Connor Dickson

I have had this work goal the past 3 years to be a sort of internal data consultant at Petco. If someone has a data-related problem to solve, they know they can come to me and I’ll discuss with them a solution and next steps.    Last week during the SQL office hours I cohost, I had a fun experience related to this.   I first shared that if you want to know if SQL could help you, think FAIL. SQL can help if you want to do something with data Faster, Automated, Independently, or with Large datasets, either in size or number.   Next I asked that people think for 5 minutes about a problem SQL could help them with.    Then people took turns a few minutes each describing a problem. I discussed with them different aspects, like how they’d use a solution and exactly which columns they’d like to see in a table.   So after a total of 20 min, I got 4 tasks that teams care about and that would probably take me like 8 hours combined to do. I shared with my team so we can think about which to prioritize, which to have junior members take on (as a great chance to work with new teams), and which to solve during future office hours. Plus, people outside our team know they can rely on us to help solve their problems.

119

8 Comments

🧵 Hearing "GraphQL" for the first time years ago at a conference made me think I needed to dust off my old TI-85 and remember how to graph equations. Not really, but I did realize I was way in over my head. I was still early in my career as an Analyst learning SQL, and now there's this new Query Language? How do you use it? Why do we have it? What is the purpose and function? What does graphing have to do with any of it? Spoiler: nothing. The "graph" refers to graph theory. Now, I'm going back to basics to learn more about GraphQL: It's a query language for APIs that lets clients ask for the data they need, nothing more and nothing less. One endpoint, strongly typed, and self-documenting. Why dust this off now? - It's everywhere, with big tech companies running GraphQL in production - Microsoft Fabric now has native GraphQL support. You can spin up a fully typed API endpoint directly from your Lakehouse or Warehouse with zero code - It changes how you think about data modeling. Designing a schema forces clarity about relationships and what consumers actually need - It's strongly typed by default, which means better tooling, autocomplete, and catching errors before they reach production Looking forward to sharing what I learn, and hearing from others on this journey. #GraphQL #APIDevelopment #DataEngineering #MicrosoftFabric #LearningInPublic

1

Fully aligned with this vision, regardless of the data-visualization tool being used, true value can only be achieved when insights are designed with—and for—end users. This means focusing on directly actionable KPIs that exist not just to look good, but to drive real decisions and spark meaningful actions.

4

Big companies have data teams. You have... yourself and maybe a bookkeeper. Good news: You don't need a data warehouse. You need 3 simple layers. 𝗟𝗔𝗬𝗘𝗥 𝟭: 𝗥𝗔𝗪 𝗗𝗔𝗧𝗔 Where transactions happen. - Bank feeds - POS system - Invoicing software - Payment processors (Stripe, PayPal) This is messy and that's OK. It's the raw material. 𝗟𝗔𝗬𝗘𝗥 𝟮: 𝗖𝗟𝗘𝗔𝗡 𝗗𝗔𝗧𝗔 Where transactions get organized. - Your accounting software (QuickBooks, Xero) - Transactions categorized - Duplicates removed - Reconciled to bank This is your single source of truth for money. 𝗟𝗔𝗬𝗘𝗥 𝟯: 𝗜𝗡𝗦𝗜𝗚𝗛𝗧𝗦 Where you actually make decisions. - Monthly P&L - Cash flow forecast - Dashboard with KPIs - Trends over time Most small businesses stop at Layer 1. Some get to Layer 2 (eventually). Few ever reach Layer 3 consistently. 𝗧𝗵𝗲 𝗴𝗮𝗽 𝗶𝘀 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻. Raw → Clean should happen automatically (bank feeds, rules, integrations) Clean → Insights should be one click (reports, dashboards) When it's manual, it doesn't happen. When it's automatic, you actually use it. Which layer is broken in your business? #DataArchitecture #SmallBusiness #BusinessIntelligence #Automation

3

4 Comments

AI is only as good as the context you give it. I've been refining a Tableau chatbot, and my biggest takeaway has been the power of Context Engineering. LLMs have a finite canvas. If you don't manage that space effectively, then you run into issues like context drift and degraded performance. To solve this, I’ve focused on learning ways to maximize the efficiency of the initial tokens passed so no critical insights are lost in the noise. As a first step I have been learning about context engineering in the form of role based tailoring. I have enhanced the chatbot to recognize whether the user is a Viewer, a Manager, or an Executive. - Smart Data Usage: We make the best use of the initial context window to prevent information loss. - Token Efficiency: We aren't just "dumping" data; we are curating the most relevant facts to keep the output sharp and cost effective. - Tailored Insights: The output matches the user's specific expectations and goals. I can now fine-tune these personas in minutes, which historically would have taken days or weeks. The speed at which we can now iterate is a total game changer!

15

‎Not all losses show up in the churn rate. ‎Some customers are still there, just not with you. ‎They’ve disengaged, quietly… and your data knows it. ‎Hello #datafam ‎Too often, valuable customers go unnoticed; transacting less, ignoring offers, silently drifting. Not because they’re lost… but because we never really understood them. ‎This Bank Analysis Dashboard was built to answer that exact problem. By connecting product performance, channel usage, revenue drivers, and customer behavior in one dynamic view to help answer questions that matter: • Which customer segments are driving revenue? • ‎Where are we losing customers and why? • ‎How are digital channels really performing? • ‎Are our products meeting the needs of each income group? ‎🔎 What the data revealed: • ‎Customer count is down YoY, yet profit is up +3.75%. Meaning fewer people are now carrying the business. • ‎High-income customers transact more but aren't the biggest revenue drivers. That crown goes to the middle-income segment. ‎• Mobile banking is growing rapidly, but Internet Banking still dominates, especially among middle-income users. ‎• Only 49% of customers engaged with offers. Among low-income users, that number drops to just 31%. ‎• Our Credit Card and Savings products perform well, but Loan Products show weak revenue uptake, hinting at trust or access gaps. What happens next? Insight means nothing without action. Some customers are still there, but no longer engaged. Use your data to spot the quiet exits before they become real losses. ‎Built in Excel, dynamic with slicers, and designed to make strategic insight immediate. Curious how it works? Watch the demo below. ▶️ [Video Attached] #BankingAnalytics #CustomerRetention #BusinessIntelligence #ExcelDashboard #Excel #DataAnalysis #DataAnalytics #Banking

37

24 Comments

Lies data analysts tell themselves: "I'll document this later." You won't. In six months you'll reverse-engineer your own work while grumbling about the "idiot that didn't document it." "This is the final version." The file is called v7_FINAL_actual_FINAL_v2. "I'm not territorial about my dashboard." Someone suggested moving a chart last week and gave 3 good reasons and 7 bad ones why it should stay in the same place. "I don't need to validate, the source data looks fine." In the history of your entire career, the source data has never once been fine (or on time). "I could explain this to a non-technical person." As long as they already understand the technical part. "I'll clean up the code before I hand it off." That code is getting handed off exactly as ugly as it is, with a comment that says "TODO: refactor" and has been there since 2022. "I just like working with data." Nah, you like being the person with answers. Smart ass. "I don't care if they use the dashboard." You always checked the view count.

217

38 Comments

Being an AE isn't always SQL queries & data modeling. Many days I'm sifting through documentation to figure out what something is supposed to mean or taking a course to keep up with the vastly changing tech landscape. 🌪️ Other days I'm debugging a failed daily job or trying to figure out why my local dbt build won't pass (usually a comma or forgetting to put something in the yaml file.) 🤦🏾‍♀️ Either way, I believe it's important to stay curious and learn from whatever your day brings! #analyticsengineer #data

3

3 Comments

Stop building dashboards that nobody trusts. 🛑 In the old world of data: Data Engineers built the pipes. Data Analysts built the charts. Everything in between was a "black box" of messy SQL and broken logic. Enter: Analytics Engineering. I’ve been diving into Module 4 of the DataTalksClub Data Engineering Zoomcamp, and it’s a game-changer. It’s not just about moving data; it’s about applying software engineering best practices to transform it. The secret sauce? dbt (data build tool). In this module, we didn't just write queries. We built: ✅ Modular SQL: Using ref() to create a lineage you can actually follow. ✅ Testing: Catching nulls and duplicates before the CEO sees them. ✅ Documentation: Automating the "What does this column mean?" conversation. ✅ Deployment: Moving from BigQuery/Postgres dev environments to production like a pro. The gap between "raw data" and "business value" is narrowing. If you aren't thinking like an Analytics Engineer yet, you're leaving reliability on the table. Huge shoutout to Juan Manuel Perafan spend a few more seconds to talk about IAM role that most educator just skip and Victoria Perez Mola for making this world-class content free. 🚀 Had moments of "productive struggle" going through this module🥹, I can celebrate Chinese new year with peace of mind now 😍🎉 #DataEngineering #AnalyticsEngineering #dbt #BigQuery #DataTalksClub #LearningInPublic

7

If your SQL query feels slow...it probably is. And before you start blaming the database, ask yourself: Are you filtering before joining or joining before filtering? Are you pulling columns you don’t need? Are you using LIKE '%value%' without knowing what it’s doing to your indexes? Are you relying on SELECT * out of habit? SQL isn’t just about getting the right answer. It’s about getting the right answer efficiently. #SQL #QueryOptimization #DataAnalytics #DataEngineering #BI

12

𝗛𝘂𝗺𝗮𝗻𝘀 𝘄𝗶𝗹𝗹 𝗮𝗹𝘄𝗮𝘆𝘀 𝗯𝗲 𝗶𝗻 𝘁𝗵𝗲 𝗹��𝗼𝗽... I’m transitioning into a new company, but right now, let's talk about the "Annotation" role... As long as AI keeps growing, humans will be needed to label, review, and refine data. That part isn’t changing anytime soon. What needs to be clearly understood is the structure of the work. 𝗗𝗮𝘁𝗮 𝗮𝗻𝗻𝗼𝘁𝗮𝘁𝗶𝗼𝗻 𝗶𝘀 𝗺𝗼𝘀𝘁𝗹𝘆 𝗽𝗿𝗼𝗷𝗲𝗰𝘁-𝗯𝗮𝘀𝗲𝗱, 𝗻𝗼𝘁 𝗲𝗺𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁-𝗯𝗮𝘀𝗲𝗱. Yes, you can sign contracts with companies, you get to work on projects whenever it's available. Yes, the pay can be good. I’ve worked on projects paying $5, $10, $17–$20 per hour. But it's mostly project-based Some projects can last two months. Others run more.. A few stretches close to a year.  𝗔𝗻𝗱 𝗲𝘃𝗲𝗻𝘁𝘂𝗮𝗹𝗹𝘆... 𝗽𝗿𝗼𝗷𝗲𝗰𝘁𝘀 𝗲𝗻𝗱!, 𝗕𝘂𝘁 𝗛𝗼𝘄 𝗽𝗿𝗲𝗽𝗮𝗿𝗲𝗱 𝗮𝗿𝗲 𝘆𝗼𝘂 𝘄𝗵𝗲𝗻 𝘁𝗵𝗲𝘆 𝗱𝗼? One of the smartest ways to create security in data annotation is diversification. Instead of depending on one platform or a single annotation type, smart annotators combine: Data annotation, data review / QA, transcription, content evaluation, AI training support, even non-ai oriented work. When you build multiple income streams: You don’t panic when a project wraps up. You stay financially steady while waiting for the next opportunity. And you have options. 𝗜𝗳 𝘆𝗼𝘂 𝗰𝗵𝗲𝗰𝗸 𝗺𝘆 𝗽𝗿𝗼𝗳𝗶𝗹𝗲, you’ll see I’ve worked across multimodal annotation: LLMs, video, audio, image. Different techniques, different industries, different project lengths... 𝗔𝗻𝗱 𝗼𝘂𝘁𝘀𝗶𝗱𝗲 𝗮𝗹𝗹 𝘁𝗵𝗮𝘁? 𝗜’𝗺 𝘀𝘁𝗶𝗹𝗹 𝗮𝗻 𝗙𝗫 𝘁𝗿𝗮𝗱𝗲𝗿  Smart work! 😉 You can book a consult with me to learn more about our algorithm (TITA). 👉🏽 https://lnkd.in/egAuK7ak #DataAnnotation #AITraining #HumanInTheLoop #AIWork #FutureOfWork #FXTrader #TechAndFinance #TITAAlgorithm #MachineLearning #LLMEvaluation #AIAnnotation #TechCareers #RemoteWork #SmartWork

10

Stop lumping data analysts in with IT. Analysts are business strategists. They don’t troubleshoot hardware. They don’t manage logins. And they’re not tucked away cranking out dashboards no one uses. Their role is to embed with marketing, product, ops - wherever decisions happen and translate data into actionable insight. If they’re siloed in IT, disconnected from the business, you don’t get clarity. You get lifeless spreadsheets. The real value shows up when analysts are in the mix: → Marketing says, “We’re rolling out a new initiative.” → Analyst says, “Here’s what succeeded before, here’s where you’ll burn budget.” That’s influence. And that’s the gap between “data support” and “growth partner.” If your analysts don’t speak the language of the business, they’re just pricey report machines. — ♻️ Repost if you’ve seen analytics buried in IT. → Follow me, I share interesting stuffs that actually makes sense

12

𝗖𝗹𝗲𝗮𝗻𝗶𝗻𝗴 𝗗𝗮𝘁𝗮 𝗜𝘀 𝗛𝗮𝗿𝗱𝗲𝗿 𝗧𝗵𝗮𝗻 𝗠𝗼𝗱𝗲𝗹𝗶𝗻𝗴 Hot take. Most people get excited about regression models and predictive analytics. But in healthcare? Data wrangling determines whether your model even makes sense. You can build the most beautiful model… But if: • ICD codes were grouped incorrectly • Null values weren’t handled • Cohorts weren’t defined properly Your prediction is noise. In my experience, 60–70% of the work is data cleaning and validation. Modeling is the final layer — not the foundation. Agree or disagree? 〰️Kosi Clean data. Clear decisions. Better healthcare.

6

3 Comments

5 things that shocked me when I built my first real training pipeline: Training rarely happens once, it’s scheduled. Data inconsistencies matter more than model accuracy. Your pipeline needs a backup pipeline. Most bugs aren’t ML bugs; they’re infrastructure bugs. Logging is your best friend until you realize you didn’t log enough. Real ML engineering is humbling. In the best way. Share your humbling moment in the comment section. #AIEngineering #Model #JoshuaAdeyemo #MLEngineering #RealWorldTrainingPipeline.

10

8 Comments

Where’s data viz headed in 2025? New tools, new formats, and new expectations are reshaping how we design with data. These are some of the trends I’m watching closely (and already leaning into). 1) 𝗔𝗜-𝗔𝘂𝗴𝗺𝗲𝗻𝘁𝗲𝗱 𝗗𝗲𝘀𝗶𝗴𝗻: There’s lots of hype right around tools like bolt.new and Lovable right now. But the results still feel too basic and formulaic to me. I’m honestly more interested in tools like Relume and Visual Electric, which aim to enhance the design process in specific ways, not replace it. 2) 𝗖𝗼𝗹𝗹𝗮𝗯𝗼𝗿𝗮𝘁𝗶𝘃𝗲 𝗗𝗮𝘁𝗮 𝗔𝗻𝗮𝗹𝘆𝘀𝗶𝘀: When Observable announced the debut of Data Canvases last week, I was stoked. We’ve had real-time collaboration in design tools like Figma for years. I’ve always wanted the equivalent for data teams and this feels like a big step. 3) 𝗔𝗻𝗶𝗺𝗮𝘁𝗲𝗱 𝗗𝗮𝘁𝗮 𝗦𝘁𝗼𝗿𝗶𝗲𝘀: More and more clients are asking us to create animated videos with their data. Think short GIFs (<30 seconds) featuring a slick chart and a single, sharp insight, designed to stand out in a crowded social feed. In a world driven by TikTok and short-form video, this makes sense creatively and strategically. 4) 𝗪𝗲𝗯𝗚𝗟 𝗮𝗻𝗱 𝗦𝗵𝗮𝗱𝗲𝗿𝘀: If you're looking to make your visuals pop, WebGL and custom shaders let you push the browser to its creative limits. We’re using them to build visuals that feel dynamic, immersive, and a little bit magical. 5) 𝗡𝗼-𝗖𝗼𝗱𝗲 𝗣𝗹𝗮𝘁𝗳𝗼𝗿𝗺𝘀: Tools like Webflow and Framer are democratizing web development and taking dev timelines from months to weeks. We’re leaning into this on specific projects more and more. 6) 𝗜𝗺𝗺𝗲𝗿𝘀𝗶𝘃𝗲 𝟯𝗗 𝗩𝗶𝘀𝘂𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻𝘀: The rise of Three.js and Blender is helping 3D find its place in data storytelling. A few years ago, I would have dismissed this as a fad. But recent examples from NYT and Washington Post, where they’ve used 3D to great effect, have convinced me otherwise. Data people, what’s on your radar this year?

18

1 Comment

Permissions don’t break loudly. They drift quietly. Most Tableau environments don’t fail because of one bad decision.   They get harder to operate because of small inconsistencies that pile up. So, here’s this week’s admin pattern. This script gives you visibility into your groups, what exists, how many there are, and where users are assigned, before you change anything. It’s not glamorous.   But it answers a foundational question: Do I actually understand the structure I’m operating? Before refactoring permissions.   Before tightening governance.   Before automating anything. Run it.   Scan the output.   Notice what surprises you. I’ll share what I noticed midweek. #TableauOps #Tableau #TableauAdmin #OperationalExcellence #DataOps

21

5 Comments

Just got back from the Kansas City Data Professionals meetup on "The Data Product Revolution" and honestly, it was exactly what I needed to hear. The whole premise hit home: we don't have a data problem, we have a trust problem. How many times have we built faster pipelines, added more dashboards, only to have teams still arguing over which numbers are "right"? Or rebuilding datasets that already exist because nobody trusts what's there? The session broke down what "Data as a Product" actually looks like in practice - not the buzzword version, but real talk about: Why ownership matters more than we think Why governance keeps failing (hint: stop bolting it on at the end) How to actually build trust into your data from the start No fluff, no vendor pitch. Just practical stuff you can actually use. Shoutout to David Traynham, Melanie Traynham, MSN, RN, CPN, and Matthew Copple, CDMP for organizing. If you're in KC and dealing with data headaches, this group is worth checking out. Great meeting some awesome data folks too - Nathan Hayden, Bonnie Why, Cynthia Stacks, and Matt Denney. Love this community! #DataEngineering #KansasCityData #DataProducts #DataGovernance #Networking

33

3 Comments