Guide
Every ROI number you have seen for shop AI was built with somebody else's inputs.
A vendor figure assumes a missed-call rate, a close rate, an average ticket, and a recovery rate. Get those wrong and a good decision looks bad, or a bad one looks excellent. This page does not give you a number. It gives you the formula and the inputs, so the number you end up with is yours and you can defend it.
Read This Part First
Why a single ROI percentage cannot apply to your shop.
Any figure you did not compute from your own inputs describes a different building, a different labor market, and a different customer mix. That is true of numbers published by vendors, by trade publications, and by this page. We do not publish an expected ROI for a shop we have not seen.
What follows is the structure of the calculation, the inputs that decide it, and where an optimistic assumption does the most damage.
The Inputs
Five inputs decide almost the entire answer.
Each of these you can measure this week from your own phone log and your own closed tickets. None requires a new system to find out.
From your phone system's call log, not from memory. Count abandoned calls, voicemails, and after-hours rings. Owners guess low, because a call that never got answered never became a person in front of them.
Of the inbound calls that reach a person, what share becomes a booked appointment. Pull it from your booking log. Do not import an industry figure, because yours is the one applied to your call volume.
Your own, from closed tickets. Be honest about the mix: if much of your work is oil changes, the blended average is lower than owners tend to quote, and the lower number is the correct one.
Count the individual recommendations a customer said no to, not the tickets. One declined ticket may hold three lines. If your system does not log declines individually, this is the first input you cannot measure, and that gap is itself the finding.
Again your own. Declines skew toward larger work than the average ticket, because a customer says no to a four-figure job more readily than to a service. More value per line recovered, but a lower recovery rate.
The sixth input is the suppression dial: what share of each pool is actually recoverable. No honest model recovers all of it. Some missed callers have already gone elsewhere. Some declined work was declined for a reason that has not changed. If the case only works at a recovery rate you would not defend out loud, it does not work.
The Arithmetic
Three terms, written in plain language.
The calculation below is the whole model. It is deliberately unglamorous: no multiplier, no compounding curve, and no assumed growth from month to month, because those are the parts that turn a defensible estimate into a sales document.
Term 3 is not revenue, it is capacity. Returning five advisor hours a week is worth something only if those hours go into work that would not otherwise have happened; if they go into a slower morning, the term is real but smaller than it looks. Nothing here accounts for the cost side, which is deliberate and is the whole subject of the companion guide on what this actually costs. A value figure without a cost figure is half a decision.
Your Inputs, Not Ours
Every field starts at zero. Nothing is pre-filled.
If we shipped this with a house example already loaded, you would anchor on our numbers. Enter yours. The panel runs entirely in this page, sends nothing anywhere, and stores nothing.
Enter your own figures to populate the panel. Until you do, every output stays at zero on purpose.
This is arithmetic on your inputs, not a projection, a forecast, or a RevvOS result. It also says nothing about what any of this would cost. For that, see what automation actually costs a shop.
A Worked Example, With Placeholders
Fill in the brackets. The blanks are yours.
This page will not put numbers in these fields, because any number we chose would be a claim about your shop that we cannot make. The structure is the part that transfers.
| Line | Your input | Where you find it |
|---|---|---|
| Missed calls per week | [ ______ ] | Phone system call log, last four weeks |
| Close rate | [ ____ % ] | Booked appointments divided by answered calls |
| Average repair order | [ $ ______ ] | Closed tickets, last quarter |
| Recoverable share, calls | [ ____ % ] | Your judgement, stated conservatively |
| Declined lines per week | [ ______ ] | Decline log, counted per line rather than per ticket |
| Average declined line value | [ $ ______ ] | Decline log, averaged |
| Recoverable share, declines | [ ____ % ] | Your judgement, stated conservatively |
| Advisor hours per week | [ ______ ] | A week of time-tracking, or an honest estimate |
| Loaded advisor cost per hour | [ $ ______ ] | Wage plus payroll burden, not wage alone |
If you fill in that table and the total does not justify a change, that is a valid result and you have saved yourself a project. Writing it down is how you say no to something that does not clear the bar, as much as how you say yes to something that does.
Where The Answer Moves
Two inputs carry most of the risk of being wrong.
Term 1 multiplies four inputs together, so an optimism error in any one propagates straight through and errors in two compound. Term 3 adds its inputs once each, so it is far harder to inflate by accident and usually the smaller term.
Missed calls per week. The number nobody measures and everybody underestimates, and the one that multiplies into everything else.
The recovery rate. It is the one input with no source document behind it, which is exactly why optimistic models lean on it hardest.
Average repair order and line values from closed tickets. Past performance is not a forecast, but it is at least evidence.
Reading A Vendor Claim
Four questions that separate a model from a pitch.
None of these are hostile. A vendor with a real model will answer all four easily, and the answers will be the same numbers you would use.
1. What missed-call volume is this built on?
If the answer is an industry average rather than a range the model tests, the output describes an average shop, not yours.
2. What recovery rate, and why that one?
A model that assumes most of a pool comes back is not modelling, it is asserting. Ask what happens to the answer at half that rate.
3. Is the forecast compounding?
Multi-year projections that grow the benefit every year assume the first year's gains multiply. Usually they repeat.
4. Does it include the running cost?
A value figure without the subscription, the implementation, and the staff time to run it is not an ROI. It is a revenue estimate with the word ROI attached.
Applied here: we gave you the structure and refused to supply the inputs, which means we cannot hand you a return figure at all. That is the intended outcome. An honest ROI for your shop depends on numbers only you have, and anyone quoting you a percentage without asking for them is describing their reference customer.
Bring your inputs
We will build the model with your numbers, and tell you if it does not clear the bar.
Four weeks of call logs and a quarter of closed tickets is enough.