Most credit unions are already using AI somewhere in their marketing, just not for the things that would move the needle most. Depending on the organization's size and technical maturity, it's usually limited to meeting note summaries, first drafts of emails, or basic member analytics. Those are all useful - but none of them touch strategy.
The next step is using AI to make better strategic decisions, not just faster tactical ones.
Every credit union sits on data that is disconnected by default. You have member survey data from vendors like Raddon. Competitive intelligence on the banks and credit unions across the street. Performance benchmarking from Callahan. Your own product roadmap and goals. Individually, each of these tells you something. Cross-referenced against one another, they tell you something much more useful.
Doing that cross-referencing by hand is genuinely time-consuming and error-prone. Too many strategic decisions at credit unions get made using a single data point sitting in its own silo while several other details that could change the outcome are sitting nearby, unconnected. The need isn't more data. It's a way to incorporate the data you already have, all at once, without it taking days.
I first built a version of this at a previous credit union using ChatGPT Projects. At the Pacific Northwest credit union where I now run marketing, I've rebuilt and expanded that system using Claude and Perplexity.
Claude Projects lets you load a set of reference files that the model treats as persistent context for conversations inside that project. I used it to build a working knowledge base of the organization:
- Recent member survey data from Raddon
- A voice-recorded walkthrough of how our teams are structured
- Existing strategic planning documents, including our long-range plan and ideal-member personas
- Every current and legacy product, along with notes on what actually differentiates each one - including where we might be wrong
- Our purpose, mission, vision, and values
- A competitive analysis created using the competitive intelligence system I built, which monitors competitors' social media, content, and websites
- That same competitor list run through Callahan to see both how competitors present themselves and how they're actually performing
- Notable partnerships and events, plus context on our SEG makeup and examples of specific Select Employee Groups
- Product brochures and collateral
One deliberate exclusion was member data. No names, no account information, and nothing that could identify an individual member. I used only aggregated, anonymous survey responses.
Before loading anything, I confirmed with our data governance team that the information I was uploading - which was mostly public or already aggregated - was appropriate to expose. I also made sure the setting that prevents our data from being used for model training was turned on. Even with that setting, there's still some residual risk of exposure. Treat anything sensitive as off-limits until you have a better answer than "I turned off a toggle."
Before trusting the system with anything strategic, I tested it. I asked basic questions about how our checking product worked or what we knew about a particular SEG, specifically to catch anything missing or wrong in what I had loaded. That validation step matters. Don't skip it.
Once I trusted what it knew, I used it for real planning at the start of the year. I asked it to draft a product marketing plan. Because it already understood our products, how they compared with competitors, and what members thought of them, the first draft came back informed instead of generic.
A few rounds of back-and-forth got it to something actionable. "Change this. Don't touch that. We can't fix this particular pain point right now." The whole cycle took maybe 20 to 30 minutes for work that would otherwise have taken hours. Loading the base files takes about ten minutes if you already have the documents on hand. After that, each new plan or analysis takes another 20 to 30 minute iteration. Multiply that across a year of planning cycles and ad hoc requests, and the hours saved add up quickly.
This isn't limited to formal planning. My team covers business development, community engagement, and brand communications, and other departments regularly bring us requests or ideas. I can validate those against the model instead of pulling together a team discussion from scratch every time.
Giving the tool a simple internal name helped. People could refer to it by name instead of calling it "the AI thing," which made it easier for leadership and my direct reports to engage with it as something real rather than an abstract project. I also tend to talk through requests out loud and dictate them instead of typing, which makes it faster to provide context.
The standing knowledge base is not the whole story. When we shifted card networks from Mastercard to Visa this year, I pulled PDFs of articles from Credit Union Times and BrandForum covering card-transition best practices and added them to that specific conversation. The base files gave the model institutional context - the additional documents gave it situational context. That combination, a stable core plus topic-specific additions when needed, is where the system starts to pay off.
I use Claude and Perplexity for different reasons. Claude Projects holds the core knowledge base and handles the heavier strategic reasoning. As the amount of knowledge grows, Claude can automatically shift into retrieval mode so the project can handle more material without blocking uploads. For comparison, ChatGPT Projects currently caps the number of files based on the plan - Plus supports 25 files per project, while Business and other paid plans support 40.
Perplexity fills a different role. Its Projects feature lets you add web links and domains as context, so I included our website along with industry sources like TheFinancialBrand.com and Credit Union Times. I've also used Perplexity's deeper research and file-creation features to turn a prompt into a finished document, slide deck, dashboard, or spreadsheet.
Moving this work into a shared space also allowed my team to work alongside me instead of routing every request through me. Watching someone on the team have their own "wait, that was fast" or "I hadn't thought of that" moment did more for internal AI adoption than any training session could have.
None of this works without discipline:
- No sensitive or member-identifiable data, regardless of which training-data settings you have turned off.
- Keep a human in the loop. Do not take the first output at face value - review it, push back, and revise it.
- Change your language, not just your tools. We don't use AI here. We work with it. Treating it like Outlook or Excel limits what you get from it. Treating it like a collaborator changes the output.
- Train your team on prompting. There's plenty of useful material available, and you can use AI itself to help draft better prompts.
Until recently, the answer to "Why not just use Copilot?" was easy. Copilot had no persistent workspace, so it couldn't hold a knowledge base. That changed in 2026 when Microsoft made Copilot Notebooks broadly available, providing a persistent workspace for reference materials. Copilot Memory can also carry selected context between chats.
The gap is narrower than it was when I first built this. However, Notebooks provides a workspace - not necessarily a reasoning upgrade. Availability and features also vary by license and tenant. Check what your organization actually has before assuming it's available. If Copilot Notebooks is available in your tenant and your data governance answer for Claude is no, start there instead of not starting.
The setup in this article is a pattern, not a product endorsement.
This is still early. The next step is a properly isolated environment where we can safely use member data itself rather than relying only on aggregated survey responses. That will probably involve a dedicated vendor rather than a general-purpose consumer AI product. We're also having conversations as a leadership team about data governance and a formal AI policy. A CRM or customer data platform is somewhere on our horizon, which could open another level of this work.
None of that is required to start.
You need data that's clean enough to use, a clear sense of what you're trying to decide, and a willingness to build the knowledge base once so you can reuse it. If your organization is not yet using AI for strategic decisions, find someone internally who is willing to dig in, or bring in someone from outside who works in this space.
The time savings are real, but the bigger win is the quality of the decisions you can make when you're cross-referencing the information you already have instead of guessing from one number in isolation.