Answering the phone is the least valuable thing your AI does.
Every operator buying voice AI today is buying call coverage. The ones furthest along stopped thinking about coverage a long time ago. They think about which work is getting done, whether or not anyone called about it. This issue is the data behind that shift, and what it's worth.
Book A Demo→Issue 02. This analysis is drawn from a slice of the swivl operator network, which now processes 300,000+ calls a month. All figures cover Q2 2026 (April 1 to June 30) unless noted otherwise.
Q1 answered one question: does this work? The data said yes, loud enough that the question stopped mattering.
What replaced it is more interesting. Nobody asks about after-hours coverage anymore. It's handled, it's invisible, it's assumed. Operators ask which of their five past-due sequences is underperforming, or why one facility keeps generating review requests but no reviews. They stopped treating the AI like a vendor and started treating it like a team member who needs better instructions.
The phone call was never the job. It was the part of the job that was easiest to see. Underneath it sits a second category of work that decides whether a facility performs: the delinquency list, the lead that went cold, the review nobody asked for, the task that genuinely needs a person. None of it generates an inbound call. All of it moves money. The operators winning right now are not the ones automating the most. They are the ones automating consistently, the same sequence every time, without fatigue.
Our own data caught the shift before we did. Escalation rates improve sharply as a facility settles in, then reverse. We could have left that out of this report. Instead, here is what happened: when we shipped ticketing, real work started getting routed to real people on purpose, with an owner, a location, and a due date. That changed what counts as an escalation.
A closing thought on how we measure any of this. "Containment," the share of calls handled without a person, is the number this industry quotes, us included. It tells you a call was picked up. It does not tell you whether the tenant's problem got solved. Treating those as the same thing is how an industry over-reports progress it has not made. Ticketing was the first real step toward fixing that, because it separates handled from finished. What we are building toward is a sharper standard: not just what percent we contained, but what percent we completed, and what percent still needed a person to close out. That is the next issue. We will show our work.
Three things happening at your facilities right now.
None of this starts with a phone call. All of it shows up in your P&L.
Escalation rates don't fall in a straight line.
We expected them to keep falling the longer a facility had been live. They didn't, and that turned out to be the more useful finding.
Escalation rate by operator tenure, across the three months studied.
The early months show exactly the pattern you would hope for. Sharp, consistent improvement as configuration settles in and the system learns the facility. Past that point the number climbs back, and the reason is a product story rather than a warning sign.
Before ticketing existed, "escalation" was a blunt instrument. A call either stayed with the AI or it didn't, and that binary was the only signal anyone had. Once the obvious inbound work was automated, the real question changed: what happens to everything the AI correctly identifies as needing a person? Ticketing was the answer. It gave operators a mechanism for routing that work deliberately: an owner, a location, a due date, and the conversation that generated it still attached. Instead of vanishing into "escalated" and never being looked at again.
So when escalation climbs in the 180+ day cohort, it does not mean the AI got worse at its job. It means the definition of the job got bigger. Facilities far enough along are running real ticketed workflows, and the work that needs a human is finally visible instead of invisible.
Escalation rate by itself will not tell you whether your deployment is working. Pair it with ticket volume and how fast those tickets close. A rising escalation number next to a shrinking backlog is a system getting more accurate, not less effective.
Roughly 1 in 5 organizations have scaled AI agents past the pilot stage. McKinsey, The State of AI: Global Survey 2026. Most companies are still in month one. This is what month seven looks like.
The work nobody calls you about is the work that pays.
Three automated workflows run on the same layer: collections outreach, abandoned cart recovery, and review or referral solicitation. None of them start with an inbound call. All of them move revenue.
Two of the three have real numbers below. Review and referral solicitation runs on the same automation layer and is the one still without its own breakdown.
Collections
A past-due tenant does not stop being a person with a schedule the moment their account goes delinquent. The operators getting the best results treat outreach as something that happens on the tenant's time, consistently, rather than on whatever afternoon someone gets to it.
Collections outcomes across the three months studied. Share of engaged conversations by action taken.
We are publishing both halves on purpose, and the gap between them is the point. An offer sent is not a job finished. Most reporting in this category stops at the first list because the first list is the flattering one, and an operator reading it has no way to tell how much of that activity turned into money. The second list is the harder number and the one that actually belongs on a board deck.
Nine percent of engaged past-due conversations ending in a completed payment, with no staff member on the line and no afternoon spent working a list, is the number to hold competitors to. Ask any vendor quoting you outreach volume what their completed rate is. Most of them do not measure it.
Abandoned Cart Recovery
Someone starts a rental online and stops. They got distracted, or they wanted to compare one more option, or the phone rang. That person is further down the path than anyone who called to ask about pricing, and at most facilities nobody ever touches them again.
Put that revenue number next to the inbound number later in this report and the argument of the whole issue lands. The inbound engine, every call answered, every lease closed with voice as the last touch, produced roughly $104k in new monthly rent over the period. Cart recovery, a workflow nobody calls you about and most operators never run, produced $60k, out of leads that were already sitting in the system doing nothing.
The average rent number is the other one worth pausing on. A recovered cart rents at roughly $120 a month against $108.67 on the average lease closed from an inbound call. It makes sense once you say it out loud. Somebody who got far enough to start a reservation already picked a unit, and they picked a real one. An inbound caller is often still shopping. A recovered cart has already chosen.
Do the arithmetic on those and something jumps out. Thirty-four percent of 1,387 calls is roughly 470 live conversations. The sequence produced 500+ rentals. More rentals than live phone conversations. Whatever closed those leases, most of it was not somebody talking on the phone. It was a link, in a text, read later.
That is not a weakness in the workflow. It is how the workflow is supposed to behave. People do not answer unknown numbers in the middle of a workday, and a facility calling about an abandoned reservation is an unknown number. They see the missed call, read the text, and rent when they are ready. Anyone selling you outbound voice as a standalone product is not showing you this number, because it is the number that says voice alone would not have worked.
Abandoned cart link click rate by send timing. Same-day was the weakest timing in the entire sequence, behind the seven-day, fourteen-day and thirty-day follow-ups.
Nearly nine times better for waiting a single day. The reason is the same reason the voicemail rate is high. Somebody who just abandoned a cart is still mid-errand. They are in the car, at the old unit, in the middle of a move. A day later they are back at a desk with a minute to deal with it, and the reminder reads as helpful rather than as surveillance. Speed is the wrong instinct here. Timing is the right one.
Before you add a person to work either list, automate it. Neither one needs judgment on the first three touches. They need consistency, and consistency is the thing a human working a list all afternoon cannot give you. Wait a day before you touch an abandoned cart, and do not run either list as a phone-only play. Then ask your vendor for the completed number, not the contacted number.
Real tenants don't stay on one channel. Neither should your AI.
Think about how people actually live now. Nobody runs on a strict 9-to-5 schedule, and nobody expects the businesses they deal with to either. If it is late and you need to pay a bill, book a room, or order dinner, you do it instantly, on your own terms, right then. Self storage has historically asked tenants to do the opposite: pause whatever they are doing and wait until an office opens the next morning for a simple answer.
That gap is where multichannel behavior shows up in the data. Tenants do not pick one channel and stay in it. They call, then text a follow-up question. They text first, then decide they want to talk to someone. A tenant reaching out at an odd hour is not an edge case. It is a nurse coming off a shift, a parent who finally got the kids down, a small business owner catching up after close. That is when working people have time, and they expect an answer through whatever channel is fastest at that moment, not the one the business prefers.
Conversations that crossed both voice and SMS grew roughly 26% over the prior three months among facilities running voice AI. That is the same behavior that makes outbound work: a tenant who ignores a call will answer a text, and a tenant who starts on text will often want to talk. Outreach that can only travel on one channel is leaving most of its results on the table.
More than three-quarters of self-storage operators say they plan to differentiate on customer experience rather than price. Storable, 2026 Self-Storage Industry Outlook. Meeting tenants on the channel they actually reach for is what that looks like in practice.
Do not run voice without SMS. A single-channel deployment loses the tenant at exactly the moment they are most willing to act.
None of this works if the phone call still stalls.
Outbound is where the untouched revenue is. Inbound is the engine that has to keep running underneath it, and it got faster.
Rental flow across the three months studied.
The conversation happens on the phone and the close happens wherever it makes sense for the tenant to finish it. Some want to book on the spot without leaving the call. Others want a link so they can compare, check pricing, or finish while doing three other things. An AI that can only complete a rental inside a single call is optimizing for its own convenience rather than the tenant's.
National self-storage street rates were down 1.5% year over year as of June 2026. RentCafe, June 2026 Self Storage Report. In that rate environment, days shaved off first-contact-to-lease are margin.
Measure time from first contact to signed lease, not just call answer rate. It is the number that connects the phone to the P&L.
We tested six ways to greet a caller. Telling them what to say made things worse.
A short look at how findings in this report get made.
Over 60 days and 2,770 calls, a deliberately sized test slice rather than a full rollout, we ran six different AI greeting scripts against each other to see which kept the most callers from asking for a human right away.
The two variants that coached callers on exact phrasing, telling them "you can say things like gate code or make a payment," backfired. Escalations jumped as much as 6.9 percentage points above control at one site.
The winner did not script the caller at all. It named the AI and listed what it could help with: gate access, payments, reservations, move-outs. No coaching, just disclosure. It was the only variant with zero backfire anywhere it ran, and at the clearest-data site it cut escalations by 2.56 points and first-exchange escalations by 4.87 points.
Network-wide, callers asking for a person up front fell from 33.6% to 25.8% over the same period. Greeting design is part of why.
People do not want a script. They want a system that already knows what it can do.
Full breakdown of all six variants: swivl blog, self-storage voice AI greetings
Seven things to do with this.
- Audit what you are asking your AI to do beyond answering. If it only handles inbound, you are paying for a fraction of it.
- Automate the delinquency list before you staff it. The first three touches need consistency, not judgment.
- Stop scripting your callers. Disclose what the system can do. Do not coach people on what to say.
- Stop reading escalation rate alone. Pair it with ticket volume and time to close.
- Do not run voice without SMS. Tenants switch channels mid-conversation and single-channel outreach loses them. Most cart recovery calls reach voicemail. The rental comes from the text.
- Wait a day on abandoned carts. Same-day outreach got a 2.47% click rate. One day later, 21.57%. Same-day was the worst timing in the whole sequence.
- Expect a curve, not a line. Set your review cadence around tenure milestones, not the first 30 days.
The AI operations platform for self storage. It handles every conversation, and turns what comes out of them into work somebody owns.
Book A Demo→
