Self Storage AI Metrics, Decoded: What Deflection, Containment, and Resolution Actually Mean

# Self Storage AI Metrics, Decoded: What Deflection, Containment, and Resolution Actually Mean

You're 40 minutes into a demo. The rep on the other side has said "82%," "68%," and "90%" in the last ten minutes, and every time the number meant something different. One was rentals completed without a human. Another was calls that stayed inside the AI. A third was calls resolved on the first try. You nod along, because arguing about definitions would take another 40 minutes and you have a tour at three.

That is the current state of buying AI for a self storage front desk. Every vendor has a headline percentage. Nobody agrees on what it measures. And the operators reading press releases in Inside Self-Storage this month have no way to tell whether 82% is better than 68% or worse than 90%, because the three vendors publishing those numbers are counting different things.

This post is the decoder ring. Five metrics your vendor will quote, what each one hides, and the questions to ask before you buy on a slide.

Why the numbers don't line up

There is no shared definition of "AI handled the call" in self storage. In one vendor's math, a call counts as "handled" if the AI stayed on the line the entire time, even if the caller never got what they wanted. In another's, it counts only if the caller's problem was solved and no follow-up task was created. In a third, it counts if the AI booked a rental with a card swipe, but not if the caller was an existing tenant paying a bill.

These aren't small definitional differences. They swing headline numbers by 20 to 30 percentage points on the same book of calls.

The broader voice AI industry has been arguing about this for two years. Containment is now its own category with its own benchmarks: enterprise procurement guidance tends to anchor around 80%, while independent research on live deployments puts most pilots closer to the 45 to 65% range. Self storage is late to that conversation and picking up its definitions from marketing decks instead of from research.

Here are the five you'll actually hear.

1. Deflection rate

What it says: The percentage of calls that never made it to a human being.

What it hides: Whether the caller's problem got solved. Deflection is a routing metric, not an outcome metric. A caller who hangs up in frustration counts as "deflected." So does a caller whose card was declined and who was handed the front-desk voicemail. Both look like wins in the number.

In the wild: Some vendors put their voice AI at "35% to 50% of inbound calls deflected." That range is at least honest about what it measures. It's a call-routing statistic, not a resolution statistic.

Question to ask: "What percentage of deflected calls generated a follow-up task or a callback the next day?" A vendor with a serious answer will have that number to hand.

2. Containment rate

What it says: The percentage of calls that stayed inside the AI from start to finish, no transfer, no escalation, no drop-off.

What it hides: Whether the caller got what they came for. Containment is a duration-and-routing metric. It says the AI kept the line, not that the AI resolved the reason for the call.

In the wild: A quarterly industry report published this spring led with a pair of big numbers: roughly 127,000 calls handled, 68% "resolved by AI," nobody on the line. That is a containment number reported as resolution. The useful question sits on the other side of it. What happened in the 32% that escalated to a live agent, and why? That is the half nobody puts on the slide, and it's the half that tells you how the system behaves when a call gets hard.

Question to ask: "For the calls that stayed inside the AI, what percentage had a customer outcome logged: payment taken, unit rented, gate code sent, ticket filed?" This is the metric that turns containment into resolution.

3. Resolution rate

What it says: The percentage of calls where the caller's problem was solved before hangup.

What it hides: How "solved" is defined. Was a payment completed, or was a payment attempted? Was a gate code sent, or was the caller told a code would arrive? Was a rental signed, or was a hold placed?

Our number: Lumio's production data across nearly a dozen operators shows nearly 6 in 10 calls resolved without a transfer, at an average call length under 90 seconds, and nearly 3 in 4 payment calls handled end to end. Those are outcome-anchored numbers: the caller's card was charged, the code was delivered, or the reservation was written back to the FMS.

Question to ask: "Walk me through your definition of 'resolved.' What has to happen inside the call for it to count?" If the definition is "the AI said something helpful and the caller didn't ask for a human," that isn't resolution. That's containment with better branding.

4. First-call resolution

What it says: The percentage of callers who didn't have to call back about the same issue.

What it hides: Two things. First, most vendors don't have visibility into callback behavior across the customer's whole phone system, so this number is estimated from call-log stitching, not measured. Second, "first call" often ignores the caller who abandoned mid-flow, tried the mobile portal instead, and never called back because they gave up.

In the wild: "90%+ first-call resolution" shows up on more than one vendor's materials. A 90% number is credible when the FMS integration is tight enough that a payment write-back is atomic and doesn't spawn a "your payment didn't post" follow-up call the next day. When the integration isn't that tight, the metric quietly absorbs those follow-ups instead of counting them. Ask which one you're looking at.

Question to ask: "How do you count first-call resolution on a payment call: at the moment the caller's card is charged, or at the moment the payment posts inside the FMS?" The gap between those two events is where "resolved" starts to leak.

5. "No human involvement" or autonomy rate

What it says: The percentage of end-to-end outcomes (rentals, payments, cancellations) that the AI completed on its own, without a human touching the transaction.

What it hides: Which categories are included. This is the newest metric in the space and the one with the most variance in what it measures.

In the wild: A recent industry announcement claimed "82% of rentals completed without human involvement." That is a rental-only metric, not a call metric, and it's a strong one for what it covers. What it doesn't describe is the rest of the phone volume: existing-tenant payments, gate access, move-out notices, complaints. If about 3 in 4 self storage calls come from existing tenants (Lumio's own portfolio data), a rental-completion metric is telling you about a slice of the smaller lead-facing cohort, not the bulk of the book.

Question to ask: "What is your autonomy rate across the full call mix, not just rentals? Payments, gate access, move-outs, and complaints included." A vendor with real depth will break it down by category. A vendor without depth will pivot back to the rental number.

Before and after: the same call book, two different narratives

Here's the shape of the game. Numbers are illustrative, drawn from the ranges above.

> Before (the headline): "This vendor's AI handles 82% of calls." > > After (the honest math): Of the calls handled, 60% were payments where a card was charged and the transaction wrote back to the FMS. 15% were gate-code requests resolved by SMS. 5% were rental completions. The remaining 20% were calls that stayed inside the AI for the full duration but ended with either the caller hanging up or a task filed for a human to complete tomorrow.

Both statements can be true about the same book of calls. The first belongs in a press release. The second belongs in the operator's judgment about whether the vendor is worth the money.

Six questions to bring to any vendor demo

Ask these in order. If a vendor can't answer at least four of them without a follow-up email, the number they quoted is a slide, not a measurement.

1. What's the exact denominator? All calls, or only calls the AI answered? Excluding wrong numbers, or including them? 2. How is "resolved" defined? Payment posted, code delivered, rental written back, or "AI ended the call and no one asked for a human"? 3. What percentage of resolved calls generated a follow-up task? The gap between resolution and containment lives here. 4. Break the rate down by call category. Payments, gate access, leads, move-outs. A single blended percentage hides where the AI is strong and where it collapses. 5. What percentage of calls the AI handled came from existing tenants versus new leads? New-lead-heavy books look artificially high on resolution because the flow is simpler. 6. Can I see the number for calls that arrived after business hours, separately from the ones during business hours? After-hours volume is roughly 1 in 3 calls in self storage, and the after-hours cohort behaves differently.

What "good" looks like across the mix

At the risk of being one of the vendors that quotes a number: across nearly a dozen production operators and thousands of real calls, Lumio's own book runs at nearly 6 in 10 calls deflected, nearly 3 in 4 payment calls resolved end to end, and an average call length under 90 seconds. Those are honest, category-aware numbers. What matters isn't whether they're higher or lower than the 68% or the 82% on somebody else's slide. It's that you can ask us the six questions above and get an answer that lines up with what a facility manager would recognize from a Monday morning at the desk.

The right number for your book depends on your call mix, your FMS integration depth, and how you count. Any vendor who can't describe those variables is quoting a slide. Any vendor who can is probably worth the demo time.

Keep reading

Ready to see the honest math on your own call book? [Book a demo](https://www.lumiostorage.com/book-demo) and we'll walk you through what "resolved" looks like across the full call mix.