Most people who talk about using AI in customer success are either selling something or haven't actually shipped anything real. I've been running customer support for a B2B SaaS company for about four years, and I want to share what genuinely changed things for us, because the early experiments were a mess.

When we first started plugging AI tools into our support workflow, we made the classic mistake of trying to automate too much too fast. We had this idea that we could reduce ticket volume by 60 percent in three months and free up the team to focus on strategic account work. What happened instead was that customers got looped in weird automated conversations, reps got confused about what the AI had already said, and handoffs were a disaster. One enterprise client nearly churned because the AI gave them a technically correct but completely unhelpful answer to a billing question, and no human caught it in time.

Here is what we changed and what actually stuck.

First, we stopped thinking about AI as a replacement for the first touch and started thinking about it as a tool for the boring repeatable layer underneath everything else. The questions that come in fifty times a day, the ones your most experienced rep could answer in their sleep, those are fair game. Password resets, how to export reports, what the cancellation policy is, how to add a new seat. Get that list of your top twenty recurring tickets and build your AI layer around those specifically. Do not try to make it generalist from day one.

Second, we got ruthless about handoff signals. The moment a customer uses words like frustrated, escalate, urgent, cancel, or mentions a specific dollar amount, the system flags it for a human immediately. No exceptions. The AI is allowed to acknowledge the message and say someone will follow up shortly, but it does not attempt to resolve anything beyond that. This alone saved us two near-churns in the first quarter after we implemented it.

Third, and this one took us a while to figure out, we started feeding the AI our actual documentation rather than generic training data. Sounds obvious but we were not doing it at first. Once we connected it to our real help articles, our internal runbooks, and even our onboarding FAQs, the accuracy went from about 60 percent satisfactory to around 85 percent in a few weeks. The tool still gets it wrong sometimes, but now it is wrong in explainable ways rather than random ones.

For tooling specifically, we went through a few iterations. We started with a well-known support platform's built-in AI, which was fine but limited. We eventually moved to a setup where we use a dedicated video tool to create short explainer clips for common issues, which we attach to AI responses for anything procedural. So instead of the AI writing out six steps to configure a webhook, it just sends a sixty-second screen recording. Customers love that. For creating those clips at scale without needing our design team involved every time, we have been using atlabs, which lets us batch-produce short instructional videos from scripts pretty quickly. That is not the centerpiece of our stack, but it plugs a real gap.

For B2C, the calculus is a little different. Volume is higher, questions are simpler, and customers have less patience for anything that feels robotic. The key there is tone calibration. Your AI responses need to sound like a human typed them even when they are templated. Run every AI response through a basic tone check before it goes live. Friendly, direct, no corporate fluff.

For enterprise B2B, the priority is not speed, it is accuracy and escalation clarity. Enterprises will forgive a slower response if it is correct. They will not forgive a fast wrong one.

The honest truth is that AI in customer success is not magic. It is infrastructure. You build it carefully, you instrument it properly, and you keep humans in the loop for anything with real stakes. Do that and it is genuinely useful. Skip any of those steps and you are just creating new problems faster than you were before.

Most support teams think they're AI-fluent.

Zapier's new rubric says otherwise.

They just published an AI fluency framework for support, and it's the clearest bar I've seen for measuring where your team actually stands.

Four levels:

🔴 𝗨𝗻𝗮𝗰𝗰𝗲𝗽𝘁𝗮𝗯𝗹𝗲: "AI is a slightly faster Google, nothing more."

🟡 𝗖𝗮𝗽𝗮𝗯𝗹𝗲: "I use AI to operate at a meaningfully higher level."

🔵 𝗔𝗱𝗼𝗽𝘁𝗶𝘃𝗲: "I orchestrate AI and build systems that elevate how I work."

🟣 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝘃𝗲: "I re-engineer how work happens."

That last one is the bar. Not just faster work. Different work. Categories of tasks that no longer exist or run without human involvement.

Most teams I talk to are somewhere between Unacceptable and Capable. Using AI for drafts and lookups, but not changing how work actually gets done.

And here's the part managers need to hear: a leader who's personally AI-fluent but whose team is still doing things the old way doesn't meet Zapier's bar. You have to prove you've led your team there too.

This is exactly why I just opened a new role focused on internal AI enablement. Not customer-facing AI. Internal. Helping my team level up, build workflows, and move from Capable to Transformative.

More on that this week.

Where does your team fall on this rubric? 👇

I posted something similar a few months ago asking for recommendations on AI chatbots to pair with Zendesk Support. Got some good replies, did a lot of testing, and wanted to share what I actually found because I had trouble finding real-world comparisons when I was searching.
...

My team was spending 40 hours a week on something AI now does in 3. Let me explain 👇

We had a customer service triage process eating one full-time employee's entire week.

Every inquiry: manually categorized, routed, logged. Slow. Error-prone. Expensive.

So we built an AI workflow in 3 days:
→ Incoming messages auto-categorized by type and urgency
→ AI Agents answered 80% of requests within seconds
→ Other 20% escalated to response templates pre-drafted for human approval
→ Human reviews and sends — still in the loop, just 10x faster

Results after 60 days:
→ Response time: 6 hours → 22 minutes
→ Team member now focused on higher-value work
→ Customer satisfaction up 24%
→ AI Agent costs: ~$200/month

This isn't science fiction. This is a basic use-case any operator can build.

The founders who win the next 5 years won't be the ones who "wait and see."

They'll be the ones who implemented while everyone else was reading articles about it saying AI isn't 'ready' yet

What process is eating your team's time right now?

Today, in trawling my archives, I came across a post I shared back in 2019 about attending an AI Summit. At the time, it felt like we were standing on the edge of something big. Fast forward to today… and it’s clear that “something big” has well and truly arrived.

As a Head of Customer Service, I’m seeing firsthand just how quickly AI is evolving, not just in capability, but in how it’s reshaping the way we engage with customers and support our teams.

But here’s the question I keep coming back to:

Is it ready to replace the human touch?
I’m not so sure.

What I'm seeing is AI acting as a powerful enabler, helping teams deliver faster, more informed, and more personalised support. It’s not replacing human connection, it’s strengthening it.

And this isn’t just in the workplace.

And what’s really made me stop and think recently… is how normal this already is for the next generation.

AI isn’t “new” to them. It’s just there.

Research shows the majority of young people are already using AI tools regularly, whether that’s for homework, asking questions, or just exploring out of curiosity. In fact, in some studies, well over half (and in some cases far more) of teens are already engaging with AI in their everyday lives.

I see it at home too.

As a mum to a 12 year old, this isn’t theoretical, it’s reality. The way they instinctively turn to AI to ask questions, sense check ideas, or learn something new feels completely natural to them.

And honestly? I find myself doing the same.

From planning road trips, to organising weekly meals, to even visualising ideas for our new home… AI has quietly become a helpful companion in my everyday life too.

So perhaps the real shift isn’t about replacement, but partnership.
AI isn’t taking away the human element, it’s giving us more space to focus on what matters most: empathy, understanding, and meaningful conversations.

Curious to hear others’ thoughts, where are you seeing AI add the most value in your world?

#ai #customerservice

Following up on my last post. Wanted to share how my post-meeting workflow actually works.

After every customer call, I used to spend time 20-30 minutes on the usual admin stuff: following up, CRM updates, next steps. Now it takes about 2 minutes.

Here is the flow:

Meeting ends → I run a custom slash command in Claude Code and drop in the transcript.

Claude reads the transcript, writes the follow-up email and adds it to my drafts in email. I never automate any customer interaction without being in the loop and approving.

Same transcript updates our CRM Meeting notes, next steps field, and any action items. Hubspot's MCP only has read access so I had Claude Code build a custom integration to write back. Our CTO reviewed and approved everything before connecting it to production.

A snippet of the instruction I use is something like: Here's a transcript from a customer call. Write a follow-up email that recaps what we discussed and lists next steps. Then give me structured CRM notes with: summary, next steps, and action items with owners.

In the instructions, I also gave Claude previous follow-up emails I used as examples.

This did take some iteration and wasn't perfect on day 1 or even week 1. The key was just getting started.

Happy to answer any questions!

Curios to hear how other folks within CS are using Claude etc. It has been a total unlock for me when it comes to renewals, comms, analysis of customers etc. Our CS tool is pretty much being used as a systems of record.

Most support teams deploy AI chatbots as their frontline solution, but customers get stuck in loops unable to resolve issues—leading to frustration and churn before reaching a human agent.

I've seen support operations where AI handles 80% of interactions but creates more tickets than it resolves. Customers repeat themselves to bots, escalate in frustration, and agents inherit conversations with zero context. The technology isn't the problem—the workflow is.

That's not an AI problem. It's a customer support systems problem.

✅ Use AI behind the scenes to prepare agents with summaries and context
✅ Route complex or emotional issues directly to people who can resolve them
✅ Eliminate customer frustration from bot loops and repetition
✅ Reduce agent ramp-up time with pre-loaded conversation intelligence

Tools like Zendesk, Intercom, Freshdesk, Gorgias, Drift, and Twilio can support this, but the outcome depends on workflow design, not chatbot sophistication.

A support system should reduce customer effort, not multiply it.

If you lead a support team, be honest: Are your AI tools making customer resolution faster—or are they creating more work for your agents?

#CustomerSupport #SupportOps #CX #CustomerExperience #SupportSystems #RevOps #OperationsLeadership

My AI adoption Journey:

Today I had the opportunity to map AI implementation end-to-end, which was incredibly encouraging.

In my early Salesforce days, I remember creating traditional workflows like Email-to-Case. Now, layering AI on top of that process felt surprisingly straightforward.

Here’s a simple AI-enabled customer support workflow I mapped:

1. Customer sends a message (chat, email, WhatsApp)
2. Conversation captured in CRM (like Email-to-Case)
3. AI processes it: detects intent, summarizes, suggests responses
4. Agent reviews and sends the response
5. AI learns from resolved cases to improve over time

For anyone learning AI, here are some suggestions:
Start small, pick the automations you already know.
Consider how to add an AI layer to make workflows smarter and easier.
Experiment in a sandbox before scaling. :)

The next step for me is prototyping this flow in my sandbox.

I am curious to hear about the practical AI implementations you have seen work well in support teams. ?

I’m not an expert by any means. I’m simply sharing what I’m learning along the way in case it helps someone else who’s also exploring this space:)

I’ve been messing around with OpenClaw for a little over a month (open-source framework for running a personal AI assistant).

I’m not a Customer Success manager, but I’ve worked closely with CS teams. The recurring pain is always context switching: Slack + helpdesk + docs + meeting notes + CRM-ish stuff. You waste time reconstructing the story.

So I tested OpenClaw specifically for that. The value isn’t any single “wow” skill. It’s having one assistant connected to everything, with memory, so you can ask one question and get a usable summary.

The skills that made the biggest difference for CS-style workflows were:
- Slack (internal escalations + stakeholder updates)
- Intercom / Front (thread context + reply drafting)
- Notion or Confluence (playbooks + account notes)
- HubSpot (account context)

Plus guardrails like ClawDefender / Skill Audit if you’re pulling real customer data.

Examples:
- “Anything urgent since yesterday?”
- “Summarize where we are on issue X, customer-safe update + internal next steps”
- “Draft the weekly customer update with owners and dates”

If anyone’s curious, happy to share the exact skill list + how I deployed OpenClaw for our CS team.