Ryan Praskievicz

My team at West Monroe has been building an Agentic AI workflow tool to tackle some of the biggest / hardest problems in the data engineering and analytics space. We're creating solutions where we have found GenAI is the most effective: > converting and generating code > generating data models > analyzing and assessing legacy code > overall enabling data engineers and data scientists Check out our latest video showing an Agentic conversion workflow, that helps migrate legacy code from tools like MS SQL, SSIS, SAS, Snowflake, Redshift, Teradata, ... -> to Databricks! *News Alert* Forbes (via Gartner) just spotlighted Agentic AI as the #1 imperative for companies in 2025. Link in comments. What we've learned on this journey building Agentic AI is that this is a skill, a muscle that your teams build. It's all about creativity and problem solving. I can confidently say that West Monroe has more people working on building these skills and applying these techniques than anyone I've run into in the market (by % of team). I'm VERY excited where we're headed next year! Link to our latest video: https://lnkd.in/gvh7UjYi Team credit: Thomas Zhang, Julia Waggoner, Lindsay Michelini, Meghan Patel, Zafeerah Siddiqua, Doug MacWilliams, Robin van Overbeek, Patrick Miller, Sam Atkinson, Brad Ptasienski, Scott Rozen-Levy, David Choi, Lana Buseman, Cory Chaplin, Eric Johnson (and so many others who have helped move this initiative along)! And partner/advisor credit to our friends at Databricks too: Franco Patano, Jeffrey Lipkowitz, Kaylee Harrison, Swaminathan Venkatesh, PhD - appreciate your guidance and support on this journey!

64

1 Comment

I think there are overlooked implications of auto model selection in ChatGPT, which seems to be rolling out slowly. First, this can massively impact OpenAI’s profitability if it can direct more queries to smaller models without adversely affecting quality. Second, this can prevent people leaving ChatGPT for other apps—like those running local models—if it’s just easier to stay in one app and get a faster model or smarter model automatically, rather than use small local models for some things and switch to ChatGPT for others. It’s even possible that ChatGPT could in future run local models on properly equipped devices for the easiest queries, which would completely avoid hitting OpenAI’s servers for those queries. So, it’s a small feature, but I think one with big implications.

15

7 Comments

AnalyticsIQ team members Dan Lescano and Mike Hattub are heading out to Chicago, Illinois for the upcoming P2PI Retail Media Summit from June 25th through the 27th. P2PI's Retail Media Summit brings together the commerce marketing community to help brands and retailers navigate the complicated, rapidly shifting media landscape. It's the perfect opportunity to connect with other professionals, learn about new emerging trends, and hear from some of the best thought leaders in commerce marketing. If you plan on attending, be sure to keep an eye out for Mike and Dan so you can say hello! #RetailMediaSummit #RetailMedia

20

1 Comment

I have a great time spinning up our own delivery data in Omni, to the point that it deserved a write up on how easy is to use and the instant insights that we get from it! Check it out in our blog!

22

1 Comment

As these models continue to get better - we really need to optimize for inference. It's the only way to allow for large scale adoption of these models for enterprise business application. #openai #gpt #genai

2

💡 Working with DTC Brands across industries, AI Is Top of everyone's Mind! 💡 I’m frequently asked about AI implementation in their respective functions. According to the latest estimates from McKinsey, 65% of companies worldwide have implemented AI models—either in the pilot or production stages. 🌍🤖 But with a plethora of AI tools and platforms out there, businesses can feel overwhelmed. How do you choose the right solution for your functional and infrastructure use cases? 📝 This McKinsey report offers a great framework for CDOs and Chief AI Officers to scale GenAI successfully. Here are my key takeaways: 1️⃣ Operating Model: Develop a model that aligns people, processes, and technology to deliver the most value. In addition, build continuous Measure, Test, & Learn cycles. 2️⃣ Component-Based GenAI: Focus on critical components in the early phases—think APIs for data inputs, vector databases, Agents, Hallucination Checkers, and more. 3️⃣ Team Org Structure: Whether you choose a centralized or decentralized AI strategy depends on your organization’s structure and communication practices. 4️⃣ Risk & Compliance Management: Establish a baseline risk assessment and compliance framework, revising it as new risks are identified. check out the full report from McKinsey here: https://lnkd.in/gbnBYefR #AI #GenAI #DataScience #DigitalTransformation #McKinsey #Innovation #Leadership

12

I've published a post on Next Basket Recommendation along with Suman Ghosh on the Epsilon Engineering Blog, where I dive into how machine learning algorithms can predict what customers are most likely to purchase next. 🎯 If you're in retail, e-commerce, or curious about how to boost customer experience through predictive models, this article is for you! 📊 👉 Read the full post here:

39

2 Comments

The Tableau world is talking about the new 2024.2 release (best release in years imho), highlighting all the great new features like the Viz Extensions (love them)...But I'm a simple data guy: What I love most about the new release is that you can now format parameters individually. I'm serious: After Tableau gifted us with the copy and paste function a few years ago (lol), we can now finally format parameters independently of each other – and it only took 9 years to implement this community idea 😅 Tableau I have a crazy idea: What if you could change the font color of navigation buttons to any color instead of being limited to 48 colors? I know you can do it!

9

#eli5 #federatedlearning Federated Learning is a cool way to teach computers new skills without having to move all the data they learn from into one place. Imagine a bunch of kids in different playgrounds, each playing with their own toys and learning how to build sandcastles. Instead of bringing all their toys to one playground to share what they've learned, each kid learns to build a sandcastle in their own playground. In Federated Learning, each "playground" is actually a device like a smartphone or a computer. These devices each have their own data (like photos, messages, or app usage), but instead of sending all this private data over the internet to a big computer (which could be risky for privacy), the learning happens right on the device. Here’s how it works: ----------------------- 1. Starting Off: A central system (like a teacher) sends out the same basic lesson plan on how to build a sandcastle to all devices. 2. Learning Locally: Each device (like each kid in their playground) practices building their sandcastle using their own toys (data). This way, they improve the lesson plan a little bit based on what works best with their toys. 3. Sharing Improvements: Once they figure out some cool tricks, instead of sending their toys to the teacher, they just send back their new building tips (a summary of what they learned). 4.Updating the Lesson Plan: The central system gathers all these tips from different devices and makes the lesson plan better. Then, it sends this updated plan back to all devices. 5. Repeat: This process repeats, with the lesson plan getting better and better each time, without ever needing to move the toys from their playgrounds. This method is super useful because it protects everyone’s privacy while still helping all devices learn better and faster together. It’s like having a giant, shared learning experience without any risk of losing or exposing your own toys! #datascience #ai

5

Huge win on #Quarterly Earnings for Palantir Technologies (#PLTR +13.26% pre-market) If you were on the #Earnings #Call, you might have heard Shyam Sankar talk about their core product: #Data #Ontology. The CTO describes how Ontology gives Palantir an unprecedented 10+ year advantage in the #AI space, not for competing with the #ChatGPTs models of the world, but by actually providing enterprise framework/software to deploy 3rd party AI models beyond your design, content writing and #meme producing teams. We are talking about creating your JARVIS from Iron Man; scaleable to every end user in your company - with access points, seamless resource sharing, and actionable analytics with tailored cross-department recommendations. And it all starts with Ontology. We've written about Data Ontology and how this can be deployed in a Marketing environment. Granted this is child's play compared to PLTR's Apollo, Foundry, and Gotham offerings; yet this is a Day 1 thinking approach to a complex and ever-evolving endeavor. Let me know what you think: https://lnkd.in/evtcWeTb

2

One of my favorite independent AI benchmarking services has come out with an extremely useful guide to choosing an AI chatbot. It can be hard to keep up with all of the AI tools and models. Artificial Analysis is becoming an essential tool to help, and they've just released their chatbot comparison here: https://lnkd.in/g2DJVvBa This is really comprehensive, and can also help you zero in on the best tool for your most frequent use cases. Worth checking out.

16

As more targeting and bidding strategies are automated inside black boxes with fully "AI-optimized" campaign types like Performance Max and Advantage+, a marketer's new job is going to be ensuring the right guardrails and filters are in place so the ad platforms are not just targeting the lowest hanging fruit and charging advertisers for conversions that would happen anyway. Nick Graham, Senior Consultant at Kepler, wrote one of the best overviews for The Drum on why marketers need to adopt this mentality, as well as specific suppression strategies they should consider to boost their campaigns' performance and spend efficiency. https://lnkd.in/g5cuh7kj "While this strategy isn’t new, it is perhaps the forgotten more sensible sibling of audience targeting". "Historically, there’s been a tendency for advertisers to go after heavy buyers. But this group is not necessarily going to drive incrementality. The suppression method offers a reverse way of targeting relevant audiences...avoiding the need to target existing customers or those who have already purchased a product or service."

2

The AI machines that we've become accustomed to via large language models from places like OpenAI and Anthropic require an incredible amount of data to train them. For the past decade or so, this data has been easily accessible - for free - on the open web. Not so much anymore, as content creators and publishers have become wise to companies whose raison d'etre is to eventually replace the humans making (or, at the very least, squeeze the human creativity out of) the very content they're using to train their models. This has led to the rise of synthetic data, but an overreliance on this method could lead to model collapse - just ask the Spanish Hapsburgs. We round out the week asking what the MTA actually did at the Grand Central 4/5/6 entrance all summer and find out where all of George Strait's exes live. This week's newsletter is live! If you like the (100% human-created) writing I put out, think about subscribing to get this in your inbox every Tuesday!

6

Finally, Evidence That Managing for the Long Term Pays Off Data and AI teams are often faced with the sacred dilemma of ROI. While value creation should indeed be the guiding principle for data, this must be considered in the medium and long term. Data, along with generative and non-generative artificial intelligence, is at the heart of business transformation, their economic models, and also innovation—for example, in design, new materials, chemistry, pharmaceuticals, and commerce, among others. The Harvard Business Review recently published a comparative analytical study by McKinsey, highlighting companies with a long-term vision versus those with a short-term outlook. This highly inspiring study is based on a sample of a large number of medium and large enterprises over a period of more than 10 years. I encourage you to read the full version on McKinsey's website, which I hope will provide CDOs with the insights needed to secure the long-term investments essential for the rapid digital transformation that companies of all sizes and industries are facing. Enjoy! #businesstransformation #innovation #investment #longterm #shortterm #data #AI https://lnkd.in/d59ptN_q

33

4 Comments

Great article by my colleague Aaron Schroeder. It talks about the interdependence of AI literacy and data literacy. “As you increase your organization’s data literacy, it will help bolster your ability to recognize and apply AI to new areas, and communicate those to the teams and individuals who will use them. Meanwhile, higher AI literacy drives better use cases and applications for AI, which in turn can help produce more valuable data.” Have a look and see how Aaron suggests increasing data literacy at your company!

4

For as much energy as #advertisers and agencies have put into supply path optimization over the years, the #programmatic supply chain is still in need of cleaning up. See the recent examples of made-for-advertising sites, Forbes and Colossus calling into question ad buyers’ abilities to control what inventory they’re purchasing programmatically. This helps to explain why Digitas North America is planning to take a stricter approach to how it sources inventory. In this piece by Tim Peterson, we speak to Leah Askew.

4

Cue the music... 🎸 It's the FINAL COUNTDOWN! Google's Universal Analytics has one week left before it's gone....and your data's going with it...unless you export. Are you ready? Here's 15 tips from my AWESOME Team at Seer Interactive to help you prepare. #GoodbyeUA #HelloGA4 #r_u_ready?

11

1 Comment