Nadeem Qureshi

𝗔𝗜+𝗢𝗽𝗲𝗻 𝗦𝗼𝘂𝗿𝗰𝗲 𝗝𝘂𝘀𝘁 𝗗𝗲𝗹𝗶𝘃𝗲𝗿𝗲𝗱 𝘁𝗵𝗲 𝗞𝗻𝗼𝗰𝗸𝗼𝘂𝘁 𝗣𝘂𝗻𝗰𝗵 𝘁𝗼 𝗩𝗲𝗻𝗱𝗼𝗿 𝗟𝗼𝗰𝗸-𝗶𝗻 The platform wars aren't coming - they're OVER. And guess what won? 𝗔𝗜 + 𝗢𝗽𝗲𝗻 𝗦𝗼𝘂𝗿𝗰𝗲 𝗷𝘂𝘀𝘁 𝗰𝗵𝗮𝗻𝗴𝗲𝗱 𝗲𝘃𝗲𝗿𝘆𝘁𝗵𝗶𝗻𝗴! According to Forbes' latest analysis, the enterprise data platform market is set to nearly double by 2032, and here's the kicker. it's all about convergence with 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗗𝗮𝘁𝗮 𝗽𝗹𝗮𝘁𝗳𝗼𝗿𝗺. Data management, Analytics, AI Apps, governance, and orchestration aren't separate conversations anymore. They're one integrated powerhouse!. 𝗪𝗵𝗮𝘁 𝗧𝗵𝗶𝘀 𝗔𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗠𝗲𝗮𝗻𝘀 🎯 🔗 𝗜𝗻𝘁𝗲𝗿𝗼𝗽𝗲𝗿𝗮𝗯𝗶𝗹𝗶𝘁𝘆 > 𝗠𝗼𝗻𝗼𝗹𝗶𝘁𝗵𝘀 The new "minimum viable stack" isn't your grandfather's data warehouse. We're talking Lakehouse + vector databases + event streams, all standardized on open formats like Delta/Iceberg. Translation? Your data speaks every language, everywhere! ⚡ 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 𝗦𝗵𝗶𝗳𝘁𝘀 𝗟𝗲𝗳𝘁 Forget quarterly compliance reports gathering dust! Policies, lineage, and privacy are becoming code, embedded right in your pipelines and AI agents. Think "continuous compliance" instead of "panic-driven audits." 🤖 𝗔𝗜 𝗣𝗶𝗰𝗸𝘀 𝗬𝗼𝘂𝗿 𝗣𝗹𝗮𝘁𝗳𝗼𝗿𝗺 𝗡𝗼𝘄 Here's the uncomfortable truth: If your platform can't feed RAG systems and AI agents with fresh, governed context across clouds, your AI will hallucinate, and trust me, your CFO will definitely notice those phantom numbers! 𝗧𝗵𝗲 𝗥𝗲𝗮𝗹 𝗪𝗶𝗻𝗻𝗲𝗿𝘀 💡 Smart organizations are building composable platform with DataGOL where: - AI is a first-class citizen (not an afterthought!) - Open source is the safety rail (not a science project!) - Data flows like water, but governance flows like code 𝗕𝗼𝘁𝘁𝗼𝗺 𝗹𝗶𝗻𝗲: The future of enterprise data isn't about picking the "right" vendor, it's about building systems so open and composable that you never have to make that bet again, that's DataGOL Ready to embrace the open + AI future? #EnterpriseData #AIStrategy #OpenSource #AIStrategy #CxO #EnterpriseData #DataStrategy

84

2 Comments

An interesting data point from the NiCE Cognigy $1 billion deal - it's apparently the largest pure play #AI company exit in European history - this according to Cognigy CEO Philipp Heltewig, revealed on today's analyst briefing call. Heltewig believes this will energize the European AI startup ecosystem. A billion dollar exit tends to get folks motivated for the art of the possible. #startup #europeanAI #AIstartup #CX #agenticAI #conversationalAI

60

This is why you still got skewed analytics data: most #GA alternatives and even #Clarity are doing this wrong (proof attached): → they will measure pagehits, even if they aren't there → for example when a page was only prerendered, but never visited You might think: "I didn't implement prerendering (let alone #SpeculationRules) yet, so I should be good" But even Google #SERP might prerender your pages (there are free tools available to check this) Attached is an example that I ran into: 1️⃣ Chrome #DevTools adds a convenient dropdown at the top right for prerendered pages 2️⃣ Allowing you to debug prerendered pages without already visiting them 3️⃣ You can see I switched to a "prerender"-ed page and viewing its network activity (outlined in red) 4️⃣ Over there, we can see Clarity JS is loaded 5️⃣ And it already starts collecting data too (outlined in red) I saw many tools doing this wrong, including Piwik PRO / Matomo, Simple Analytics, Fathom Analytics et cetera. Amongst those that I tested, Google Analytics is the only one doing this right. In your case, do other tools show more pageviews compared to GA? Or actually less? (which could happen in case of bfcache)

63

34 Comments

Since we launched, our mission at Google’s Data Commons has been to organize the world's public data and make it universally accessible and useful. But key public data remains scattered. For example, let’s say I’m trying to understand whether low income nations are properly funding healthcare internally. Analysts need to search government funding records, international organization records (World Bank, OECD, WHO), funder nation records … , then track down assumption, clean, and hopefully stitch together disparate datasets manually … or in most cases, give up. We're changing that. Today, we are launching the Data Commons Model Context Protocol (MCP) Server, a new way to make all of Data Commons’ interconnected public datasets instantly accessible and actionable for AI developers, data scientists and organizations everywhere. This capability further supports the larger ambition of Data Commons: using real-world statistical information as an anchor to help reduce Large Language Model (LLM) hallucinations. -> The game-changer? You don't write custom code or wrestle with APIs. You just ask a question in plain language, and MCP connects the data for you. It shifts data-driven decisions from complicated to practical. -> Real-World Impact: Global organization ONE is already leveraging this with the launch of the free ONE Data Agent. This first-of-its-kind tool uses AI and human expertise to unlock trusted health financing data in seconds—a powerful new resource for those shaping policy and driving change. Let’s go back to our example above. By asking ONE’s Health Finance tracker “In Nigeria, how much do households spend out of pocket on health compared to the government, per person”, we can quickly see that Nigerians spend 5x out of pocket compared to their government. In high-income countries, the government pays 3x more. The data shows 2.5 B people live in countries where they are spending more out of pocket and *not one Low Medium Income Country (LMIC)* has the government spending more. This, when all that is needed is $86 USD per person on health to change this trend …  Now we have access to the data, and hopefully can advocate for the right aid and the right change. See the full details in our announcement on the Google Developer blog. We’d love to hear your thoughts! -> https://lnkd.in/efr_6RC7 And see the Health Financing tool from the One Campaign: -> https://lnkd.in/eVjzKbHV   Finally, this impactful work took a village. A special shout out to Keyur Shah from the Data Commons team and Jorge Rivera from The ONE Campaign. These amazing souls worked nights and weekends on risky, early tech to make this happen. In addition, kudos to Marta García Ferreiro, Dan Noble, Christie Ellks, and the Data Commons team as well as David McNair, Micaela Iveson , and The ONE Campaign who are the ideal, inspirational partner to work with. #DataCommons #TechLaunch #PublicData #TeamWork #OneCampaign #Africa #Data

172

13 Comments

I started this autumn wanting to learn more about machine learning and using LLMs in BigQuery. But where do you begin? Read all the documentation? Follow a step-by-step guide? Watch a video? I’d say: find yourself a problem. By defining a few real-world problems you think you’ll need to solve with the tool you want to learn, you’ll get the right starting point. Once you have a problem, you can begin trying to solve it — and through that, you’ll see the gaps in your knowledge that need to be filled. I managed to define a few problems I want to solve with machine learning, and I found some solutions using the AI and ML features in BigQuery, with both strengths and limitations. The goal is to present these at this autumn’s upcoming Digital Analytics Meetup – stay tuned! In the meantime: find yourself a problem.

49

5 Comments

Many of this year's conversations at Cannes Lions International Festival of Creativity highlighted the importance of data and AI in driving creativity and inspiring change in the advertising industry. At the same time, however, marketers have shifted their focus away from truly understanding consumers and are now treating them more as data points for building their AI models. And what about the role of AI in consumers' lives? How has AI changed their consumption of content, purchase behavior, and decision-making? 💡 In their search for quick and easy answers to their questions, consumers are increasingly using LLMs over traditional search engines. This means that it is now more important for marketers to think about 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐯𝐞 𝐄𝐧𝐠𝐢𝐧𝐞 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐚𝐭𝐢𝐨𝐧 (GEO) vs traditional SEO. 💡 Consumers are increasingly 𝐫𝐞𝐥𝐢𝐚𝐧𝐭 𝐨𝐧 𝐀𝐈 𝐭𝐨 𝐜𝐮𝐫𝐚𝐭𝐞 𝐭𝐡𝐞𝐢𝐫 𝐝𝐞𝐜𝐢𝐬𝐢𝐨𝐧-𝐦𝐚𝐤𝐢𝐧𝐠 across a variety of categories. These categories range from ones traditionally associated with longer purchase cycles (e.g., financial services, education) to ones where they are open to fully delegate their decision-making to AI agents (e.g., CPG, travel). 💡 Consumers don't follow traditional moments of truth and are instead increasingly making their decisions based on the 𝐢𝐧𝐬𝐭𝐚𝐧𝐭 𝐦𝐨𝐦𝐞𝐧𝐭 𝐨𝐟 𝐭𝐫𝐮𝐭𝐡, where the whole funnel collapses from minutes, days, and months to sometimes mere seconds. 💡 Consumers are more likely 𝐭𝐨 𝐦𝐚𝐤𝐞 𝐮𝐧𝐩𝐥𝐚𝐧𝐧𝐞𝐝 𝐚𝐧𝐝 𝐢𝐦𝐩𝐮𝐥𝐬𝐞 𝐩𝐮𝐫𝐜𝐡𝐚𝐬𝐞𝐬, as with the amount of information they now have at hand, traditional purchase planning becomes extremely complex (due to both the amount of information and number of choices available). Hence, consumers are driven more by 𝐧𝐞𝐞𝐝 𝐬𝐭𝐚𝐭𝐞𝐬 rather than specific products and categories. Which consumer trends driven by the evolving role of AI do you observe in your categories? -- Thank you P&G Alumni Network - Official Page for the interesting conversations this summer! It's always inspiring to brainstorm and share perspectives with my amazing ex-P&G colleagues. ✨

53

23 Comments

How to use the right #AIframework to make sure your agents are actually aligned with enterprise? Sense about it: You might be tempted to jump straight into building agents, chasing the hype and expecting fast results. But soon you hit walls. Your agents fail to run, outputs feel inconsistent, and you find yourself debugging errors instead of seeing real value. Why does this happen: Stop always blaming Frameworks, MCP, and LLM for bad #AIAgents, You need to really understand how an agent actually takes action. The Following Framework can be help for every AI Agent Builder. It breaks down an agent’s workflow into 4 natural steps: Think: Just like our eyes and ears help us notice cars before crossing a road, agents must perceive their environment. In practice, they gather signals from your prompt, web data, documents, or from MCP Servers to know what’s happening. Plan: Think of how you plan your route before a trip — checking maps, traffic, and timing. Agents do the same by reasoning with methods like CoT, GoT, or Reflexion to decide the best sequence of steps. The Best example can be seen in CUA Agents like Manus and OpenAI Operator. Action: Just as you physically walk, drive, or call someone to get things done, agents take concrete actions. In reality, that means calling APIs, generating reports, updating systems, or scheduling tasks. Reflect: Like reviewing how your last team meeting went to identify what things needed to be changed for the new quarter or performing sprint retrospective. Similarly, Agents analyze feedback and refine their approach, improving results over time. This continuous loop is what makes agents evolve from simple automation to true autonomy. And if you’re building your own, remembering this gives you a mental checklist, - Did my agent collect from the right inputs? (Sense) - Did it map out a clear sequence of steps? (Plan) - Did it carry out the right tools or actions? (Action) - And did it learn from the results? (Reflect) If your agents are not within the loop, You will get a mental feedback of fixing that particular component rather than rebuilding or scrapping the entire project. Here’s the key: You’ll see why most enterprises fail at scaling agents; they skip the cycle and overestimate what their agents can do. This will show you how to design agents with a scalable framework at the core, ensuring they’re robust, compliant, and enterprise-ready. #EnterpriseAI #AIAgentEngineering

61

3 Comments

🛍️ The next big channel in commerce won’t be a website, a marketplace, or an app. It will be AI agents that buy, book, and solve for your customers. 🚀 From Physical to Agentic: The Next Great Commerce Shift In our latest blog in the Agentic Commerce series, I explore how commerce has evolved: 1️⃣ Commerce 1.0 – Physical to Digital: The rise of e-commerce, mobile, and omnichannel. 2️⃣ Commerce 2.0 – Platform Revolution: Marketplaces unlocked infinite availability by orchestrating third-party supply. 3️⃣ Commerce 3.0 – Agentic Commerce: A new channel emerges — AI agents that anticipate, compare, and act. They don’t just inform customers, they become the interface for shopping, service, and transactions. 📌 Key takeaway: In the Agentic era, success won’t be about having the best storefront — it will be about being agent-ready, able to connect instantly with AI agents that shop, book, and solve for your customers across this new channel. 💡 That’s exactly why we’re building Mirakl Nexus — to enable retailers and merchants to own the future of commerce. 👉 Read the full article here: https://lnkd.in/ef7jkHuG #AgenticCommerce #AI #MiraklNexus #EcommerceInnovation #FutureOfCommerce

90

Several strong 2025 articles on training embedding models for retrieval (and other tasks, as KaLM article), with a focus on step-by-step 'recipe'-style guides that detail how the models were trained, what techniques worked, and what didn’notm, rather than just reporting state-of-the-art results KaLM-Embedding-V2: Superior Training Techniques and Data Inspire A Versatile Embedding Model https://lnkd.in/g3kcKV-m CSMF: Cascaded Selective Mask Fine-Tuning for Multi-Objective Embedding-Based Retrieval https://lnkd.in/gqrTAHt8 Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models https://lnkd.in/g_XwPtAb

130

6 Comments