A computing platform that detects when a user is experiencing difficulty at an ATM, identifies the specific usage issue from sensor and presence data, and dynamically generates and transmits targeted assistance to the machine — before the user has to ask for help or abandon the transaction.
ATMs are equipped with error detection — card reader failures, PIN entry timeouts, low cash — but have no mechanism to identify when a user is having trouble understanding or navigating the interface. A first-time user who doesn't know the card orientation, an elderly user unsure which receipt option to select, or someone confused by a foreign-language menu all generate the same non-response that the ATM cannot interpret. The machine waits, then times out.
A central computing platform receives user presence information from the ATM's sensor systems — indicating that a user has been detected and is active at the machine. The platform analyzes this presence data to identify a specific usage issue: card insertion error, PIN confusion, menu navigation hesitation, receipt selection pause. It then generates targeted user assistance information and transmits it back to the ATM system for display.
The assistance is dynamic — generated in response to the detected issue, not retrieved from a static library. A user struggling with card orientation receives card insertion guidance. A user stalling on the amount selection screen receives a prompt for the most common amount choices at that ATM. The help is specific to what's happening right now, not a general help menu.
The system has two major components: the ATM sensor layer (which detects user presence and interaction state) and the computing platform (which receives the presence data, identifies the usage issue, and generates targeted assistance). The ATM itself is a data source and a display endpoint — the intelligence lives in the platform.
This separation allows the platform to aggregate patterns across multiple ATMs — learning what types of issues occur at which machines and times — while keeping the ATM's own compute requirements minimal. The platform's assistance generation can incorporate user account context, transaction history, and machine-specific interface quirks.
The ATM's sensor system generates presence information that characterizes the user's state: time since card insertion, number of PIN re-entries, duration on a specific screen, touch input patterns. This granular presence data is the raw signal from which the platform identifies usage issues — without requiring the user to self-report or press a help button.
The presence information includes both positive signals (user is actively interacting) and hesitation signals (user has been on the same screen for longer than the 95th percentile interaction time). The platform uses both types to classify the issue — active repeated input errors vs. passive hesitation carry different assistance implications.
The platform classifies the usage issue from the presence data pattern. A user with three card re-insertions in thirty seconds has a card orientation issue. A user dwelling on the language selection screen for 45 seconds may have a language barrier. A user who progressed to amount entry then reversed has changed their mind and may need guidance on how to start a new transaction.
Each classified issue maps to a specific assistance type. The platform doesn't just detect that something is wrong — it identifies what is wrong precisely enough to generate assistance that addresses the exact step where the user is struggling. The 19 claims cover the full issue-identification-to-assistance pipeline.
Repeated card re-insertions → platform identifies card orientation error → sends card insertion guidance animation to ATM screen.
Repeated PIN failures + dwell → platform identifies PIN entry issue → sends PIN entry guidance with option to reset or use alternative auth.
Excessive dwell on selection screen → platform identifies decision paralysis → sends simplified option summary highlighting the most used choices.
No interaction on language screen + dwell → platform detects language selection hesitation → transmits language options prominently with visual cues.
The platform generates assistance information dynamically based on the identified issue, the specific ATM's interface, and the user's context. The generated content is transmitted to the ATM system for display — replacing the current screen or overlaying a help layer that guides the user through the identified stumbling point.
The assistance generation can incorporate machine-specific state: the ATM's current step, the interface version in use, and the options available at that machine. A user struggling with an amount selection at an ATM that supports custom amounts gets different guidance than a user at one that only offers preset amounts. The platform knows the difference because it manages the machine's session context.
Because the platform manages multiple ATM systems, it accumulates a dataset of usage issues per machine: which steps cause the most hesitation, at which times, for which user segments. This cross-machine learning improves issue identification accuracy — a step that consistently causes 60-second dwell times at a particular ATM model is pre-flagged as a high-hesitation step, lowering the threshold for triggering assistance.
The platform can also proactively push interface adjustments to specific machines: if a step is known to cause widespread confusion, the machine can display preemptive guidance before users encounter the issue rather than waiting for presence signals to indicate a problem has occurred.
Platform aggregates issue frequency across all managed ATMs — identifies which steps, machines, and times generate the most usage issues.
Issue identification thresholds are calibrated per machine and step — a 30-second dwell on a known confusing screen triggers earlier than on a simple confirmation step.
Platform can push preemptive guidance to an ATM for a known high-issue step before a user exhibits a hesitation signal — preventing the issue before it starts.
User's account context (first-time ATM user, international card holder, accessibility needs) informs both issue identification sensitivity and the type of assistance generated.
Proactive usage issue detection and dynamic assistance generation transforms the ATM from a passive interface into an actively supportive one — reducing transaction abandonment, improving completion rates for first-time and at-risk users, and eliminating the need for on-site staff to intervene in routine usage issues.
No forward citations on record as of June 2026. US12223479B2 was granted in February 2025 — forward citations typically begin appearing 12–24 months post-grant as practitioners and examiners reference newly published claims.