待翻譯：The Future of Data Analytics: Why AI is rewriting the Analyst’s Job Description

AI 服務暫時不可用，以下為來源正文，待恢復後補全翻譯。

The Future of Data Analytics: Why AI is rewriting the Analyst’s Job Description | Databricks Blog Skip to main content We are already seeing that AI is automating time-consuming technical tasks like SQL and dashboarding, driving the ability for business users to more quickly get answers. The analyst’s role is shifting toward irreplaceable human skills: framing problems, applying organizational context, and translating insights into strategic decisions. To capture this value, organizations must redefine hiring and success metrics, moving from technical proficiency to decision influence and question quality. The Data Analyst role has been declared dead more times than we can count. AI will write queries, build dashboards, and generate insights. So why bother hiring analysts at all? Because that argument confuses the task with the job. What’s actually being automated is the work that consumed analysts’ time but never delivered business value: wrangling data, rebuilding dashboards for every new stakeholder, writing ad hoc SQL for one-off requests. The analyst who only does those things is being automated. The analyst who frames business problems and drives decisions is becoming more and more valuable. This isn’t a prediction. It’s already happening. And in this article, we’ll explain why AI isn’t ending the analyst role; it’s restoring it to what it should have been all along. Between us, we’ve hired 50+ analysts, led multi-disciplinary delivery teams, and designed AI-enabled decision frameworks across industries. We compared notes recently, and the same pattern kept surfacing: the analysts who thrive aren’t the ones writing the best SQL. They’re the ones asking the best questions. Platforms like Databricks AI/BI are converging toward analytics driven by natural language. You can describe what you want in plain English and get a dashboard, a metric, or an insight. Text-to-SQL tooling is solving a problem the industry has wrestled with for decades: the time and technical know-how it takes to get from a question to an output. That's a real win. But closing this gap only sharpens the dimension that was always more important – the quality of the insight itself. There has always been, and always will be, a different gap: knowing what question to ask in the first place. The most important step in being a great data analyst is not just answering what is being asked, but refining and shaping to reach an understanding of what really needs to be answered. How many times have you been asked to get the data for a data point, only to discover afterwards that it wasn't really what the business was trying to assess? How We Got Here: The BI Tool Trap This has been a slow, gradual process. Initially, analysts served as liaisons between the business and IT teams that owned Data Warehouses, at times acting more like Business Analysts. But then with the advent of tools like QlikView, Power BI, and Tableau, these business-facing analyst teams were no longer wedded to legacy data warehouses to produce their dashboards. This enabled quick iteration but led to a different set of problems. Analysts became dashboard builders and data fixers, acting at times like shadow-IT. Suddenly, the role no longer required them to be merely great communicators who understood the value of the business's data. Their time was consumed by: Manual data wrangling and pipeline firefighting Rebuilding dashboards for each new stakeholder Writing adhoc SQL for one-off requests Bridging gaps between BI tools and upstream data The skill that organisations actually needed, business problem framing, went underdeveloped. Analysts were busy and stretched, but often struggled to show their value. We’ve all seen countless rounds of restructuring in data teams, as people struggle to understand their value. There's a persistent tension at the heart of the role: technical experts are often terrible at understanding the business, while business-savvy individuals are less likely to be strong technically. Organisations tried to solve this by hiring people who could do both, but such people are rare. We’ve seen this throughout our experiences. Usually, leaders settle on a mix of skills across the team rather than individual skills, ie having some very technical people who aren’t strong at business context and less technical people who are stronger at communication. But usually in smaller teams this leads to these individuals feeling disgruntled in their roles, with both sets of people needing to pick up both types of work. The more business-focused get frustrated by technical challenges and the technical employees getting frustrated at having to speak to the business. A lot of this has led to analytical teams that focus on surfacing data rather than answering questions. The core value proposition of analysis, turning information into decisions, was diluted by the mechanics of getting data from A to B. The result of this is a generation of analysts who can tell you what happened, but not why it matters or what to do about it. AI Simplifies Things Again Every wave of tooling promised to free analysts from the mechanics of delivery. In practice, it buried them deeper. AI changes this. Not by replacing analysts, but by automating the work that buried them. Platforms like AI/BI Dashboards with Genie Code allow users to describe what they want in plain English and get a working, consistently styled dashboard in minutes. Natural language interfaces like Genie One mean stakeholders can ask ad hoc questions directly, without filing a request or waiting in a queue. Metric definitions can be generated and refined automatically, with AI acting as a coach to sharpen them against the actual business need. These are not incremental improvements. They remove entire classes of friction that, over time, distorted the analyst role beyond recognition. The technical tasks have become commodities. We’ve seen this play out in practice. An analyst at a public sector organisation needed to build a customer segmentation model, a previously two-month effort involving SQL development, data prep, and iterative cycles with engineering. Using Genie Code, they built it in half a day. The AI handled the technical execution; the analyst focused on defining the segments that actually mattered to the business. That’s the shift: not fewer analysts, but analysts spending their time on the work that moves the needle. But speed alone is not the value. An AI that answers the wrong question perfectly is still wrong. What this new wave of tooling makes clear is that the bottleneck in analytics has never been SQL or dashboards. It has always been judgment. Someone still needs to: Define what the business should actually be measuring, and why Frame the right question before anyone touches the data Validate that the output makes sense in context Interpret the data points into a recommended course of action AI does not do any of that. That's the analyst's job. It always was, we just never gave them the time to do it. AI simplifies execution. It does not own intent, accountability, or consequence. That responsibility remains firmly human. “With AI augmenting processes and automating routine tasks, human judgment, accountability, and decision ownership remain central” - Capgemini Research Institute, The multi-year AI advantage: Building the enterprise of tomorrow, p.6 And this is where the analyst's role does not disappear, but re-emerges. The Analyst Reborn: Less SQL, More So What With technical tasks automated, the analyst's value shifts to work only humans can do: Problem structuring The most valuable thing an analyst can do is help a stakeholder articulate the right question before anyone touches data. This is harder than it sounds: the Capgemini Research Institute found that "only 33% of leaders can articulate their needs to a Gen AI system." If you need to measure a marketing campaign, measuring click-through rate may sound great, and you could ask the AI to measure this for you and believe you have a successful campaign. If you developed the campaign, you may not need to question this further. But you may not have asked the right question. What if the email was offering a 90% discount? Well then, you would need to look at the total revenue picture, too. The analyst is needed to help frame the problem and play devil's advocate, and without needing to know SQL, the analyst can focus on understanding the business and acting as a translator. Context and judgement A number without context is just a number. An analyst who knows the business can tell you that a 5% drop in retention is alarming, but a 5% drop in a metric whose definition you just changed is meaningless. AI doesn't have that contextual awareness. A good analyst does. This is what domain knowledge means in practice. Not just familiarity with the industry, but knowing that last quarter's revenue spike was a one-off promotion. Or that a key account migration is distorting the data. Or that a particular field has been manually corrected in the CRM for years because the upstream system was never fixed. AI has no memory of any of this. As AI scales the volume and speed of insight generation, this gap does not close; it widens. The more outputs AI produces, the more you need someone who can hold each one against institutional reality and ask whether it actually makes sense. Human interpretation is the remediation of AI’s two biggest flaws: Hallucinations, often wrapped up in excessive optimism and confidence in the way that AI answers. Probabilistic Nature - which can impact the reliability of AI. The analyst with deep domain knowledge is the one who looks at a perfectly formatted dashboard and says: "This number is wrong, and I know exactly why." That capability is not a prompt. It is built on years of familiarity with the business. Storytelling and influence Data doesn't drive decisions; stories do. The analyst's job isn't to present a table of numbers. It's to walk into a room and say, "Here's what's happening, here's why, and here's what I think we should do." Research shows that combining qualitative and quantitative data is always more persuasive than numbers alone; people remember stories, not statistics (Heath & Heath, Made to Stick). Getting to the right business decision requires strong persuasion skills and an understanding of the audience (HBR: Data Science and the art of persuasion). No LLM is replacing that in a boardroom. Governance and trust As AI generates more insights, someone needs to ensure they're grounded, auditable, and correct. Alongside a good cataloging tool such as Unity Catalog, the analyst becomes the quality layer, curating AI outputs, validating them against domain knowledge, and flagging when the model is hallucinating or the data is wrong. The latter is a very common problem. A lot of data isn't pure. Manual data entry, new lookup values, out-of-system events. These all introduce noise that AI isn't aware of. The strong analyst knows this and treats the data accordingly. Orchestration The modern analyst doesn't write the SQL; they direct the AI agents that do. They curate the Genie spaces, define the right metrics, structure the knowledge bases, and design the analytical workflows. Think of it as moving from player to coach, which is what they should have been all along. What This Means for Organisations The implications are practical and immediate: Hiring: Stop screening analysts for SQL proficiency. Start screening for curiosity, business acumen, and communication skills. Technical skills are no longer the be-all and end-all. Consider training your existing analysts on critical thinking skills. Time to Value: With the advent of AI, organisations should see improvements in time to value. However, they need to be aware that this will come with a risk of misinterpretation due to hallucinations or missing cont [truncated for AI cost control]