How I'm Actually Using AI in Customer Success

There’s a moment most customer success managers run into every day, usually around 6pm. You’ve just come off back-to-back conversations. You have a queue of recaps you owe customers, action items from previous days you still haven’t sent, and project plans that would actually be useful if they were up to date. But there are 24 hours in a day, eight of them belong to sleep, and the math doesn’t work. So what do you do? You push through the action items that are critical for that day. You leave the rest as a task for the next day, or whichever day you finally get free time to chip into the administrative burden.

This isn’t a time management problem. This is a structural one. CSMs, especially the ones trying to be high-performers within their organizations, are being asked to do a lot more than the job historically required. CS touches everything. And touching everything takes time.

This is where AI comes into play.

Not a replacement. An assistant.

The framing that’s worked for me is to think about AI the same way I’d think about hiring a personal assistant. If you’ve ever written a job description, you know the exercise. You define what they own. You define what they don’t touch. You define the standards you expect. You define what gets escalated back to you. That’s the same work you do with AI. You’re not just turning a tool on. You’re hiring something to work alongside you, and the quality of the output depends on how clearly you’ve defined the role.

What this looks like in practice

A few of the things I’m actually doing, day to day.

Meeting recaps. I use Claude to pull from my meeting transcripts (via Granola) and generate recaps in my tone (via Skill files). The same source material gets shaped into whatever the customer lives on, a Slack message, an email, a Teams message. The point is that the recap goes out the same day, while the context is still warm and the customer hasn’t moved on. The 30-minute “go back, re-watch the Gong, write the recap” routine is gone.

Executive mapping for expansion. I use Clay in Claude to source potential executives I should be expanding into but haven’t engaged with yet. The same workflow reviews my previous meetings to flag whether a given executive has actually shown up in a conversation, and then highlights opportunities within the account at director and above, inside my target personas, where I don’t have anyone actively engaged. Instead of me building that picture from scratch every time, the picture gets handed to me and I decide what to do with it.

Drafting outreach in my voice. This is the one most people give up on too early. The default complaint about AI-generated copy is that it sounds like AI. That’s true, but it’s also a solvable problem. The reason it sounds like AI is that nobody has taught it to sound like them. I’ve invested in tone skills that capture how I greet people, how I structure a proposal, the words I never use, the rhythm of how I close. The result is drafts that don’t carry a heavy burden to edit. I’m not rewriting the message. I’m reviewing it, adjusting one or two things, and sending. The investment up front pays back every single week. If you walk away from AI because the first output sounded generic, that’s on the input you gave it, not on the tool.

Pre-meeting briefs. Every morning I get a brief pulled together for each team I’m meeting with that day. Previous conversations, recent calls, open threads, the status of the project plan. It tells me what we discussed last, what’s still outstanding, and what I should be walking into the call to accomplish. Instead of toggling between five tools in the ten minutes before a meeting, I show up ready.

None of these are AI doing the customer relationship. All of them are AI clearing the runway so I can do it.

What I’ve actually reclaimed

The honest measure of whether AI is working for you is what you’re now spending time on that you weren’t before.

For me, the answer is the part of the job that actually moves the needle. More time on account planning, the real version, where I sit with my portfolio and think about which accounts have expansion ceiling, where the risk is, and where my hours should be going over the next 90 days. More time on commercial execution for renewals and expansions, which is where the strategic upside of the role lives and the first thing that gets crowded out when administrative load runs hot. More time being present in the actual meeting, because when you know the recap is going to write itself, you stop frantically taking notes and you start listening.

The work I’m doing more of is the human work. That’s the point.

What I’m piloting next

The pilot I’m most curious about is using Notion AI agents to update project plans after a call. The flow is straightforward. The meeting happens, the note taker captures the transcript, and an agent reads the transcript against the existing project plan in Notion and proposes the updates. New action items added. Completed items checked off. Status notes attached to the right workstream. I still review before anything publishes, but the gap between “we discussed it on the call” and “it’s reflected in the plan” closes from days to minutes. If this works the way I think it will, the project plan stops being something I maintain and starts being something that maintains itself between my reviews.

The work that stays

The market is asking more of CSMs than it ever has. The accounts are larger, the expectations are higher, the technical sophistication required is real, and the commercial accountability is rising. None of that is going away.

What AI gives you is not a way out of that work. It’s a way to keep up with it without losing the parts of the job you actually got into the role for. The administrative tax used to fall on you personally. A lot of it doesn’t have to anymore. The hours you reclaim get to go back into the human side of customer success. Building the relationship. Reading the room. Knowing when to push and when to listen. Earning the renewal. Earning the expansion. Being the person the customer remembers when something works, and the person they call when something breaks.

That part of the job was never the thing AI was going to take from you. It’s the thing AI exists to give you more time for.

Treat it as your assistant. Not your competitor. Not your replacement. Just the help you actually needed.