In this instructional video, Kedeisha Bryan argues that data analysts must evolve beyond traditional technical tasks to master four AI‑centric skills in order to stay employable as the field rapidly changes. She begins with a personal backstory—transitioning from low‑wage delivery work to leading a community of 80+ data professionals—to establish credibility and motivate viewers. She then details each skill: 1. Prompting and workflow design – analysts should act as unofficial AI product managers, crafting repeatable, system‑level prompts, building templates, and designing end‑to‑end AI‑enabled workflows that reduce manual “file‑hunting” work and scale analysis. 2. Output validation and AI QA – because AI often sounds confident while being wrong, analysts must implement a “logic bridge” of checkpoints, verify outputs against business knowledge, and protect their reputation by pressure‑testing AI results. 3. Data context and knowledge framing – most business knowledge is undocumented; analysts must create context libraries (metric definitions, domain vocabularies, policy notes) to ground AI models, preventing generic or hallucinated answers. 4. AI experimentation and prioritization – adopting a product‑management mindset, analysts should ask three critical questions before launching AI projects (is AI the right tool?, is the current process truly broken?, how will success be measured?) to avoid costly failed initiatives. Throughout, she cites industry research (MIT, Gartner, Forrester, Salesforce) and quantifies the financial impact of AI hallucinations (≈ $67 billion in 2024). She emphasizes that judgment, problem‑solving, and the ability to steer AI, rather than merely using tools, will differentiate analysts in the job market. The video ends with calls to action directing viewers to additional resources and a follow‑up video.
Chapters
If you wanna become a data analyst, you need to understand that this job is changing fast. And the analysts who stand out in this job market will not just be the ones with the technical skills, but the ones who know how to work with AI the right way. Because in this new era, that difference can determine whether you get hired or stay unemployed. So in this video, I'm gonna break down the 4 must-have AI skills for data analysts. If you're a subscriber, welcome back. But if this is your first time here, I'm Kedisha, and 6 years ago I was delivering pizzas and Amazon packages for $8 an hour. And today I am running a community of over 80 current and aspiring data professionals, and I've helped thousands of other career changers break into data analytics without going back to school and wasting more money on courses that just don't work. And if you ever want to work with me to land your first role in data analytics, click the link in the description. Now let's get into it. Okay, so skill number 1, prompting and workflow design. If you think your job is just running analysis, pulling numbers, building dashboards, handing over reports, you're already behind because your job in the new AI era is to act like an unofficial AI product manager. You decide what gets built, how it works, and what success looks like, and whether the output is trustworthy before anyone acts on it. And this shift matters more because companies are spending serious money on AI and many still have very little to show for it. They have all these tools, but what they lack are people who know how to make these tools even useful. And that starts with how you think about prompting. We are way past the era of just one-off ChatGPT questions. The companies getting real value from AI are building repeatable workflows, automations, and internal tools that their teams can use every day. And that means you need to think beyond prompts and start thinking in systems. So this year your role shifts from doing every step manually to supervising the process. You set the goal, you evaluate the output, and the system handles the middle. So what does this look like in practice? Before you send anything to a model, you give it context. Who is making the decision? What do they actually need to know? What happens if the answer is wrong? And that context matters more than people think. Then break the task into steps. One giant prompt backed with 5 questions usually give you one muddy answer. Break it down and you get something that you can actually work with. And build templates. If the quality of the output depends on who wrote the prompt that day, you don't have a workflow, you just have a guessing game. AI does not slow down when your process is broken. It actually speeds up the mistakes. A bad workflow doesn't give you slow, obvious errors. It gives you fast, confident, and repeated ones. And the more you automate it, the more expensive that gets. So you're not here to write prettier prompts. You're here to build systems that still hold up when decisions run through them. Think about your last real data project and how much time was actually spent on the analysis and how much was actually spent hunting down the right files, cleaning broken inputs, and reformatting a report you've already ran 10 times. For most of us in data, a huge part of the week never actually touches insight. That is actually what workflow design fixes. Companies building AI-enabled workflows are seeing serious returns just by simply reducing that manual repeated work. And once it compounds, that impact just gets bigger. One financial analyst can run simultaneous analysis across 50 markets, work that used to probably take a team of 10 an entire week. This is what happens when you build the right system instead of prompting your way through the same work over and over. The analyst who gets the workflow right early is not just faster, they are the person preventing a category of failure most teams do not notice until it has already cost them time, trust, or money. All right, skill number 2, output validation and AI QA. Researchers at MIT found that AI models are 34% more likely to use words like definitely, certainly, even when they're completely wrong. So that means confidence is not a trust signal when we're working with these AI models. And I call this a confidence paradox, and it matters because most of us have spent our entire lives reading confidence as competence. That instinct works better with humans than it does with AI with no emotional intelligence. AI can sound certain without being correct, and if you don't understand the work well enough yourself, you won't know how to tell the difference. And if you can't check the AI, their malfunctions becomes your problem. And the only way to protect your work and your reputation is to build a habit of verifying before anything important ever gets acted on. Anytime an AI output is about to touch a real decision, ask one question first. Does this actually hold up? Not does it sound polished, not does it sound smart, but does it actually hold up against what you already know about the business or context of the project? Does that trend even make sense given last quarter? Does that recommendation violate a real constraint? Does it ignore a pattern your company has seen for years? The system I use for this is called a logic bridge. It is a set of checkpoints that connects what an AI gives you and to what you already know is true. So if AI recommends a pricing change that breaks your entire margin floors, you can catch it. If it flags a trend that ignores known seasonal patterns, you can catch that too. You are not not here to admire the output. You are here to pressure test it. Businesses lost an estimated of about $67 billion in 2024 due to AI hallucinations, not because the tools existed, but because people trusted the output without even checking it. A wrong number made it into a slide deck, a recommendation got approved because it sounded reasonable, and a team acted on analysis that was confidently wrong. And in most organizations, nobody clearly owns that validation step. And this is where trust gets built. When you are the analyst who says, I checked this against our actual data, I tested the logic, and here's breaks, you become far more valuable than the person who just simply moved the fastest, just blindly trusting AI. Trust beats speed when real decisions are on the line. Any analyst can paste an AI output into a slide, but the analyst who gets pulled into a room is the one who says, I checked this, I found the weak spots, and I fixed them. That's not extra work. That is where your edge comes from. All right, quick pause. Now, if you want to land your first job in data analytics and avoid the lengthy mistakes that I did, click the link in the pinned comment where you can find the exact steps that my students use to land their job in data analytics without going back to school. All right, now let's get back to it. All right, skill number 3, data context and knowledge framing. How much of what your company knows actually lives somewhere a model can reach? For most companies, not much. A huge amount of business knowledge lives in emails, meeting notes, policy documents, old planning decks, and undocumented decisions sitting in someone's head. You cannot just paste that into a prompt and expect the model to magically understand your business. When you hand AI a business problem with no context at all, it just fills in the gaps with generic assumptions and just guesses things on their own. Your job is to fix the context problem before the output problem shows up. So start by building a context library of your most important and local tasks. Think of it like briefing a document for the model. Include metric definitions, business rules, domain vocabulary, background of how things work, and the assumptions that matter in your environment. The more grounded the model is in your actual business, the less generic the output becomes. Now remember, don't include anything that's proprietary information. Then do the harder work. Go after the knowledge that has never been organized. That might be last year's planning notes, the email that contains why a metric is calculated a certain way, the business rule everyone follows but no one documented. The work is not very glamorous, but this is what makes AI useful inside a real company. Because once you give the model real business context, the output starts reflecting your reality instead of generic nonsense. Salesforce research from 2025 found that only 43% of business leaders are fully confident in the accuracy of their organization's data. So if leaders already don't trust their data, what do you think happens when you just layer AI on top of it? AI does not fix weak inputs. It actually amplifies them. The companies getting poor AI output are often not dealing with bad models, but they're probably just dealing with bad inputs. Weak context, and just missing definitions. The model cannot tell the difference between a data quality problem and a real signal, but you can. And that is why the skill matters so much. Your value is not just in what you can do technically, it's also in what you understand about your domain, the business rules, and the context that makes analysis useful. Every hour that you spend building that understanding will make you much harder to commoditize. The model does not have that layer unless you give it one. Okay, skill number 4, AI experimentation and prioritization. This last skill moves you higher up the value chain because it is fundamentally a product management A lot of AI projects are failing for a simple reason. The hype moved faster than the judgment. Companies bought in because everyone else was buying in, not because they had a clearly defined problem that AI was actually the right tool to solve. That is why Gartner is projecting that at least 40% of AI projects will be abandoned after the first real test because of costs, poor governance, or just weak business value. That is not mainly a technology problem, it's really just a prioritization problem. The analyst who can slow that down and ask the right questions before money gets committed becomes much more valuable. Before any AI initiative, get real about real resources and ask 3 questions. First, is this actually a problem AI could solve? A lot of things being handed to AI would be better solved by just fixing the process, cleaning the data, or just simplifying the workflow. Second, can you prove that the current approach is actually breaking down? If the current system still works, then you may be solving a problem that does not even exist in the first place. Third, how will you know this even worked? What exact metrics should be improved, and by how much, and by when? If you cannot answer that upfront, you are not ready to even start. The analyst who walks into the meeting with those 3 questions is not just helping the project, They are shaping whether the project should happen at all. Only 34% of organizations are truly reimagining how they work around AI. The rest are just layering AI on top of the same weak processes and then wondering why their results are so disappointing. Leadership trusts people who ask hard questions, test properly, and bring evidence, not just yes-men who just wanna follow the hype. That is what separates someone who is excited about AI between someone who can actually help a business use it well. Every experiment teaches you something. Even the ones that fail sharpen your judgment, improve your instincts, and even give you real examples to talk about lot of interviews, performance reviews, and promotion conversations. Over time, that track record is what separates someone who has been around AI from someone who actually knows how to work with it. When people see layoffs or hear companies blame AI, a lot of them take that as proof that the window is closing, but that's not the full story. Forrester projects that 55% of companies making all these AI-related layoffs will regret them, not because AI does nothing— it absolutely can create value in the right setting— but many companies are making these permanent decisions based on what they hope AI will eventually do, not what they can reliably do right now in complex judgment-heavy work. And when that gap becomes so obvious, they still need people who can think, validate, frame context, and make sound decisions. That creates a new opening for analysts who actually know how to work with AI in real business environments, because the people with the most leverage these days are not the ones who know all the tools and all the skills and went to all the university degree programs. They're the ones with judgment, problem-solving skills. They know when to trust the output, they know when to question it, which experiments are worth running and which ones will waste months. That kind of judgment does not come from watching content All day. It comes from doing the work. Tools get replaced, judgment does not. The analyst who can direct AI instead of the one just using it and just blindly following it is the one who stays viable no matter what the tools look like next year or even 10 years from now. So in the new job market era, it's not about who knows everything. It's about who can build a workflow that holds up, who catches the errors before they hit a decision, and who gives AI the context it actually needs, and who knows when to say yes to a project and when to walk away. And that job is now. If you want help building that kind of profile, one that actually gets you hired and keeps you growing once you're in, watch this next video. I'll see you in the next one.
