Alarm & security · AI ROI

The six numbers that tell you whether an AI investment actually worked

Every vendor has a slide that says ROI. Most cannot tell you which line on your financial statements it touches. This tool asks you for one thing they never do: the number they are promising to move, and by how much. Then it prices that promise against your own loaded costs.

As featured in SDM Magazine: AI ROI for alarm companies: 6 numbers that actually move RMR, attrition and margin ↗ (opens in a new tab)

Your operating baseline×What the vendor promises×Your readiness=What you'd actually collect

A planning estimate, not a forecast. Nothing you type leaves your device.

Loaded with an example dealer, including a set of illustrative vendor claims. Replace every number with yours.
Which numbers is this tool supposed to move?

Most tools touch one or two of these. Pick only what this vendor is claiming, then run the tool again for the next one you are considering.

Leave all numbers selected to measure them all, or unselect the numbers you’re not trying to move with a product to meet your current goals.

Your company

Monitoring, service and inspection. Leave out one-time installation revenue.

Recurring plus installation and everything else. Used for the margin math.

Alarm companies trade at a multiple of RMR, commonly the high 30s to mid 40s depending on size and mix. Use your own if you have a recent valuation.

The investment under test

License, integration, training, and the staff time to run it. Price the one tool you are evaluating right now, not your whole AI budget. If you only count the license, you will flatter the result.

1 RMR growth rate

Not "more growth." A number. If they won't give you one, enter 0 and see what the tool says.

This number, if the promise holds—
How to measure this at your company

The formula. (RMR at month 12 − RMR at month 0) ÷ RMR at month 0. Use gross RMR before attrition, then read it alongside number 2 so you can see whether growth is real or just replacing churn.

Where it lives. Your monitoring platform's RMR report, reconciled to billing. Reconcile it. The two rarely match on the first try.

How long to give it. Twelve months. A quarter of RMR growth tells you almost nothing.

What actually moves it. More qualified proposals out the door, higher close rate, or higher RMR per sale. Ask which of the three this tool touches.

2 Customer attrition

RMR lost over twelve months as a share of starting RMR. If you don't know it, that is itself the finding.

Every point of attrition is RMR you have to replace before you can grow.

This number, if the promise holds—
How to measure this at your company

The formula. RMR cancelled over twelve months ÷ RMR at the start of those twelve months. Track gross attrition separately from net, or acquisitions will hide the problem.

Where it lives. Cancellation log reconciled to billing. Count the accounts that quietly stopped paying, not just the ones who called to cancel.

How long to give it. Four to six quarters. Attrition bends slowly and seasonally.

What actually moves it. Catching service problems before the customer does, contract renewal discipline, and the quality of the first ninety days after install. Ask which one this tool touches.

3 Cost-to-serve per account
Your cost-to-serve: —

Captured, and deliberately excluded from the model. The published productivity research measures customer service contacts, not alarm signal handling. Those are different workloads, and pretending otherwise would inflate your result. If a vendor claims signal-handling savings, price that separately and ask them for their evidence.

This is the best-evidenced of the six. Peer-reviewed field research on AI-assisted support agents found roughly 15% more issues resolved per hour, and the most credible published figures for genuine contact deflection sit near 10–15% year over year. A promise well above that range deserves a hard question.

This number, if the promise holds—
How to measure this at your company

The formula. (Customer service + billing + admin cost) ÷ active accounts. Fully loaded, so wages plus payroll tax, benefits and workers' compensation.

Where it lives. Payroll by department, divided by your active account count. Most dealers have never calculated this. The baseline above may be the most useful number on this page.

How long to give it. Twelve months. It moves quietly and it moves down slowly.

What actually moves it. Fewer contacts per account, or the same contacts handled faster. Those are different mechanisms with different evidence behind them. Ask which one is being sold to you.

4 Truck roll avoidance

Service and installation both. Installation rolls belong here: the tech who waits hours on support, leaves and comes back, or arrives without the right equipment is a second trip you paid for.

The default is burdened payroll of about $55 an hour divided by 80% utilization, which is roughly $69. Paying someone for eight hours and getting six productive ones means the six carry the cost of all eight. This varies widely across the country: California runs high, and your workers' compensation class code moves it more than most owners expect. Work yours out below rather than trusting the default.

Fuel, maintenance, tires, commercial auto insurance, and depreciation on the van, divided by the rolls it makes in a year. Most owners guess low here because only fuel feels like a cost. In California, with fuel where it is, a fully costed roll commonly lands between $200 and $300 once the technician's time is in it. The cost per roll line below adds the two together for you.

Cost per roll: —
Work out your loaded technician rate

Set to 80% by default, deliberately toward the conservative end. Most shops land between 75% and 85% once drive time, waiting on support, restocking and training come out. Burdened payroll divided by this share is what an hour on a truck actually costs, and it is where installation rolls hurt most. Set it to 100% if you want the straight payroll number instead.

Loaded rate: —

No fault found, resolvable remotely, a return trip for parts or equipment, or a return trip because nobody could reach support in time. Only this share is addressable. A real fault fixed in one visit is not a truck roll anyone can avoid, and counting it would inflate your result.

No published research measures AI-driven truck roll avoidance in this industry. There is no benchmark here and this tool does not invent one. Whatever rate your vendor commits to, put it here, and get it in writing.

This number, if the promise holds—
How to measure this at your company

You will have to build this one. Nobody publishes it, which means the only credible number is your own. It is also entirely measurable, which is why it belongs on this list.

Add six disposition codes to your dispatch and install close-out, and make them required:

Resolved first visitNo fault foundReturn trip: parts or equipmentReturn trip: waiting on supportFalse alarmCustomer error

Baseline before the tool goes in. Ninety days minimum. If you start measuring after go-live you will never know what changed.

How long to give it. Twelve months, so you clear seasonality and the ramp.

What contaminates the read. A change in field headcount, a shift in account mix, a weather quarter, or a new install cohort still in its shakeout period. Note these on the chart as they happen, or you will argue about them a year from now.

5 Sales cycle length

Quote to signature to activation. If signing still takes a week of email, no amount of AI upstream fixes that bottleneck.

This number, if the promise holds—
How to measure this at your company

The formula. Median days from quote issued to first billable month. Use the median, not the mean, or one stalled municipal job will swamp the number.

Where it lives. Three timestamps: quote sent, contract executed, RMR activated. Most dealers can produce the first and third and have no idea about the second, which is usually where the time actually goes.

How long to give it. Ninety days, and it is the fastest of the six to show movement.

What this model does and doesn't count. It counts the RMR you collect earlier because activation came sooner. It does not count deals you win because you were faster than a competitor, which is real but not something this tool will invent a number for.

6 Gross margin trend

This one takes no input. It is the reconciliation: everything above, minus everything the tool costs you, read against your revenue. It is where the theater finally meets the books, so the tool computes it instead of asking you to promise it.

Margin effect: —
How to measure this at your company

The formula. Twelve-month rolling gross margin, before and after. Not one month, and not the month the vendor asks you to look at.

The discipline that makes it count. Put the full cost of the tool in cost of sales, including the staff time to run it. Tools get bought out of one budget and run out of another, and that is how a subscription disappears from the margin conversation.

What it tells you. If numbers 1 through 5 moved and margin didn't, the value went somewhere else, usually into headcount you didn't reduce or capacity you didn't redeploy. That is a management finding, not a software finding.

Whether you collect it depends on more than the tool

Two towers do not make a bridge. The technology is one tower, your people are the other, and what carries value across is the architecture between them: redesigned workflow, managers with time to coach, and a deliberate adoption plan. Five questions, and they change the numbers above.

—

Take the measurement kit with you

Your numbers are above, free and unwalled. The kit is the part you keep: a one-page printout with your figures, the disposition codes, the measurement windows, and a vendor commitment worksheet to sign before you buy.

  • Your six numbers and what each promise is worth
  • How to measure each one at your company, and over what window
  • The six disposition codes to add before go-live
  • A commitment worksheet: which number, how much, by when, measured how

Your kit opens right here, ready to print. No newsletter unless you tick the box.

Growth Consultant Services · growthconsultantservices.com

These results are modeled estimates based on the numbers you entered and the promises you were given. They are not a forecast, a guarantee, or a valuation. Actual returns depend on your current state, implementation quality, organizational readiness, and factors outside this model.

This tool asserts no effect sizes of its own. Every promised rate above is one you entered. Where published research exists it is described in plain language next to the relevant question; where it does not exist, the tool says so rather than inventing a benchmark.

Before you act on any of it, talk with an advisor and qualified counsel.

© 2026 Growth Consultant Services. All rights reserved.