Most companies make people work through layers of help before they reach anyone who can fix the problem, so those with the hardest problems try the hardest. Job Thomas, Automattic’s interim Chief Customer Experience Officer, says AI didn’t create that pattern, it made it easier to see, and that a deflection rate is “a measure of how good we are at not talking to customers.” I run response reviews for the WordPress.com support bot. What I check isn’t the number, it’s whether someone got the right help quickly.
-
+
+
+
+
+
+
-
+
+
+
+
+
+
Most people have been stuck with a support bot that won’t let them through. I work on the opposite problem: teaching ours to notice when it should step aside.
I wrote a post on the WordPress.com blog about how it makes that call. Here’s the short version.
Before the assistant answers anything, it runs a few checks. Have you already asked for a person? Have you asked this same question before? Have you contacted support about it recently? Is this the kind of topic a bot shouldn’t handle on its own? If the answer to any of those is yes, it starts moving you toward a human instead of trying again.
It also offers a person without being asked. That happens when a problem needs someone to look into an account properly, when the assistant isn’t confident in its own answer, or when what it already suggested didn’t work.
And if you ask for a person outright, it doesn’t argue with you. It might ask one or two quick questions in case it can fix the problem on the spot, but the offer to hand over stays on the table. You choose.
The part I care about most is what happens after that. When it hands you over, the whole conversation goes with you — what you asked, what it tried, what didn’t work. Nobody has to explain their problem twice.
The number that surprised me: about one in four conversations ends up with a person, and in about half of those, the assistant is the one who suggested it. It offers first more often than people expect.
That’s the goal, really. Solve what it can. When it can’t, get out of the way quickly.
-
> grep –lessons-learned ./experience/*
It’s been about nine months since I took on the AI Integration Specialist role, and I wanted to share some of the things I’ve learned along the way. These aren’t polished frameworks, they’re real observations from doing the work every day.
Accuracy is trust
When an AI bot gives a wrong answer, it doesn’t just fail to help, it actively erodes trust. Users who receive incorrect information from our Bots are less likely to engage with the bot next time. That’s why we treat accuracy as the single most important metric. Getting it right matters more than responding fast.
Review everything, assume nothing
One of the biggest shifts in my thinking has been around evaluation. You can’t improve what you don’t measure, and you can’t measure what you don’t look at. Our regular feedback reviews are the foundation of everything we do. Every conversation the bot has is an opportunity to learn.
Cross-team alignment is non-negotiable
AI work touches almost every part of our support workflow. If the AI Squad operated in isolation, we’d quickly drift out of sync with what the business actually needs. Regular check-ins and close collaboration with various stakeholders keep us grounded and focused.
The human side of AI leadership
Leading a squad isn’t just about data and strategy. It’s about people. Providing constructive feedback, supporting professional growth, and making sure everyone on the team feels valued and heard, that’s where the real impact happens. A motivated, well-supported team does better work, and that shows up in the bot’s performance.
Still learning
Nine months in, I’m more convinced than ever that the intersection of AI and customer experience is where some of the most meaningful work is happening. I don’t have all the answers, but I’m learning something new every day. And that’s exactly where I want to be.
-
+
+
+
+
+
+
> forecast –ai-support-2026
After nearly a year of working full-time on AI in customer support at WordPress.com, I’ve been thinking a lot about where this is all heading. Not in a hype-cycle, buzzword kind of way, but based on what I’m actually seeing in our data, our conversations, and our day-to-day work.
AI is getting better at knowing what it doesn’t know
One of the most important developments I’ve seen is AI getting better at recognizing its own limitations. Early on, bots would confidently give wrong answers. Now, the best systems know when to say “I’m not sure about this, let me connect you with a human.” That self-awareness is a game-changer for user trust.
The human-AI handoff is the new frontier
The moment a conversation moves from bot to human support is critical. Do it too early, and you lose the efficiency benefits of AI. Do it too late, and the user is frustrated. Getting that transition smooth and seamless is one of the biggest areas of focus for teams like ours, and I expect it to be a major differentiator for support organizations in the coming years.
Quality evaluation needs to evolve
How we evaluate AI support quality today is still largely manual. Our AI squad reviews conversations, scores them against criteria, and identifies patterns. That’s valuable, but as volume scales, we’ll need smarter ways to evaluate at scale. Think automated quality scoring, anomaly detection, and real-time accuracy monitoring. The teams that figure this out first will have a serious advantage.
AI doesn’t replace empathy
No matter how good AI gets, customer support will always need a human heart. AI can handle the “what” and the “how” but the “I understand how frustrating this must be” still needs to come from a real person. The future isn’t AI vs. humans. It’s AI and humans working together, each doing what they do best.
What I’m excited about
I’m genuinely optimistic about where we’re heading at WordPress.com. The work we’re doing with our Bots, the investments in AI quality, and the cross-team collaboration across divisions and teams, it all points toward a support experience that’s faster, smarter, and more caring. And I’m proud to be part of building that future.
You must be logged in to post a comment.