Delegation, Context, and the Elevator Test

Dan Ahrens likes to say he was doing customer success before it was called customer success. Twenty-five years in technology gave him a career that refused to sit in one box: renewals, presales technical advisory, support, project management, even a stretch as the director of an IT organization, seeing the vendor relationship from the buyer’s side of the table. It wasn’t until he moved into enterprise B2B SaaS around 2012 that the whole messy collection of things he’d been doing finally had a name. He spent five and a half years at Gainsight to get close to the epicenter of the discipline, ran the digital customer experience team at Atlassian, led CS at Forcepoint, and eventually ran entire customer success organizations. Today he advises and builds independently.

That range is the whole point. Because Dan has personally done nearly every job AI is now being asked to do, he doesn’t treat it as magic or as a threat. He treats it like a new hire — a very capable one with almost no life experience — and the questions he asks about it are the same ones he’d ask before delegating to a person: Does it have the context for this task? Have I tested its judgment? Am I actually making someone’s life easier, or just moving the work around?

From Everything to Customer Success

Around 2016, when the CS category finally cohered, Dan went looking for the center of it. Gainsight was the thought-leadership company in the space — the conference, the blogs, the videos everyone pointed to — so he went to be there. What followed were roles at increasing scale, and scale is where his instinct for automation was forged.

At Atlassian, his digital customer experience team was responsible for every customer without a named CSM — roughly 245,000 of Atlassian’s 250,000 customers, about 98.5% of the paid base, served almost entirely through low-touch email and in-app experiences. At Forcepoint, the book was over 13,000 global customers and $400 million in ARR. No human manages a book like that by hand, so the only real question was how to automate without making the experience worse.

His answer is an analogy he keeps coming back to: the elevator.

“Most people enjoy riding an elevator as opposed to taking the stairs — it makes your life easier. But what about an elevator that stopped literally at every floor on the way up and every floor on the way down? You may then choose to take the stairs.” — Dan Ahrens, Customer Success Leader & Advisor

That’s the trap of digital CS, and the test he applies to every automation: is it the elevator that gets you where you’re going, or the one that stops at every floor, adding extra steps that don’t provide true value?

What’s Actually Working

Dan’s most useful AI work has been aimed at the tasks that carry real human judgment — not the ones that are obviously mechanical. His favorite example is a renewal-review process he used to run by hand.

At one company, a Gainsight automation created a call-to-action whenever a customer’s renewal hit 120 days out. That part worked. What didn’t scale was the exception-catching: once a month Dan would personally read through CSM notes, checking whether the status fields and open text updates reflected reality and whether anything was quietly slipping through the cracks before it reached executive reporting. Valuable, and a couple of hours of manual work every time.

So he built a custom Claude agent for it. He approached it exactly the way he’d approach handing the work to a manager or lead on his team: here’s the table of renewals, here are examples of the emails I’ve sent, here are the things I look at, here’s the logic. Then he iterated. Sometimes it was overly critical and he had to dial it back; other times it let things slip and he recalibrated.

“Do the same things you would do if you were going to delegate this to a human. Here’s the table, here are some examples, here are the logic rules — and then give me a couple iterations to check its work.” — Dan Ahrens, Customer Success Leader & Advisor

The payoff wasn’t just saved time. Because the analysis no longer cost him a couple of hours, it stopped being monthly. Now it runs every Monday morning and drops a bulleted list of accounts needing attention into the whole team’s inbox — which turned a once-a-month audit into a weekly proactive habit, and closed the window where an at-risk account could go a full month unnoticed.

Underneath the tooling is a conviction about where AI shouldn’t go. Dan is wary of what he calls AI slop — ten people rebuilding the same dashboard in slightly different colors. He points to a line from Gainsight’s CEO that has stuck with him: most B2B software is still, underneath the UI, competing with Microsoft Excel. His current side projects chase exactly that gap — the pain points where the only “solution” today is someone hand-building a spreadsheet that’s stale the moment they export the next CSV from Salesforce.

Don’t Hand a Toddler a Power Saw

The clearest lesson in Dan’s thinking is about the gap between what a system knows and what we assume it knows. He describes overhearing specialists talk shop — real English words that carry no meaning to an outsider — and says that’s exactly the mistake we make with AI: we judge it from inside our own knowledge, assuming it has the same context we have.

His frame for it is disarmingly concrete.

“You wouldn’t give a toddler a power saw — they don’t have the life experience or the context. We give humans tasks commensurate with their training and judgment. We have to be that deliberate about what we hand to an AI agent.” — Dan Ahrens, Customer Success Leader & Advisor

He traces the instinct back to his very first real AI experiment, years before it was a work tool. On a Friday afternoon in late 2022, when ChatGPT was new and the news was that it could write code, he asked it to build the old game Pong. The first version just sent the ball off the screen — he hadn’t told it to play against the computer. His prompting was poor. He fixed the context, and then his nine-year-old daughter wandered in and asked if the ball could be a cat emoji’s head. It could. Two lessons landed at once: the output is only as good as the context you give it, and — a Gainsight value he still quotes — a beginner’s mind asks for things experts wouldn’t think to try.

That’s why he’s cautious about pushing AI straight into customer-facing judgment before it’s earned it. A smart toddler is still a toddler. You give it the training, you give it the context, and you run the testing loops until you trust the response — the same graduation from supervision to autonomy you’d give any employee.

What’s Next

Dan’s forward bet is that the real returns come from augmenting the people already in the seats, not replacing them — he points to the companies that cut too aggressively and had to hire back. The sweet spot he keeps naming is the “low-signal, high-cost” work: checking all your accounts every day for a single LinkedIn job change, say. Almost always nothing — until the one hit that actually matters. Too expensive for a human, perfect for an agent.

He also expects token costs to force a reckoning that he mostly welcomes: fewer people spinning up redundant one-off tools, more organizations thinking deliberately about which internal problems are actually worth an AI solution. Governance and reuse over everyone building their own secret sauce.

But his one piece of advice for post-sales leaders comes back to where he started — the human at the other end of the automation. In sixteen years, the single biggest predictor of a customer’s long-term success he’s seen isn’t the product; it’s the onboarding experience. So the highest-value thing AI can do isn’t to shift homework onto the buyer under the banner of “automated onboarding.” It’s to watch where a customer actually is in their journey and give them the right guidance at the right moment — the thing an implementation specialist used to spend hours doing by hand. Make it easy to get up and running, and everything downstream gets easier. That’s the elevator working the way it should.

The Sonora Project captures how CS leaders are navigating AI — the real experiments, practical wins, and hard-won lessons shaping the future of customer success.