The data analyst's honest guide to working with AI

The data request that broke the data analyst

When we spoke to Tyler Goulet from Aisle he told us he spends his days looking for stories in data. As the customer success lead, he needs to know which brands are seeing the best results, which patterns they can turn into best practices, and which angles make compelling case studies. Before his team adopted better tools, every one of these questions meant looping in the engineering team and waiting.

Here is the reality for most data analysts today. The 43rd data request of the week arrives. You know it will take two hours to pull together: understanding which tables track what events, writing SQL, debugging the query, optimizing for performance, and finally formatting results for stakeholders. Then someone asks a follow-up question and the cycle starts again. 

Everyone's talking about AI solving your data problems. But for someone evaluating a new AI-powered data analysis tool, separating the hype from reality is becoming a challenge. 

The AI promise vs. the AI reality check

Picture a typical Tuesday for Sarah, a data analyst at a mid-sized tech company. She starts her morning with what should be a simple question from the marketing team: "What was our conversion rate last quarter by channel?"

First, she logs into the data warehouse to check which tables have the data she needs. Then she switches to the BI tool to see if someone already built this view. They haven't. She opens her SQL editor to write a query, but realizes she needs to validate the date ranges first, so she's back in the data catalog. After finally running the query, she needs to export the data, format it in a spreadsheet, and create a chart in yet another tool. By lunch, she's switched between six different platforms and spent three hours answering a question that should have taken minutes.

This isn't a one-off bad day. It's every day. Sarah's calendar shows back-to-back meetings where stakeholders ask questions she knows she could answer quickly if she didn't have to navigate this maze of disconnected tools. The actual analysis, the insight generation that drew her to this work, happens in the gaps between all the tool-wrangling and data prep. Most weeks, she gets maybe one afternoon of genuine analytical thinking.

The tools meant to help aren't helping. Each platform promises to make data more accessible, but adding more tools just means more logins, more context to remember, and more places where data might be slightly out of sync. Sarah finds herself explaining to stakeholders why the numbers in Tool A don't match Tool B, when she should be explaining what the numbers mean for the business.

Analysts like Sarah aren't asking to be replaced by AI. They're asking to stop doing work that doesn't require their expertise. They want to move quickly without breaking governance rules. They want tools that show their work rather than operate as black-box magic that spits out answers with no explanation. Here's where most AI business analytics solutions fall short. 

Sarah and her colleagues have already taken matters into their own hands. During a recent team happy hour, someone admitted they'd been pasting company data into ChatGPT to get quick analyses. The confession opened the floodgates. Nearly everyone at the table confessed to similar workarounds using personal API keys to access data tools, copying datasets to free online analyzers, and finding creative ways to bypass the official (and painfully slow) data access request process.

This shadow IT problem isn't happening because Sarah and her team are reckless. It's happening because when your stakeholder needs an answer today and the official process says you'll get database access in two weeks, you find a workaround. When your choice is between missing a deadline or bending the rules, most people bend.

The question isn't whether AI will change data analysis. It already has. Walk into any analyst team's Slack channel, and you'll find people sharing ChatGPT prompts, discussing which AI tools work best for different queries, and helping each other navigate the gap between what they're supposed to use and what actually gets the job done.

The real question is whether companies will provide tools that are accurate, governed, and actually solve the problems that drive analysts to build workarounds in the first place.

The three questions to ask any AI data tool

1. Can you see the code it generates?

General AI tools sound impressive. They respond in natural language, seem to understand context, and deliver answers quickly. The problem is they often make things up. When ChatGPT generates data analysis, it might hallucinate metrics, invent table structures, or create plausible-sounding statistics that don't exist in your actual data.

The best AI tools for data analysis don't hide their work. They generate SQL or Python code that you can inspect, verify, and adjust. This transparency matters for three reasons.

First, you need to verify assumptions and business logic. When an AI tool tells you revenue increased 23% month-over-month, you need to see what it counted as revenue, how it handled refunds, and which date field it used. Without seeing the code, you're trusting a black box.

Second, business rules change. A metric definition that's correct today might need adjustment next quarter. If you can't see or modify the code, you're stuck recreating everything from scratch.

Third, you learn from the code. Even experienced analysts benefit from seeing how AI approaches a problem. It might use a more efficient join, a cleaner aggregation, or handle edge cases you hadn't considered.

This isn't just about verification. It's about maintaining control. Data analysis requires judgment about what to measure and how to measure it. AI should accelerate your work, not make decisions for you.

2. Does it connect to your actual data infrastructure?

Many AI tools require copying your data somewhere else. This creates immediate problems: security risks from data leaving your infrastructure, stale data that's hours or days behind, and another tool to maintain in your stack.

The alternative is AI that connects directly to your data where it lives. This means two things in practice.

First, it should integrate with your existing data infrastructure without requiring you to rebuild everything. obé Fitness implemented Fabi alongside their Snowflake and dbt setup. The tool integrated seamlessly, giving their director of data and analytics a single environment to work in rather than endless context switching.

Second, and this matters for early-stage companies, it should work regardless of whether you have a data warehouse. You can't afford to ignore your product and revenue data as an early-stage operator. Yet your data lives in silos. Getting insights traditionally requires either sifting through disjointed tools or hiring a data professional to aggregate everything into a warehouse. Then you need an analytics expert to write SQL and Python. Then you need a $50k contract to build dashboards in a BI tool that's impossible to use.

Modern AI-powered tools skip this entire chain. You can connect direct applications like Amplitude, PostHog, Google Analytics, or Google Ads. Or, if you've already centralized everything in a data warehouse such as Snowflake, BigQuery, or Postgres, connect to that instead. The tool adapts to your infrastructure rather than forcing you to adapt to it.

This approach eliminates a major friction point. Teams no longer wait for data engineering to build pipelines before they can analyze anything. They connect their tools and start asking questions.

3. Does it reduce busywork or just create different busywork?

The honest test of any AI tool is simple: what tasks actually disappear, and which just change form?

AI should eliminate the need to write repetitive SQL queries for similar questions, perform manual data cleaning and normalization, format results for different stakeholders, and switch context constantly between tools.

Real transformation stories show what this looks like in practice.

Aisle's product manager, Chirag Garg, was handling 5-10 data requests per week from the brand team. Each took 30 minutes to two hours. That's 15 hours per week on work that didn't align with product and engineering priorities,s but still had to be answered. Before AI-powered analytics, running complex queries across their brands took days of back-and-forth: understanding event tracking, writing and debugging complex SQL, optimizing for performance, and formatting results for stakeholders.

After implementing Fabi, this multi-week process was compressed to three days of exploratory data analysis. Dashboards and reports that once took hours or days now complete in 10 to 15 minutes. That's a 92% reduction in analysis time. The brand team now answers its own questions, eliminating 40-50 monthly data requests to product or engineering. Within the first month, 100% of brand managers were using the tool for their weekly reporting.

Tyler explains the shift: "I'm constantly looking for stories in our data—whether it's identifying which brands are seeing the best results, uncovering patterns we can turn into best practices, or finding compelling angles for case studies and content. Before, those insights required looping in our engineering team. Now I can run those analyses myself in minutes. It's completely changed how quickly I can move from question to actionable insight, and it's made it easy to build repeatable workflows that I can run whenever I need them."

Parasail faced a different challenge. As an early-stage startup, they needed enterprise-level analytics without the resources to support them. Matt Carnali and his team implemented Fabi, resulting in a 90% reduction in dashboard creation time. The learning curve was 4-5 hours compared to 2-3 months for traditional BI tools. Matt highlights the speed: "Fabi.ai took us four to five hours to learn. A traditional BI tool would probably have taken us a few months to get any sort of dashboard stood up."

What changed wasn't just speed. The team moved from a "finger in the wind" approach to scientific precision in their decision-making. Board and executive team meetings now rely on comprehensive analytics rather than intuition.

What actually improves (and what doesn't)

AI excels at specific tasks. It generates starting-point queries from natural language. "Show me revenue by customer segment for Q3" becomes working SQL in seconds. It automates repetitive analysis patterns. If you run the same weekly report with different parameters, AI can handle the variations. It creates visualizations from data without manual charting. It distributes insights through scheduled reports to Slack, email, or other channels where your team works.

Analyst judgment still matters in critical areas. You determine which questions to ask. AI can answer "what was our conversion rate?" but it won't know whether that's the right metric for your current business challenge. You understand business context and nuance. When numbers look unusual, you know whether that's an error, a seasonal pattern, or an important signal. You validate AI-generated logic against business rules. The code might be syntactically correct but still measure the wrong thing. You make strategic recommendations from insights. Data shows what happened; analysts explain why it matters and what to do about it.

The setup time test

If a tool takes months to implement, you're already losing. The "time-to-first-dashboard" metric matters. Traditional BI tools require building data models, defining metrics, and creating visualizations before anyone can use them. This takes months. AI-powered tools should let you query immediately.

From connection to first insights can take as little as 10 minutes for a technical founder connecting to production data. Aisle trained their non-technical brand team in 15 minutes and achieved 100% adoption among brand managers within the first month. People don't need to become AI data analysts. They need to connect their data, ask questions in natural language, and turn insights into workflows and dashboards in a few clicks.

The implication is significant. Many companies delay analytics initiatives because implementation seems daunting. If your timeline is "we'll set up proper analytics next quarter after we hire someone," you're losing insights every day. Modern AI BI tools compress that timeline to hours or days, not quarters.

Building workflows, not just dashboards

One-off analysis helps answer specific questions. Repeatable workflows solve ongoing needs. The difference is crucial.

Better AI data analysis tools let you chain analyses together, save them as workflows, and automate them. You start with exploratory data analysis, refine it until you have something useful, then turn it into a dashboard or automated report with a few clicks.

Chirag at Aisle uses this for automated daily and weekly scheduled reports. AI-generated summaries go to Slack and Google Sheets for easy consumption by the brand team. This isn't manual work disguised as automation. It's genuinely hands-off once configured.

The workflow capabilities should include the ability to share your work. When you build an analysis, others should be able to see your process, understand your logic, and adapt it for their own questions. This accelerates learning across the team. New analysts can see how experienced ones approach problems. Adjacent teams can build on existing work rather than starting from scratch.

Red flags: When AI isn't the answer

Let's be honest about limitations. AI for data analysis isn't always the right choice.

Traditional BI makes sense for large enterprises with dozens of analysts and established data governance, for companies with complex, highly customized visualization requirements that demand pixel-perfect control, and for organizations with deep integrations into enterprise software stacks that aren't going anywhere.

You should wait for AI data analysis tools if your data quality is too poor. Garbage in, garbage out still applies. AI can't fix fundamentally broken data. If you don't have clear questions you need answered, AI won't magically reveal what matters. Start with the business problems you're trying to solve. If you're looking for AI to replace strategic thinking, you'll be disappointed. AI accelerates execution; it doesn't substitute for knowing what to measure and why.

The honest answer: Is AI ready for data analysis?

AI is ready for data analysis if you choose purpose-built tools over general AI, verify the code AI generates, maintain governance and oversight, and use AI to augment analyst expertise rather than replace it.

The question isn't whether AI will change data analysis. It already has. For companies like Aisle and Parasail, AI-powered analytics didn't eliminate their need for analyst expertise. It eliminated 40-50 hours per month of busywork, letting analysts do what they were hired to do: generate insights that drive decisions.

You can view analyst inefficiency as a cost to be tolerated or a problem to be solved. The technology exists to solve it. The teams that implement it gain speed, retain talent, and make better decisions faster than competitors who don't.

That's not hype. That's just better infrastructure.

Try it yourself

Test any AI analytics tool with these three questions:

1. Show me the SQL you generated for this query
2. How do I verify this metric calculation?
3. Can I adjust the code if the business logic needs to change?

If a tool can't answer these clearly, keep looking.

Ready to see what AI-powered analytics looks like with complete transparency? Connect your data sources to Fabi, ask questions in natural language, and build workflows without needing to be a BI expert or having a legacy BI budget. Get started with Fabi for free in less than five minutes.