Udi Schlessinger

If your datasets are poorly labeled, you’re inviting a host of issues like noise, bias, and inconsistencies that can seriously hurt your model's accuracy and reliability when it’s out in the wild. That's why data labeling has evolved from being just a preliminary step into a vital part of engineering and operations in the AI lifecycle.

7

Lately I’ve been jumping between a few AI and data-intelligence projects at Snowflake, and one thing keeps hitting me over and over again: the hardest part of AI has very little to do with the model itself. Models are cool. They look great in slides and polished demos. But the second you plug them into real workflows, real products, and real customers… things get interesting really fast. Here are a few things I keep running into: Data looks clean until the moment you actually rely on it Every team has its own idea of how something “should work” A perfect solution on paper turns into something completely different in production Integrations sound easy, right up until you start doing them for real And my personal favorite: “We’ll fix that later,” which usually translates to “we’ll rebuild this twice” What’s funny is that these challenges barely change from company to company. Whether it was Snowflake, ServiceNow, Cisco, or even earlier in my research days, the pattern is pretty much identical: AI only becomes useful when teams communicate well and work through the messy parts together. It’s not glamorous. It’s not magic. It’s just alignment, transparency, and adjusting when reality doesn’t match the plan. And honestly, that’s what I like about this field. The tech is fun, but the real progress comes from people actually building things together. Just wanted to jot this down before I forget. Curious what others are seeing in your teams lately. For you, what’s been harder: the model or the people part? By the way, here’s a video that showed up in my feed and fits this topic pretty well 👇 https://lnkd.in/g3wTuzb6

2

Last week, I spoke at the Data Behind the Scenes conference #DataBS It's a conference run by Randy Au for data practitioners who are tired of polished case studies that skip over the actual work. Instead you get honest takes on project realities: the messy bits, the trial and error, the things that go wrong. I spoke about the 10 red flags that your multi-market research data went sideways. We've all had that moment, staring at survey results that look technically correct but feel fundamentally wrong, then having to explain why three countries' worth of data is unusable. Links to my talk, as well as the full playlist, are in the comments

16

5 Comments

We're facing more and more questions about the use of synthetic data in market research. I'd recommend this blog post, which really resonated with me. So much of the conversation is "how closely can this replicate human data?" But there are so many other considerations researchers should be thinking about. https://lnkd.in/eSQfSTPj

31

3 Comments

WWW.ELIXIRDATA.co and the Value of its built in AI Agents: ElixirData’s built-in AI agents are like having a team of data-savvy analysts that never sleep. Each agent watches one cost-heavy process, spots leaks, and tells you (or the system) what to fix—before the money walks out the door. 1. Spend-Agent – scans every PO, invoice and contract line. If the same SKU is bought at two different prices, or a discount tier isn’t applied, it flags the variance and can auto-send a “claim credit” email to the supplier. Typical catch: 1–3 % of annual addressable spend recovered in 90 days. 2. Energy-Agent – ingests IoT meter data + utility tariffs. When a shift or building is trending above its kWh baseline, it recommends throttle-down schedules or peak-hour avoidance. Plant example: $420 k saved in year-one by shifting two production lines to off-peak. 3. Inventory-Agent – fuses MRP, sales forecast and freight lead-times. If safety-stock is higher than the 95 % service-level target, it proposes a transfer order instead of a new buy, cutting carrying cost. Retail case: freed $6.8 M cash and cut obsolescence write-offs 28 %. 4. Maintenance-Agent – reads sensor vibration, temperature, and past work-orders. Predicts failure 7–10 days out, letting you schedule minor repairs instead of emergency shutdowns. Chemicals client: unplanned downtime ↓ 32 %, saving $1.9 M in lost-batch margin. 5. Compliance-Agent – tracks regulatory changes and cross-checks them against current master data. Catches missing tax codes, expired certs or mis-classified hazardous goods—avoiding fines and rush rework. Bottom line: the agents surface waste you didn’t know existed, quantify the saving, and (with your one-click approval) can trigger the corrective action—turning raw data into real money, 24/7. Interested Parties,leave me a message here in LinkedIn and I will arrange a briefing Session to Present and Demonstrate

1

2 Comments

Recently got asked about our data modeling approach - first piece of advice isn't novel to those in the space: read The Data Warehouse Toolkit by Ralph Kimball & Margy Ross! Beyond that, here's our quick framework for handling complex datasets: **1. Large source systems with 50+ tables** Map the business process first. At Uber, we had thousands of tables - but only 20-30 powered 80% of decisions. Start with core tables, add complexity only where needed. **2. Schema changes** • Version control everything in dbt or similar modeling framework tool • Test in staging / development environments first • Document breaking changes religiously **3. New business requirements** Ask three questions before adding "just one more column": - Does this fit our existing model? - Who else needs this in 6 months? - What's the maintenance cost? Building clean data models isn't about perfection - it's about building something that evolves without requiring yearly rebuilds. What's your biggest data modeling challenge? #DataEngineering #Analytics

17

2 Comments

The Databricks Startups team answers founders’ most common AI questions—where to start without hype, how to stay compliant with sensitive data, what to build versus buy, how to cut hallucinations, and how to keep costs in check. 💡 Get step‑by‑step starting points, governance patterns for sensitive data, a pragmatic build‑versus‑buy lens, tactics to curb hallucinations, and cost controls that scale. 🔗 Full post: https://lnkd.in/ehRa3Yp5 #Databricks #DatabricksForStartups #AI #GenAI #MLOps

63

Just-in-Time UI: At Aligned Labs we frequently share reports about the data collection, fine-tuning and eval work that we do with clients.  They often need different views or refined analysis, which is why we built Whiz-bang Boom! Our new tool takes a CSV or JSON file and generates a shareable, interactive report. You can easily modify the report or focus the analysis with a simple prompt, and it creates a new revision. You can organize reports into spaces and customize those spaces with a simple prompt. Try it out. https://wizbangboom.com #businessIntelligence #AI

25

2 Comments

Document-centric tasks sit at the core of many enterprise, business, and government workflows. Search is important, but the real challenge is going beyond retrieval, enabling systems to reason across documents, verify facts, and handle multi-step information tasks. Great to see a new model from Databricks moving in this direction. I expect we’ll see many more models, including open-weight ones, designed specifically for document-driven workflows, an area with huge potential across enterprise and government use cases.

23

1 Comment

GPT-5 is here. What does that mean for the average market researcher? Many of the key improvements OpenAI has made with GPT-5 are upgrades to usability, so if you haven't gotten very far on your AI journey yet, there's never been a better time to jump in. But where do you start? Here are three things that will pay immediate dividends. 1. Learn the ChatGPT Platform Here are some things you can do within the ChatGPT platform: attach documents and spreadsheets, have ChatGPT generate in depth research reports on a specific topic with Deep Research, output documents and spreadsheets, design CustomGPTs to encode portions of your workflow and thinking, and many more. If you don't know what these things are, where they live in the platform, or how to use them, then there's a lot more that ChatGPT could be helping you with. 2. Test Your Assumptions about What AI Can Do Previously, the default model for most users of ChatGPT was not the best model, so unless you knew where to select your model and which of the confusingly named models was the most powerful (o3 is better than 4o...is that not obvious??), you may not know what AI is actually capable of. With dynamic switching between models now handled automatically, and a more powerful model underpinning everything, you'll get a much better idea of where and how AI can actually help you if you go and actually try out three or four representative tasks. 3. Make Your Requirements Explicit AI can't read your mind. While ChatGPT continues to get better and better at executing the technical aspects of a huge range of tasks, it still doesn't know certain critical details, like the unspoken priorities of the client, the norms in your company, or your sense of taste. If you want AI to take these things into account when generating outputs for you, you need to tell it. Communicating these things is not easy for most people, so when you're testing out the new model, give some thought to the gap between what you're assuming is obvious and what the model couldn't possibly know without you telling it, and start trying to articulate it. AI is, and will continue to get better. But it's a tool, and it's up to you to make sure that you learn how to use it effectively. Dig Insights

30

1 Comment

I’m moving from theory to practice on my #SyntheticDataJourney, developing benchmarks for synthetic outputs across established ML models and emerging LLM-based approaches. The first barrier we’ve uncovered isn’t model performance. It’s data structure. Data readiness is directly tied to the type of synthetic model you plan to apply. For traditional ML approaches, this is simpler. Structured numeric inputs often work, even when representing categorical variables. LLMs are different. They require context – A LOT of context. Not just metadata, but clarity around what each variable represents, how respondents interpret questions, how scales function, and how elements of the questionnaire connect. We can’t assume the model will “figure it out.” A clear mapping of question intent, response meaning, and structural relationships is table stakes. This has parallels to best practices with data governance and communication - if humans can quickly understand the data as a stand-alone product, LLMs are likely able to do the same. If you haven’t evaluated your organization’s data readiness for synthetic models, that may be the real starting point. Check out Tony Seale's post (link in comments) on AI readiness for another perspective. #MarketResearch #DecisionScience #DataQuality

16

2 Comments