Charles Green

Many data teams succumb to the temptation of cutting corners and skip testing altogether. This seems like a time-saver now but becomes costly later. I encourage you to stop thinking of rigorous testing as a delay or nice to have; it’s an investment in building robust and reliable platforms. Every hour spent validating your code upfront saves you 10s of hours troubleshooting, fixing issues, and untangling tech debt later on. Prioritizing testing is setting yourself up for success with: - 𝗥𝗲𝗹𝗶𝗮𝗯𝗹𝗲 𝗗𝗲𝗹𝗶𝘃𝗲𝗿𝗮𝗯𝗹𝗲𝘀: Quality data pipelines that deliver accurate insights. - 𝗥𝗲𝗱𝘂𝗰𝗲𝗱 𝗧𝗲𝗰𝗵 𝗗𝗲𝗯𝘁: Less bugs is less time spent scrambling to fix issues. - 𝗘𝗻𝗵𝗮𝗻𝗰𝗲𝗱 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆: Focus on innovation instead of constant firefighting. - 𝗠𝗼𝗿𝗲 𝗧𝗿𝘂𝘀𝘁: Peers and stakeholders gain confidence in your deliverables. To truly implement testing requires an intentional effort; committing testing time in project planning, holding peers accountable and having an agreed testing strategy and framework. The long-term ROI of a test-centric approach is clear: fewer bugs, less time firefighting, more predictable outcomes, and stronger relationships between the business and data teams. As an engineer, I don't measure my value by what I do quickly, but by the reliability and longevity of the work I do. Are you happy with the amount of testing you implement in your day-to-day work? Would you change it if you could?

11

3 Comments

AI won't replace SQL Server DBAs, and here’s why. Despite what everyone is telling you. I am repeatedly seeing some AI folk claiming DBAs are the next job to be automated into oblivion. This is bullshit. Here's why: For marketing emails or blogs, AI errors are tolerable. No one will lose their job over them. But with critical infrastructure, you can’t afford to roll that dice and hope it’s good. What happens when AI makes a mistake in your production database? - System downtime - Introducing new bugs - Corruption - Data loss - Downtime Even a 1% error rate is catastrophic when a single change can take down your entire server. We’ve been testing AI for lots of different scenarios recently. And we have found that you still need a senior DBA to understand what it is doing. You can get more value out of a senior DBA, but you can’t remove it. We’re talking ChatGPT, Claude, Perplexity… all of them. But don’t get me wrong, AI is great and our usage is increasing. But for business critical infrastructure, senior DBAs is still what you need.

36

3 Comments

SQL Server isn’t slow. Your indexing strategy is. After years of working with SQL Server systems, one pattern keeps repeating. Indexes exist. Performance is still bad. Because SQL Server isn’t seeking. It’s scanning. Here are 6 common indexing mistakes quietly killing performance: 1️⃣ Indexing everything “just in case” Too many indexes confuse the optimizer and slow writes. More indexes ≠ more seeks. 2️⃣ Getting column order wrong The right columns in the wrong order turn seeks into scans. 3️⃣ Using functions on indexed columns Wrap a column in YEAR() or LOWER() and SQL Server can’t seek — it scans. 4️⃣ Missing indexes on JOIN columns No usable index means no seek. Just scans, hash joins, and memory pressure. 5️⃣ Ignoring fragmentation and maintenance Fragmented indexes can’t seek efficiently. They behave like scans. 6️⃣ Indexing low-selectivity columns If an index points to half the table, SQL Server stops seeking and starts scanning. Here’s the rule that never changes: Good indexes = index seeks Bad indexes = index scans And scans are where performance goes to die.

37

HFM is a data engine, not just a reporting tool. Put energy into loading clean data, setting up strong mappings, and designing a scalable dimension structure. Flashy reports can’t fix messy inputs. Better input = better output. #OracleHFM #DataGovernance #CloseExcellence

7

1 Comment

Most data teams are wasted on dashboards. If that’s all they’re building, you don’t have a data team. You have a support desk. Most companies treat data like an internal helpdesk: → “Build me a dashboard for X.” → “Can you pull numbers on Y?” → “We need a weekly report for Z.” This turns analysts into order-takers, not problem-solvers. The best teams I’ve worked with operate differently: ✅ They challenge the questions being asked ✅ They surface uncomfortable truths departments don’t want to hear ✅ They focus on business outcomes, not pretty charts ✅ They say “no” to vanity metrics, “yes” to leading indicators Example: → Marketing requests a campaign performance dashboard. → Report generator builds it. → Truth seeker asks: - “What is the problem you are trying to solve?" - "How would you action the data?” → Instead of following orders, they focus on solving the problem. Often the problem isn’t missing data. It’s missing clarity on what success looks like. Your data team should be neutral ground: → Revenue drops get investigated, not explained away. → Assumptions get tested, not assumed. → Truth outranks comfort. What’s one hard truth your data team quietly knows that leadership isn’t ready to hear?

71

18 Comments

You know what cracks me up? The idea that giving feedback after an interview is somehow too dangerous to risk. “Oh no, we can’t tell them why we didn’t move forward, liability!” Right, because saying “we needed more SQL experience” is clearly a lawsuit waiting to happen. So instead, companies do the corporate equivalent of sneaking out the back door. After five interviews, two assessments, and a small blood sample, you get the famous silence. Or worse, the vague “we’ve decided to move forward with other candidates.” Other candidates where? Doing what? Were they real or just a concept? Meanwhile, job seekers are left guessing. Was it the experience gap? Was it my answer to “tell me about yourself”? Was it the fact that I said synergy with too much sarcasm? We dissect every word like detectives solving a crime. And the crime is… being left hanging. Here’s the thing, feedback isn’t liability. Feedback is humanity. Even if it’s brief and bland “we went with someone with more direct industry experience” at least you can file it away, adjust, move on. Instead, companies choose radio silence like it’s an advanced HR strategy. The irony is the best hiring managers know this. They close the loop. They deliver feedback with kindness and clarity. They know a simple, “You were great but we need XYZ” isn’t going to end in court, it’s going to build goodwill. Some even earn long term fans who might apply again later or recommend someone else. But the liability fear has spread like glitter. It’s everywhere, sticky and irrational. “Don’t give feedback” has become default because no one wants to risk a legal boogeyman that rarely appears. The result? An entire generation of job seekers wandering LinkedIn, haunted by unanswered questions and rejection emails that read like auto-generated breakup texts. What’s wild is how low the bar actually is. You don’t have to write a novel or justify every decision. Just something. “You made it far but we went with a candidate with more direct X experience.” Done. Two sentences. Closure. No therapy bill. To the managers and recruiters who still close the loop, thank you. You are rare, brave creatures. You spare people endless self doubt and keep the process human. You probably don’t even realize how much credibility and goodwill you’re earning. To everyone still getting ghosted, it’s not you, it’s the system. Silence says nothing about your value. It says everything about a culture afraid of its own shadow. And maybe one day, companies will realize the real liability isn’t feedback, it’s burning bridges with good candidates who might have been perfect for the next role. Until then, we’ll keep joking about it, supporting each other, and writing posts like this because someone has to say what everyone’s thinking. So here’s to normalizing closure. It’s not that hard. If you can run seven rounds of interviews, you can send one honest note at the end.

54

14 Comments

Stop copy-pasting pipeline code. Start generating it. Every data team has that one pattern they implement 50 times. Customer data ingestion. Event stream processing. CDC pipelines. Same logic, different parameters, copy-paste-pray it works. You'd be surprised how often this happens. 🔹 𝗧𝗵𝗲 𝗰𝗼𝗽𝘆-𝗽𝗮𝘀𝘁𝗲 𝘁𝗮𝘅: Research shows developers spend nearly 50% of their time on maintenance alone. How much of that is fixing copy-paste inconsistencies? 🔹 𝗧𝗵𝗲 𝗿𝗲𝗮𝗹 𝗰𝗼𝘀𝘁: When you update the pattern, you're hunting down 50 implementations. Miss one? Production breaks. 🔹 𝗧𝗵𝗲 𝘀𝗼𝗹𝘂𝘁𝗶𝗼𝗻: Data Applications. Define patterns once. Generate implementations instantly. Update the blueprint, every instance updates automatically. Here's what this looks like at Ascend, and it happens to be one of my (and our customers') favorite features: ✅ 𝗬𝗔𝗠𝗟-𝗱𝗲𝗳𝗶𝗻𝗲𝗱 𝗯𝗹𝘂𝗲𝗽𝗿𝗶𝗻𝘁𝘀—parameterize anything from transforms to tasks ✅ 𝗜𝗻𝘀𝘁𝗮𝗻𝘁 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻—entire pipelines from a few parameters ✅ 𝗩𝗲𝗿𝘀𝗶𝗼𝗻𝗲𝗱 𝗽𝗮𝘁𝘁𝗲𝗿𝗻𝘀—update once, propagate everywhere ✅ 𝗘𝗻𝗳𝗼𝗿𝗰𝗲𝗱 𝘀𝘁𝗮𝗻𝗱𝗮𝗿𝗱𝘀—no more "creative" implementations The impact? Teams are replacing dozens of similar pipelines with a handful of Data Applications. Result: dramatically less maintenance, faster deployment, consistent quality. 👉 The best code is code you don't have to write—or maintain. #DataEngineering #Automation #CodeReuse #DataOps

11

3 Comments

Some SQL Server environments just won’t budge. CDC locked down. Log access restricted. DBAs saying “no” for security reasons. That used to mean you couldn’t stream changes — or had to build brittle workarounds. Not anymore. Artie now supports **SQL Server Change Tracking (CT)** as a native replication method. CT captures primary keys and version numbers for changed rows directly from SQL Server — no elevated permissions, no direct log access — and streams those changes in real time. Why it matters: ✅ Lightweight replication with minimal database overhead ✅ Works in managed or restricted SQL Server environments ✅ No DBA intervention or complex permissions required ✅ Keeps your data fresh with the same reliability as CDC Low-latency syncs. Zero friction. Change Tracking just made SQL Server replication a whole lot easier. 📖 Full changelog in the comments.

10

1 Comment

You trained the model. You trusted the output. But can you trace the data that led to the decision? 👀 AI in analytics isn’t dangerous because it’s wrong. It’s dangerous because it’s convincing, even when it’s wrong. 📰 This article breaks down the real risks of AI-powered insights and how to design for trust with traceability, governance, and human oversight. Read it here: https://lnkd.in/eKuQnv7J

10

You don't need to take another SQL course. You need to practice what you already know. Most data analysts only utilize about 30 different SQL keywords. You can probably learn the top 30 SQL keywords in under 2 hours--but that would not make you "SQL-literate". "Learning SQL" has two parts to it: 1) Learning the syntax 2) Thinking critically so you can take a business problem, understand what data you need to pull to solve the problem, and write an efficient query to get that data The latter is much more complicated and time-consuming to learn. And yet, it's easy to get caught up in thinking that we need to spend more time taking more courses. What you really need is to practice using SQL to solve business problems. This week, take 5 hours to work on a real-world practice project. Try not to fall into the mindset of trying to get as much done as possible in 5 hours. Instead, focus on ensuring each question is answered completely. This will be much more valuable than taking 5 hours to watch another SQL tutorial.

210

25 Comments

“If you can’t measure it, you can’t understand it. If you can’t understand it, you can’t control it. If you can’t control it, you can’t improve it.” – H. James Harrington   This quote sums up why so many SQL Server environments drift into trouble. → Backups silently fail. → Jobs run longer and longer. → Security gaps widen. → Performance slowly degrades.   But no one sees it, because no one is measuring.   At The SERO Group, measurement is where we start. We begin with a SQL Server Health Check. Then we continue with: ✅ Daily health checks ✅ Monitoring & alerting.  ✅ Backup validation. ✅ Patch status. ✅ Security posture. ✅ Trends over time.   We measure so we can understand. We understand, so we can control. And we control, so we make things better.   Better health. Better security. Better reliability. Better recoverability.   That’s how stability happens—on purpose, not by luck. 

26

8 Comments

The #1 reason data migrations fail has nothing to do with technology. Everyone who thinks data projects need to take months or years is setting you up for failure. You are either: - Migrating from legacy to cloud data warehouse - Changing or optimizing the new data infrastructure - Getting acquired and merging systems together Everyone is stuck in planning, claiming it will take months or years to do it. Most companies treat it like moving houses a "lift and shift" of all their old junk into a new, more expensive location. But in reality, you don't do everything; 70% of it needs to be removed anyway. Instead of a purely technical plan, lead with a business-first framework. This applies to any data initiative, from simple to complex. Before start you need to have data strategy dialled in **1. Strategy Blueprint** Before you move a single byte, create a business case for every data asset. For any data initiative, you should have a clear end user, results, and use case written next to it. Conduct workshops with stakeholders to understand the priorities of all units. (Data Strategy should already be done, but most haven't done it.) Here, you'll notice +50% are irrelevant, useless, and also parts that are a priority right now. **2. Migrate by Value** Don't migrate "data warehouse" or use technical jargon. Migrate "the Marketing ROI Model" or "Commercial patient insights for X leadership." Structure the project into phases, each delivering a validated, high-value asset to a specific business unit. Based on the first step, you know what the lowest-hanging fruit is. Each phase is funded by the success of the previous one. That way, you get quick wins in weeks and build momentum throughout the process. **3. Validate & Verify:** Your new platform is worthless without 100% trust. The only way to earn it is to run the new system in parallel with the old one and prove mathematically that the numbers match. Document the entire process. This is non-negotiable. Sign off and agree that the new data is correct *before* retiring the old system. **4. Feedback & Adoption** The project isn't finished when the new system is live, but when the old system is turned off. Conduct weekly feedback calls, ensuring trust is there Documentation of the new whole process Check back if achived the outcomes according to timelines. This way you gone get quick wins, build according to value. 🧬 Repost if you've think data migration fail due no strategic plan. → Follow for more on building data teams that drive real business value.

21

3 Comments

Just a reminder: Data Governance isn’t the same as Data Management. If we’re building a data product, DG Team is not the one filling out the metadata forms, documenting access controls, or uploading definitions into the catalog. DG team is the one saying: “Here’s what metadata is needed.” “Here’s why it matters.” “Here’s who should be doing it.” Expecting DG to do the documentation is like asking the librarian to write the books. We organize, guide, and ensure discoverability but the content? That’s best done by the one's who know their domain in and out or perhaps a team effort. We should keep the roles clear so we can all move faster, together. #datagovernance #datamanagement #datacatalog

243

17 Comments

When someone asks: “Why does this number look off?” …it could be a DAX issue. But it could also be one of 20 other things. Let’s be real though, getting to an accurate final number involves way more than just writing a good measure. It’s the entire system: Data input — Was the right thing entered at the start? ETL / ingestion process — Was the data even pulled in correctly? Stored procedures — Are transformations happening upstream that you didn’t know about? Data freshness — Are you looking at updated data or something from two days ago? Business processes — Like RCM lag or delays, data entry practices/timing, office hours/PTO, etc. Power Query steps — Did something get removed or filtered too early? DAX measures — Are they doing what you think they’re doing? Page-level filters — Are slicers or hidden filters skewing the view? When someone spots a "wrong" number, the real question should be: "Where in the process is this going sideways?" Accuracy isn’t a formula. It’s a chain of decisions, dependencies, and assumptions. And any weak link can change the story you’re telling. #DataAnalytics #PowerBI #DataQuality #ETL #DAX #DashboardDesign #RCM

8