Findable first: how a Kiwi financial services firm put AI to work
Seventeen people, five offices across regional New Zealand, and a leadership team that knew AI was going to change how they serve customers. The question wasn’t whether to adopt it. The question was where to start.
They started in the right place, and that’s the whole story.
I call it findable first: before you put AI on top of your business, make your information findable, governed and well described. Get that right and the AI starts earning its keep almost immediately. Skip it and even the best tools underdeliver. Findable first is the difference.
Where do you start with AI in a small business?
Not with the AI, as it turns out. We started by understanding the business.
Before any technology moved, we sat down with the leadership team and worked through how the firm actually operates: where the hours go, which jobs are repetitive, where customers end up waiting, and which improvements would genuinely move the needle. Out of that came a short list of targeted use cases, ranked by the difference they’d make. Not AI everywhere. AI where it counts.
Discovery also made one thing clear: the foundations had to come first. The firm’s leadership wanted better, faster outcomes for their customers and less time lost to mundane work, and none of the use cases would deliver that sitting on top of ungoverned data. So rather than buying licences and hoping, the directors committed to doing the groundwork properly: a governed document platform, a permissions model designed for a regulated industry, and a team trained to use what was built.
That decision took discipline. The groundwork phase is the part nobody puts in a brochure. It’s also the part that determines whether everything after it works.
The groundwork: governance before AI
The first months were deliberate, structured work. Security groups and permissions designed up front. More than 1,700 client folders moved to a properly governed document platform, cut over in a single overnight switch so the team simply arrived Monday morning to find everything in its new home. Sensitive material placed on a need-to-know footing, exactly as a regulated industry expects.
In financial services, where identity documents and personal financial information are part of daily work, this isn’t box-ticking. It’s the foundation of customer trust, and it’s what lets you say yes to AI with confidence.
Then the clever part: teaching the files to describe themselves.
Making thousands of documents findable
We used AI to read the content of every document and fill in a set of metadata labels: what type of document it is, who it belongs to, the dates that matter, when an identity document expires. Ten labels in all.
The build was methodical. The firm’s internal lead took it one label at a time, testing each against real documents and refining until the extraction was consistent, then moving to the next. Slow is smooth, smooth is fast.
Six weeks in, the firm’s director sent me this: “We have our 10 categories now. Very, very happy with the data it’s extracting with our labels.”
Ten labels doesn’t sound like much. What it means is that thousands of documents that once depended on institutional memory can now be found, filtered and questioned in seconds by anyone with permission to see them. The document library stopped being a filing cabinet and started being the firm’s brain.
What did the firm actually gain?
Once the groundwork was in, the wins came quickly, and they came from the team rather than from me. That’s the point.
The director set up an automated routine that files his email. It has quietly worked through more than 10,000 messages and, in his words, “just keeps chugging along.” Hours of administrative work every week that simply no longer exists.
Next on the list is handing document signing workflows to an agent, taking another piece of repetitive admin off the team’s plate. As one of the directors put it: “It should make that so much quicker.”
The team has been through Copilot training, delivered online so all five offices could join the same session, and recorded so it keeps teaching long after the session ended. We built the content around the work the team actually does every day, their inboxes, their documents, their meetings, rather than a tour of features. The result is that the enthusiasm isn’t confined to the technically minded few. It’s spreading across the team, and that’s the adoption signal I watch for.
And every hour saved goes back where it belongs: into customer conversations, faster turnaround, and advice rather than admin.
The Cairn approach: people, process, technology. In that order.
This engagement is the Cairn approach working the way it’s meant to.
People first. The directors led from the front and used the tools themselves. An internal champion owned the build, so the capability lives inside the firm rather than inside their consultant. Training was a team commitment, not an optional extra.
Process second. Governance before migration. Permissions before AI. A clear position on what stays inside the controlled environment.
Technology last, and by then it was almost the easy part. When Copilot and the knowledge agent arrived, the data was governed, findable and described. The tools landed on prepared ground.
For any business wondering what this looks like in practice, the order of operations was:
Leadership commits and leads from the front.
Understand the business: map where the time goes and pick the use cases where AI will make a real difference.
Govern the data: platform, permissions, security, compliance position.
Make information findable: migrate, structure, and describe documents with AI-filled metadata.
Train the whole team, and aim the training at your strongest users.
Put AI to work on the mundane: filing, signing, searching, drafting.
Reinvest the saved hours in customers.
What’s next
There’s a next chapter coming. The biggest use case identified at discovery, proactive customer contact driven by the firm’s own data, was deliberately sequenced for when the firm moves to its new core customer system later this year, one that opens up proper integration. Because the foundations are already in place, that build starts from halfway up the hill: automated document handling, workflow connections, and saved hours turned into even better service. Holding the highest-value work until the platform was right was a disciplined call, and the right one.
Most firms buy the AI first and sort the data out later. This firm flipped the order, and the results followed.
Copilot didn’t make their files smart. The groundwork did.
If you’re working out where to start with AI in your business, give me a bell.