When I think back on the support-efficiency project, what actually stands out isn’t the “AI” part at all; it’s the realization that many operational problems look technical on the surface but are, at their core, questions about workflow, attention, and emotional bandwidth. Our North America support function was essentially run by one person who carried everything from Trustpilot reviews to product questions to tickets that really belonged in sales operations or accounting. None of this was unusual for a lean fintech team, but it did create a day-to-day environment where urgent work mixed with repetitive work, and important issues often had to wait behind tasks that simply consumed time without adding any real value.
I started paying closer attention when I realized how often our support rep was pulled into these loops—opening Trustpilot, crafting well-phrased responses, double-checking tone, responding to feedback at odd hours—and how little visibility the rest of the organization had into that workload. It wasn’t that the work was complex; it was that it accumulated quietly, and because every response required context and careful phrasing, it drained more mental energy than anyone realized.
That was when I offered to help. Not because I wanted to “bring AI into support,” but because I already had the automation framework in place from my other projects, and I could see exactly where small interventions would save hours without requiring the team to change how they worked. So I began with the simplest part: reviews. Every new Trustpilot review triggered an internal email, which meant I had a stable signal I could hook into. I built a workflow that intercepted the email, captured the review, sent it into a dedicated Slack channel, and paired it with two AI-generated draft responses—one slightly more formal, one more conversational—so that the rep could simply choose one, make a few edits, and post.
The point wasn’t to automate judgment; it was to remove the mental tax of staring at a blank reply box ten times a day. And almost immediately, the feedback loop improved. The rep responded faster, management could see what was happening without asking for updates, and our online presence became more consistent—not because anyone suddenly had more time, but because the repetitive part of the job shrank.
Once that piece was working smoothly, I extended the same logic to other review surfaces like G2 and Capterra, and eventually to a portion of our inbound support tickets. Tickets were interesting because many of them were not actually solvable by support alone; some required sales operations, others needed accounting or product input, and before, the rep had to manually route each one by guessing which team would handle it. I added a light AI layer that performed a first-pass interpretation of each ticket—just enough to determine whether it should stay with support or be directed immediately to another team. It wasn’t a deep classification system, but it removed the back-and-forth that wasted time and frustrated everyone involved.
The more I worked on this, the more I kept coming back to the same observation: effectiveness rarely comes from sophisticated architecture; it comes from understanding the shape of a team’s actual work. Automation only helps when it is molded around real habits, real pain points, and real constraints. In this case, AI was simply a quiet supporting actor inside a workflow that already existed—it helped soften the edges of a repetitive job, gave management a clearer window into activity, and prevented issues from sitting idle while someone tried to figure out who they belonged to.
If there was a “lesson learned,” it was that the hardest part is not building the automation; the hardest part is asking the right questions. What exactly slows people down? What patterns repeat daily? Which steps are meaningful, and which are just muscle memory? Once you understand that, the solution becomes almost obvious. You use automation to clear the noise, and you reserve human effort for the parts where human judgment actually matters.
Looking back, the project wasn’t large in scope, but it had an outsized impact because it addressed something everyone felt but nobody had time to articulate. Support worked more smoothly, the rep felt less buried, our online presence became more consistent, and teams that rarely interacted with each other suddenly had a shared rhythm. It was a small example of how operational empathy, paired with the right tools, can quietly transform a workflow without announcing itself as “a big initiative.”
And honestly, that’s the kind of work I enjoy the most.