February 13, 2026

If you feel like AI terminology is multiplying faster than your ability to track it, you are not alone. The goal for executives is not to keep up with every new acronym, it is to understand the basic categories so you can ask better questions and make better decisions.

You can think of AI capabilities in a handful of practical families. Most vendor offerings mix several of these under the hood, but this list helps you decode what you are really buying.

Traditional machine learning

  • This learns patterns from historical data to predict what is likely to happen next.
  • Examples: credit risk scoring, fraud detection, deposit and cash forecasting, member churn prediction.
  • What it needs: enough past data with known outcomes, plus reasonable data quality. Bad data makes it confidently wrong at scale.

Predictive analytics

  • This is often machine learning plus statistics, presented as forecasts and “likely outcomes” for business users.
  • Examples: loan demand forecasts, staffing forecasts for the contact center, product uptake projections.
  • What it needs: agreed business definitions, consistent historical data, and clear owners who will use the predictions in planning.

Generative AI

  • This creates new content rather than just analyzing existing data.
  • Examples: drafting member emails, summarizing call notes, creating first drafts of policies, producing talking points for branch staff.
  • What it needs: clear rules about where it is allowed, what data it can see, and which outputs require human review before going to members or regulators.

Natural language processing (NLP)

  • This allows systems to understand and work with human language in text or speech.
  • Examples: extracting data from loan documents, routing member messages by topic, scanning communications for emerging complaints or compliance signals.
  • What it needs: exposure to your actual documents and vocabulary, plus validation that accuracy is good enough for the decisions you plan to support.

Computer vision

  • This interprets images and visual documents.
  • Examples: ID verification, check image analysis, reading paystubs or tax returns, verifying signatures.
  • What it needs: consistent image quality, a representative set of examples to tune against, and clear fallback paths when confidence is low.

Recommendation and personalization engines

  • These sit on top of your data to decide “what next” for each member.
  • Examples: next best offer in digital banking, collections strategies tailored to member behavior, personalized financial education nudges.
  • What it needs: reliable member profiles, clear guardrails around fairness, and monitoring so it does not steer value only to already profitable segments.

Robotic process automation (RPA)

  • This is rules-based automation, not “intelligence,” but it is often bundled in AI discussions.
  • Examples: moving data between systems, generating fixed-format reports, reconciling the same fields every day.
  • What it needs: stable screens and business rules, good documentation, and simple exception paths.

AI agents and agentic AI

  • This is what I prefer to call emerging Gen 2 automation. Instead of just following a script, an agent is given a goal and can break that goal into steps, call other tools, and adjust when it hits something unexpected.
  • Examples: handling a multi-step onboarding process end to end, preparing a hardship recommendation by pulling data from several systems, coordinating tasks across several internal tools.
  • What it needs: clearly defined goals, limits on what it is allowed to change, strong monitoring, and people who can intervene when something looks off.

You will see many more labels, but most of them are variations or combinations of these families.

This is my opinion, but I like comparing RPA As Gen 1, Agents As Gen 2.

  • RPA is Gen 1. It is a very fast clerk that follows the same checklist every time.
  • AI plus agents are Gen 2. They start to act more like a junior analyst who understands the goal, can choose tools, and can adapt within boundaries.

As you move from Gen 1 to Gen 2, you gain flexibility and potential value, but you also take on more responsibility for oversight, testing, and governance. This is where frameworks from NIST and expectations from NCUA around AI risk management and documentation become important.

Instead of chasing every new AI type, you can build a simple habit: ask the same core questions every time AI enters the conversation.

  • What business outcome are we trying to improve in lending, member experience, operations, risk, marketing, or finance?
  • Which family of AI does this actually use, and is that appropriate for the outcome and risk level?
  • What data does it rely on, and is our data fit for that purpose, given the potential impact if it is wrong?
  • What happens when it is wrong, who feels that impact, and who remains accountable for the decision?
  • How will we monitor performance and explain this system to our board, examiners, and members if asked?

Leaders who consistently ask these questions usually do not need to know every technical detail. They force clarity from vendors and internal teams and ensure AI is used in a way that matches their maturity, resources, and member obligations.

Which of these AI “families” are you already using without fully naming it, and what is one question from this list you want to start asking every time a new AI proposal lands on your desk?