Cubby's voice AI agent Patty answered just over 280,000 inbound self-storage calls in the first seven months of 2026, and 47% of them ended without a human ever joining the line, per a Cubby blog post published August 3, 2026. That is production volume from an agent with write access to the facility management system, not a phone tree with a warmer recording.
The post arrives as swivl reported 68% containment on 126,920 calls in Q1 2026 and 10 Federal pushed AI call handling toward 80% resolution rates. Cubby's contribution is definitional: it separates inference-based agents from IVR, rules engines, and keyword chatbots, then publishes hard numbers on what its FMS-native agent actually resolves.
What Production Tasks Is Patty Handling Today?
Cubby's August 2026 post catalogs AI work running at live facilities, not roadmap promises.
| Job | What the AI does | What stays human |
|---|---|---|
| Inbound calls | Verifies callers, takes payments, issues gate codes, books rentals | Policy exceptions, disputes, callers who want a person |
| Pricing | Recommends street rates and existing-tenant increases | Approving the number, owning the relationship |
| Call quality | Grades every call across six criteria with summaries | 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 |
Patty sits in the inbound calls row. She completes calls end to end: lease lookups, balance checks, payments through a PCI-compliant keypad flow where the agent mutes while the caller enters card data, autopay enrollment, gate and lock codes after identity verification, unit availability, reservations, live transfers, messages, and tasks for managers.
The 47% no-human figure is the metric Cubby says operators should demand from any vendor: what percentage of calls resolve, and how do you define "resolve"?
How Is Cubby Defining AI Versus Automation?
Cubby's post spends significant space on taxonomy because, as it notes, at least seven self-storage vendors announced AI agents between November 2025 and July 2026, and the machinery behind those announcements ranges widely.
IVR and phone trees execute menu logic. Press two for billing. A warmer voice does not change the mechanism.
Rules engines and RPA run predefined sequences: late fees on day eleven, nightly billing runs, rate syncs to the website. Cubby's platform contains plenty of this. It is not AI.
Keyword chatbots match scripts to phrases and route everything else to a form.
Inference-based agents generate responses from learned patterns and handle inputs nobody scripted. That is Patty's category.
Cubby also draws a deliberate architecture line: the model handles conversation while hard system permissions govern actions like payments and fee posting. Identical outputs for money-related tasks still run through rules-based machinery where consistency is non-negotiable.
Why Does FMS Integration Change the Containment Math?
The difference between answering a call and finishing one is write access to the system of record. Cubby argues that finishing means the caller hangs up with the task done: payment posted, gate code issued, unit booked.
Patty works inside Cubby's FMS, so actions post directly to the ledger. Vendors that answer calls but cannot write to the FMS leave staff completing the action after the AI hands off, which deflates the containment number operators actually care about.
Cubby cites an ISS Spring 2026 observation in its post: the same chatbot product resolves 80% of customer inquiries when trained on rich integrated data but only 30% on disconnected sources. Same product, different outcomes based entirely on data quality.
That framing aligns with the broader industry push toward platform-native automation rather than bolt-on contact center tools, a shift 10 Federal documented moving from 10% to nearly 80% AI call resolution over 18 months and that Modern Storage Media's August 2026 remote-management guide treats as baseline infrastructure.
What Can Patty Not Do, and Why Does That Matter?
Cubby lists limits vendors rarely volunteer. None are failures. They are design choices.
Patty does not negotiate lien settlements or discuss auction terms. She does not touch raw card data; payments run through a PCI-compliant keypad capture flow. Policy skills ship turned off, and operators configure which actions the agent can take. She will not produce identical wording on every call because responses are generated, not scripted. She does not replace on-site staff for locks, cleanouts, unit walks, or auctions.
The post also states plainly that AI does not fix bad data. An agent is only as good as the records underneath it.
For operators evaluating vendors, Cubby proposes five questions: which parts are inference versus rules, what percentage resolve without a human and how is that defined, which actions can the agent take in the system of record, what happens when it does not know, and can you hear three unedited recordings including one that went badly.
The Numbers Worth Writing Down
- Patty calls handled (Jan-July 2026): 280,000+
- No-human resolution rate: 47%
- Data published: August 3, 2026
- Vendor AI announcements (Nov 2025-July 2026): 7+
- Call grading criteria: 6 dimensions, every call
- swivl Q1 2026 comparison: 68% containment on 126,920 calls
- ISS integrated-data resolution cited: 80%
- ISS disconnected-data resolution cited: 30%
- PCI payment flow: Keypad capture, agent muted during entry
Ask for the Number and the Definition
280,000 calls in seven months is a real sample. Forty-seven percent resolving without a human is a real containment rate, even if it trails swivl's 68% on a different network and quarter.
The actionable takeaway is not which vendor wins a headline comparison. It is that FMS-native write access, explicit resolution definitions, and integrated data separate agents that finish calls from demos that sound good on a webinar. Cubby published both the number and the boundary conditions. That is the standard every operator should hold the next seven vendors to.