I get asked almost every day what the future of the analyst role looks like. Will we have jobs a year from now?
My honest answer: yes. But the job is going to look different. And I think how different it is depends a lot on which direction you choose to grow.
I see two roles emerging as increasingly important for analytics professionals as AI continues to reshape how organizations work with data. They’re related, and in smaller organizations they may actually be the same person. But they’re distinct in what they require and where they create the most value.
Before I get to those two roles, though, I want to be direct about the one that’s going away.
The Role We Have to Leave Behind
My first job as a web analyst was at Adobe. Every Monday morning I’d come in, open Omniture Discover, and spend three to four hours pulling updated metrics for the week across a bunch of different KPIs. Then I’d copy-paste everything into an Excel spreadsheet dashboard and send it off to my stakeholders.
This was a completely legitimate and necessary job function at the time. Someone had to do it, and I did it well.
But this is something we can leave behind. Today you can build that dashboard once, use a global chat agent to help set it up inside your analytics tool, and put a dashboard agent on top of it that reads it on a regular cadence: monitoring the health of the business, emailing stakeholders automatically, flagging anomalies the moment they appear, surfacing suggestions for where to dig in. It does all of that at a consistency and scale no individual analyst running a Monday morning ritual ever could.
And that brings me to the harder truth: if your primary job right now is to write SQL and pull data, you are going to be out of a job within the next 12 months. That is not hyperbole. Those are tasks that can already be done by an agent today, and the capability is only improving. The data puller role is being automated. The question is what you’re stepping into instead.
The Data Governor
The first emerging role is one I wrote about recently in more depth: the data governance analyst. If you haven’t read that post, you can do so here, but the short version is this: AI doesn’t fix bad data, it amplifies it. And the analyst who owns data quality and taxonomy is the one who makes AI-driven analytics actually trustworthy.
I’ll give you a quick example I referenced in that post. Early in my career at Google, I was asked what should have been a simple question: how many people came from our paid search ads and signed up for a trial of Google Apps for Business? Simple in theory. In practice, there were multiple Start Trial buttons on the site built by different developers at different times, each with a slightly different event naming: different capitalization, different spacing, different conventions. On top of that, the UTM parameters for paid search were inconsistent across campaigns. Answering the question required manually downloading data and trying to piece together rows that may or may not have captured the right combination.
As analysts, we have the business context to recognize that purchase_complete and complete_purchase are the same event. We know the history. AI doesn’t have that context. It sees two events and treats them as two events, and the answer it returns is built on that fragmented signal.
The data governance role is the person who prevents that problem from happening in the first place, and who works to clean it up when it does. It means owning the event taxonomy, pushing back when tracking ships without a naming convention, building schemas that are defined once and actually used, and treating governance not as a quarterly cleanup project but as an ongoing practice embedded in how the team works.
This role becomes more important as AI becomes more central to how organizations access their data, not less. The quality of the answers an AI system returns is only as good as the quality of the data it’s reading. Someone has to own that quality. That’s the governance analyst.
The Strategist
The second role is where I think the most exciting version of the analyst’s future lives.
The strategist is the person who closes the loop. They’re not doing the mechanical retrieval work. They’re reading the conditions, using AI to accelerate the investigation, and making the judgment calls a model won’t make on its own.
The days of finding an anomaly, a broken link, or a website issue and then filing a Jira ticket, essentially tossing the problem over the fence, are gone. The analyst who will stand out, the strategist, will take it upon themselves to not just surface an issue, but drive a solution.
Becoming the strategist is not purely a mindset problem. The tooling matters too. For a long time, running the full arc of an investigation required stitching together three different platforms: one for analytics, one for session replay, one for experimentation. Google deprecated Optimize years ago and never really replaced it. That fragmentation made end-to-end ownership genuinely hard. Platforms like Amplitude are changing that, pulling analytics, session replay, and experimentation into a single connected workflow. The harder part, once you have the tooling, is developing the judgment to know which signal is worth chasing, which question behind the question actually matters, and when to push back on how a problem has been framed. That takes time and repetition. It is also, notably, the part AI is worst at.
Part of what makes this role newly credible is that AI changes the economics of investigation. The reason analysts historically handed off anomalies was not lack of curiosity. It was that running the full arc, funnel analysis, session review, root cause, experiment design, etc., could take two or more weeks and require coordinating four different people. By the time you had an answer, the moment had passed. AI compresses that timeline dramatically. An analyst who used to surface a finding and move on can now own the entire investigation in an afternoon. That changes what ownership actually means.
Here’s what that actually looks like in practice. Say your dashboard agent surfaces a 15% drop in checkout completion on iOS. You don’t just file a ticket. You pull up the funnel to pinpoint exactly where users are falling off. You use session replay to watch what’s happening across thousands of sessions: validation errors at the zip code field, rage clicks, quiet abandonment. The pattern becomes clear: Canadian users can’t submit their postal codes because the field doesn’t accept alphanumeric formats. Now you have the where and the why. You spin up an experiment to validate the fix, you ship it, and you own the outcome.
That is the full loop. You didn’t just surface a number. You orchestrated the investigation, interpreted the evidence, drove the decision, and measured the result. The AI helped you get there faster. But you called the shots.
That shift, from pulling data on request to owning the question and the answer and everything in between, is the strategist role. It’s more visible, more impactful, and a lot harder to automate.
The One Thing AI Can’t Do
The biggest takeaway here though is this:
AI can crunch every data point. It can find patterns no human team could surface manually. It can synthesize 10,000 user sessions into a single finding in minutes. These are real capabilities, and I’m not dismissing them.
But it cannot make your stakeholders care about the user. It cannot look a product manager in the eye and explain why the drop at step three is a trust problem, not a UI problem. It cannot rally a team around a finding that’s inconvenient for someone’s roadmap.
The strategist is also the person who gets into the room. Product reviews, roadmap conversations, post-mortems. Historically, analysts got handed findings to prepare for those meetings, not seats at the table. Getting into the room, and staying there, requires speaking in outcomes rather than metrics. “Checkout completion dropped 15 percent” means something to an analyst. “We are losing Canadian users at the postal code field and it is costing us roughly 200 signups a week” is a different conversation entirely. AI will not make that translation for you.
That’s the job. It always has been.
The analysts who are going to matter most in this next era aren’t the ones who are best at prompting a model. They’re the ones who built something worth prompting on top of, and who know what to do with the answer once they have it.