Software

Why most ATM AI projects fail

If the model makes incorrect assumptions about underlying conditions, the results from an AI query can lead you in the wrong direction. Here's where ATM AI fails and how they can succeed.

presented by

Photo: istock

November 4, 2026 by ATM marketplace

Summarize withChatGPTGrokPerplexityClaude

Artificial intelligence has taken center stage over the last few years. It's been a groundbreaking advancement in how we approach and complete business projects, including ATM management.

We can embrace the advantages of AI tools: speed, streamlining and efficiency. But we can't ignore the importance of proper interpretation. AI models can extract data, but without context, the ability of that same model to provide insights is limited.

And it's not just a case of "garbage in, garbage out." Even when the data is good, if the model makes incorrect assumptions about underlying conditions, the results from an AI query can lead you in the wrong direction.

Where AI is being used in ATMs

AI can help businesses standardize decisions, preserve ATM data, and move quickly from analysis to action. In the world of ATM management AI excels at:

Documentation
Best-practice documentation can be built out and standardized with AI. In addition, new audiences can be accessed by leveraging AI to summarize and reformat existing documentation.

Automation
AI allows new information to be integrated faster without additional IT development, enabling businesses to use existing tools quickly and consistently.

Quality assurance
OCR (Optical Recognition) can be included as part of quality control mechanisms. AI can also help identify patterns for engineering changes.

Analysis
AI can assist with analyzing data at scale and flagging inconsistencies for review.

Where AI projects can fail

Take the example of gathering data for ATM transactions that were denied. AI is a great resource to find the transactions, furnish error codes, and aggregate the data in a spreadsheet.

In this example, your data is solid and the output is reliable...but it's incomplete. You have the denial rates at your fingertips, but context is missing, leaving you with the "what," but not the "why."

Even when individual denial codes can be interpreted by AI, it takes a human looking deeper into that data to find the underlying reason for a denial. A human will have the experience to better analyze the data in the context of outside influences and their impact on ATM transaction activity.

Are transactions being denied because of user error or device failure? Is it something a third party did? AI can pull the codes, but will it know how the codes affect your overall fleet?

Are devices running out of funds quicker because a high lottery jackpot caused higher than usual ATM cash withdrawals? This single event could have a big impact on uptime, but this background information is unlikely to be known or incorporated by an AI tool when doing its analysis.

ATM transactions make multiple stops and engage a variety of sources. AI doesn't have access to the details of every stop, only the end result. It still takes a person to put the disparate parts together and provide a thorough investigation.

Keys to AI success

AI has enabled us to research, review, and analyze data faster than ever. It's a powerful tool that's here to stay, but it's still evolving. Knowing its limitations can help strengthen, rather than hinder, ATM projects.

Not only do human-in-the-loop guardrails and the establishment of protocols help prevent miscues, they help keep projects on track. Some best practices for AI data projects include:

• Identify and isolate: Ensure that inputs are clear, processes are defined, and exceptions can be isolated for human handling.
• Scope and constrain: Define your goals and the expected format and content of the output. Be clear about exclusions.
• Define and instruct: Provide detailed, explicit instructions.
• Source validation: Determine where the data will be sourced and on what cadence.
• Train and improve: Review output and refine the instructions. Dynamic variables should be revisited regularly.

Bridging the data gap in ATMs

Smaller financial institutions and independent ATM deployers might not have the budget or personnel to fully establish and maintain comprehensive data readiness practices. One way to help prevent failure in data analysis is with Brink's ATM Managed Services.

Brink's helps manage, operate, and enable secure financial transactions. Our services include ATM performance management via the Brink's Insight Center, which processes relevant metrics on the health and performance of endpoint devices and provides users with accurate operational insights.

Brink's ATM Managed Services uses AI strategically with the goal of reducing cashouts and idle cash, preventing ATM failure, improving monitoring, detecting security threats, optimizing service dispatch, and improving asset decisions. Brink's ATM Managed Services focus on lowering total cost, increasing uptime, reducing idle cash, extending asset life, and simplifying operations.

The long-term vision is not simply an ATM that uses AI. It is an ATM operation that can anticipate needs, recommend actions, automate routine decisions, and continuously improve performance. It's a portfolio of services that helps ensure successful ATM operations.

Ready to get more from your ATM data?

Learn how Brink's ATM Managed Services combines technology, operational expertise and actionable insights to help improve uptime, optimize cash, and simplify ATM operations.

Explore Brink's ATM Managed Services →

Included In This Story

---

Brink's

Brink’s ATM Managed Services provides outsourcing help that’s vendor agnostic. We provide services regardless of the hardware or software on current devices. Being able to offload ATMs as-is without replacement ensures that outsourcing costs stay manageable and predictable.

Request Info

Learn More