You've sat through six AI demos this year and you still can't say what changes on Monday morning when the front office opens.
The confusion has a straightforward cause: at least four different technologies are being sold under one word. Between November 2025 and July 2026, seven different vendors across facility software, contact centers, and marketing announced an AI agent of some kind, and the machinery behind those announcements ranges from a phone menu with a nicer voice to a model that can carry an unscripted conversation and post a payment to your ledger. Every one of them got the same adjective in the press release.
So, definitions first. Then the work AI in self storage is doing today, and the limits still standing — including two we don't expect to move.
What AI Is, Precisely
AI means software that infers. Instead of executing instructions a person wrote in advance, a model produces its response by drawing on patterns learned from prior data, which is why you can hand it something nobody scripted — a rambling voicemail, a question phrased sideways — and still get a useful answer. Handling the case no one anticipated is the defining trait, and it makes a good first test in any demo.
Three technologies routinely get presented as AI, and each works differently under the hood:
IVR and phone trees. Menu logic. Press two for billing. A warmer recording doesn't change the mechanism.
Rules engines and scheduled triggers. If a balance is outstanding on day eleven, apply the late fee. This is robotic process automation — RPA — and most of the "automation" already running in your operation is exactly this. More below.
Keyword chatbots. A script matched to phrases. It handles the questions someone thought to write down and routes the rest to a contact form.
All three are legitimate tools, and Cubby's platform contains plenty of the second one. The trouble starts when a phone tree gets marketed as an agent, because the operator evaluating it ends up asking the wrong questions.
How AI Works vs. How Typical Software Works
Conventional software runs on explicit instructions: a person anticipated the situations the program would meet and wrote down what to do in each one. An AI model works from learned probability instead, generating the response most likely to be useful given everything it has absorbed. The practical differences look like this:
Conventional software / rules | AI (inference) | |
|---|---|---|
Handles inputs nobody anticipated | No | Yes |
Same input, same output every time | Yes | Not guaranteed |
Auditable line by line | Yes | Reviewed after the fact |
When it misses | Stops and reports an error | Produces a best guess |
Gets better when | Someone rewrites it | The model or its data improves |
Neither column wins. Predictability is exactly what you want when late fees post on schedule and the ledger has to balance; flexibility is exactly what you want when a prospect calls with a question that fits no script. The design question for a storage platform is where to draw the line between the two, and a well-built one draws it deliberately: at Cubby, the model handles the conversation while the actions available to it are governed by hard system permissions that behave like the left column. That architecture split is the substance behind the term AI-native, which otherwise risks being a label every vendor wears.
Where RPA Fits, and Why It Isn't AI
Robotic process automation is software that repeats a sequence of steps a person used to perform by hand. The nightly billing run, the late fee on day eleven, the report that lands in your accountant's inbox on the first, the rate change syncing to your website — RPA, all of it. It never gets bored, never mis-keys a number, and never calls in sick.
A useful shorthand: RPA follows a path someone mapped in advance, while AI is handed a goal and finds its own path. Change the layout of a screen and an RPA bot stalls; change the phrasing of a question and an AI adjusts.
Most operations need both, pointed at different work. Karl Graham of Luminus Capital runs his own call center seven days a week, twelve hours a day, and calls it a strength — but "you scale that by the person," and the eight facilities Luminus acquired last year each brought a flood of first-billing-cycle calls no staffing plan can absorb. He'd planned to hire night-shift coverage before concluding an AI voice agent was "a much easier solution": a computer answering the phone at 2am, so no call goes missed. His larger bet, made on Students of Storage: "anything you do at a computer in the storage industry in the next 12 to 24 months is just going to be completely 100% automated." We'd take the under on "anything," but the direction is right.
What AI Is Doing in Self Storage Right Now
Everything in this table runs in production at real facilities as of August 2026. Read it as a photograph rather than a boundary — the list has grown every year since voice agents first appeared in the industry, and it will be longer by the time this article needs its first update. (The platform-level view of how these fit together lives on our self storage AI page.)
Job | What the AI does | What stays with a human |
|---|---|---|
Inbound calls | Verifies callers, looks up leases, takes payments, issues gate codes, books rentals | Policy exceptions, disputes, callers who want a person |
Pricing | Recommends street rates and existing-tenant increases from occupancy, demand, and tenure | Approving the number, owning the relationship |
Call quality | Grades every call across six criteria and writes a summary | The coaching conversation that follows |
Analytics | Surfaces leading indicators before occupancy moves | Deciding what to do about them |
Tenant communication | Drafts notices, receipts, follow-ups | Final say on exceptions and the hard conversations |
Three of these deserve a closer look.
On the Phone: Answering vs. Resolving
Plenty of products can answer a storage call now. Far fewer can finish one. Finishing means the caller hangs up with the thing done — payment posted, gate code issued, unit booked — and that requires the agent to have write access to the system of record, which is an architecture question no voice demo can settle. (It's also the crux of the build-vs-buy question for AI call handling.)
Cubby's agent — operators know her as Patty — works inside the FMS, so she completes calls end to end: lease lookups, balance checks, payments through a secure keypad flow, autopay enrollment, gate and lock codes after identity verification, unit availability, reservations, live transfers, messages, tasks for managers. In the first seven months of 2026, Patty answered just over 280,000 calls, and 47% of them ended without a human ever joining the call. When you evaluate any vendor, ask for their version of that number and how they define it. Both halves of the answer are informative.
In Pricing: What the Model Actually Sees
The pricing model reads unit-type occupancy, length of stay, competitor rates, and move-out history, then produces a recommendation. A person approves it. Street-rate moves and existing-tenant increases carry very different risks — the second lands on someone who has paid you reliably for three years — so the approval step is doing real work, and we've kept it. (How the recommendation gets made is its own article: AI-based pricing for self storage, explained.)
In Oversight: Grading Every Call Instead of Sampling Six
Call review used to mean a supervisor sampling a handful of recordings a month and extrapolating. AI call grading covers every call — greeting, friendliness, listening, empathy, problem solving, efficiency — with a summary attached to each record, so coaching conversations start from the full picture rather than from whichever six calls got pulled.
What AI Doesn't Do in Self Storage
Vendors rarely volunteer this half of the evaluation, which is a good reason to spend time on it. None of what follows is cause for alarm. Each limit is either a deliberate design choice or a plain fact about how the technology works, and knowing them is how you deploy AI with confidence instead of crossed fingers.
It Doesn't Negotiate
With the collections skill enabled, Patty will tell a tenant in the lien process what's owed and route them to payment — and even then she never negotiates a settlement or discusses auction terms. With it off, she won't discuss the balance at all. Lien law leaves no room for a generated answer; that conversation belongs to a person who can be held to what they said. (Documented in the skills reference.)
It Doesn't Touch Card Data
Phone payments run through a PCI-compliant capture flow: the agent is muted while the caller enters the card number on the keypad, and the tones route directly to the payment processor. The agent receives only non-sensitive outcomes — success or failure, the last four digits, a receipt ID — never the card itself. This tends to be the second question operators ask, so it's worth stating plainly; the full mechanics are here.
It Doesn't Decide Your Policy
The skills that encode policy — discussing lien balances, closing sales, canceling reservations, scheduling move-outs, changing autopay or coverage — all ship turned off, and the operator flips the switches. Sales closing is a three-way choice: always, after hours only, or never. You also shape how she talks: her persona is a written coaching note covering tone, phrasing, and when to escalate, and Knowledge entries teach her each facility's specific rules. The judgment encoded in all of that is yours. The agent carries it out.
It Won't Say the Same Thing Twice
Ask the same question two different ways and you may get two differently worded answers, because responses are generated in the moment rather than read from a script. Our own documentation describes the persona as "a short coaching note, not a rulebook," and that's the right mental model — you shape how the agent speaks and handles situations the way you'd coach a new hire, without dictating her sentences. For work that must come out identical every time, like fee amounts, legal notices, and payment math, the platform uses the rules-based machinery from earlier in this article, where identical is guaranteed.
It Doesn't Replace the Person on Site
Locks, cleanouts, unit walks, auctions, the tenant crying in the office. AI moves the phone and the ledger; the property still needs its people, and nothing about a language model moves a mattress.
It Doesn't Fix Bad Data
One of the sharper observations at ISS Spring 2026 landed on exactly this point: "The same chatbot product, trained on rich integrated data, resolves 80% of customer inquiries. Trained on a handful of disconnected data sources, it might resolve 30% or worse. Same product. Completely different outcomes based entirely on data quality." An agent is only as good as the records underneath it, which is why the FMS being the source of truth — leasing, payments, access, and calls in one system — has quietly become the most important AI feature there is.
Which of These Limits Will Fall, and Which Won't
Most of the boundaries above are engineering problems, and engineering problems erode. Expect the range of calls handled, the quality of document extraction, and the breadth of available actions to keep improving on a steady curve.
Two limits sit on different ground. Generated output can't come with a guarantee of sameness, so the division of labor — inference for conversation, rules for money — will persist no matter how capable the models get. And accountability stays human: someone must answer for the outcome of a lien conversation or a rate decision, and a model can't. When you read a vendor roadmap, sort the promises into those two piles. The first pile is credible. The second deserves an eyebrow.
How to Tell a Working AI From a Good Demo
Another line from ISS 2026 worth keeping in your pocket: most AI success stories told in this industry over the next few years will be "not fabricated, exactly — aspirational." Five questions separate the aspirational from the operational (and if you want the long version, we published twelve questions to ask self storage AI vendors):
Which parts of the product are inference and which are rules? Both are good answers; a vendor who can't tell you is the concern.
What percentage of calls resolve without a human — and how do you define "resolve"?
Which actions can the agent take in the system of record, and which ship off by default?
What happens when it doesn't know — transfer, message, or improvise?
Can I hear three unedited recordings, including one that went badly?
Everyone has a highlight reel. The fifth question is where the conversation gets honest.
Frequently Asked Questions
What is AI in self storage? Software that infers — it handles calls, questions, and data patterns nobody scripted by generating the most probable useful response from learned patterns. Phone trees, rules engines, and keyword chatbots work from predefined instructions and belong under automation.
What's the difference between AI and automation in self storage? Automation, including RPA, executes steps defined in advance and excels at predictable, repeated work like billing runs and late fees. AI handles work that can't be scripted, like open-ended phone conversations, at the cost of word-for-word consistency.
Is RPA the same as AI? No. RPA replays a defined sequence — batch billing, scheduled fees, report exports — with perfect consistency and no understanding. Most existing storage "automation" is RPA, and that's a compliment.
What does AI do in self storage today? In production right now: resolving inbound calls end to end, recommending street rates and existing-tenant increases, grading every recorded call, surfacing occupancy indicators, and drafting tenant communication. The list grows every year.
What can't AI do in self storage? Negotiate lien settlements, handle raw card data, set its own policy, guarantee identical wording on every call, or do physical work at the facility.
Will AI replace self storage managers? No. It absorbs the phone and billing volume; managers keep the judgment calls, the property, and the relationships.
Latest
Industry Insights
A Reservation Is Not a Rental. But Could It Be Even Better?
Cubby Team
・
Read More
Industry Insights
Self Storage Data Services: What Your FMS Should Be Telling You
Cubby Team
・
Read More
Industry Insights
Self-Storage Payment Processing, Explained: Rates, Spreads, and What "Integrated" Actually Means
Cubby Team
・
Read More