabilizes.

Book a retention review — bring 12 months of settlement data and we'll generate sample weekly briefs against your actual fleet.

The settlement is a safety mirror nobody looks in

What the AI is looking for

Each weekly settlement gets cross-referenced against:

Current-period earnings vs the driver's prior 4-period average — a

drop is a quitting signal, especially if the driver's mileage was comparable

Mileage trend vs cohort median — drivers tracking 20%+ below their

own cohort for 3+ weeks running are quitting in 60 days

Deduction pattern — unexpected deductions (advances not previously

taken, escrow draws, equipment damage charges) often correlate with driver financial stress

Disputed line items — a driver who's disputing the same line item

every week is a driver who feels mistreated

Lane mix — drivers consistently getting low-RPM lanes from one

specific dispatcher have a specific human conflict to address

Home time variance — actual vs requested home time, especially when

variance is widening

Communication frequency — Telegram bot interactions, voice calls,

email exchanges. Drivers who stop talking are usually deciding.

The AI runs on a per-tenant ML model trained on the carrier's historical quit data, so it learns the carrier's specific quitting patterns rather than relying on generic indicators.

The per-driver, per-problem report

The AI's output is a structured weekly brief delivered to the HR / fleet manager:

The brief surfaces the why behind the score, not just the score. The manager can have a specific conversation, not a vague "how's it going?"

Why this works (and surveys don't)

Three reasons settlement-based AI catches retention risks better than quarterly engagement surveys:

1. Settlement data is unbiased. A driver who's on the edge but doesn't want to admit it answers the survey "fine, thanks." Their settlement tells the truth.

2. The signal is ambient. The driver doesn't have to do anything special. They just keep working. The AI picks up the pattern from data they're generating anyway.

The conversation, not the warning

Centrix supplies the conversation framework:

*"Hey, your miles have been a little light the last couple weeks.

Anything going on?"*

Listen. Common answers: home-time gap, equipment frustration, pay

confusion, dispatcher friction, personal stress.

The action is small and immediate: re-route to a lane closer to home,

swap trucks, walk through the next settlement, transfer dispatcher, refer to the carrier's EAP.

What the math looks like

A 100-driver fleet at industry-baseline 95% turnover replaces 95 drivers/year × $14K = $1.33M/year on the hiring treadmill.

Identify the top quit-risk decile weekly: 8-10 drivers
Conversations recover 30-50% = 3-5 drivers/week
36-60 retained drivers/year × $14K = **$504K-$840K of replacement

cost avoided**

Plus the qualitative wins: stable dispatcher-driver relationships, fewer disrupted lanes, less recruiting pressure, more experienced drivers in the truck.

Where to start

If you're 50+ drivers and don't have a structured weekly retention process:

Connect Alvys settlements + Samsara + Ensilog. The AI needs the

integrated view.

Run the weekly brief in shadow mode for 30 days. The manager sees the

recommendations but isn't expected to act on every one — calibrate signal vs noise.

Pick the top 5 drivers per week for actual conversations. Build the

muscle. By month two, the conversations are routine and the recovery rate stabilizes.

Book a retention review — bring 12 months of settlement data and we'll generate sample weekly briefs against your actual fleet.

Centrix's AI Reads Every Driver's Settlement Every Week — Catching Retention Risks Before the Notice

The settlement is a safety mirror nobody looks in

What the AI is looking for

The per-driver, per-problem report

Why this works (and surveys don't)

The conversation, not the warning

What the math looks like

Where to start

Frequently Asked Questions

Centrix's AI Reads Every Driver's Settlement Every Week — Catching Retention Risks Before the Notice

The settlement is a safety mirror nobody looks in

What the AI is looking for

The per-driver, per-problem report

Why this works (and surveys don't)

The conversation, not the warning

What the math looks like

Where to start

Frequently Asked Questions