Occam

17.02.2026 14:30 HiPPOs still hand down most decisions in a company – even when surrounded by piles and piles of reports with metrics galore. Marketers still use the “funnel” to imagine and allocate budget, people, actions – even when there is multi-decade data that the funnel is a lie. Company after company misses trends and goes kaput – even after investing in multi-million dollar data projects to build clean rooms, unified consumer view cloud-based business intelligence platforms. ☹️ The core challenge isn’t data scarcity; it is insights latency. Which impacts your ability to follow my advice to deliver IAbI, and not data. Here’s the traditional analytics workflow being practiced in your company: 1. Report Generation. Hours and hours of standard reports/dashboards. 2. Manual Analysis. Hopefully, laborious segmentation of known knowns missing subtle non-intuitive patterns. AKA High-dimensionality data challenge. 3. Insight Extraction. Identifying the most important findings, extracting context from Marketers, Finance, Sr. Leaders, creating presentations. 4. Exec Last-Mile Barrier. Data competing with other priorities, insights missed, misinterpreted, translation to action challenging – to say the least. A process that is fundamentally reactive and linear. It struggles with tracking hundreds of variables per user engagement, non-linear patterns . Humans are ill-suited to find a needle in the haystack – identifying truly significant anomalies or emerging trends within massive datasets. ☹️ ☹️ It does not have to be this way. anymore. It is time to hand control of Marketing Analytics over to AI! Online, offline, digital, everything. 😊 AI can act as a force multiplier and overcome the above limitations. A. Pattern Recognition at Scale. Machine Learning algorithms are awesome at finding complex, non-linear relationships and hidden clusters within massive high-dimensionality datasets. B. Automating the Mundane. As I’m sure you’ve seen in your use of ChatGPT, Qwen, others, AI can automatically generate insights from routine data, it can flag anomalies instantly, and surface the most statistically significant changes. C. Predictive Power. Rather than reactive, what happened, AI is exceptional at what’s likely to happen, thus solving the insights latency. It is worth noting it can personalize this at a massive scale – dynamic segmentation, tailored experiences, value something impossible with manual rules. AND, do this uniquely for the needs for every human in your org! D. Continuous Learning. AI’s real superpower. AI models adapt as new data flows, they constantly refine their understanding of user behavior and system performance – at a massive scale. Handing control of Digital Analytics over to AI achieves this profound shift: From Analyst-as-reporter to Analyst-as-strategist. From data puking and insights hunting to validation and activating action. 😊 😊 😊 It is time to rebuild analytics from the ground up. You’ll remember I originated the 10/90 rule of Analytics 20+ years ago. “If you have $100 to invest in smart decisions, invest $10 in tools and implementation, invest $90 in humans who will analyze the data!” Here’s my new 10/90 rule for success via investment in Analytics: “If you have $100 to invest in smart decisions, invest $10 in brilliant human analytical strategists, invest $90 in AI activation.” In fact, over time the $100 is likely to reduce to $80, then $70, and maybe less… While the quality of decisions, the scale of intelligence and automation, will exponentially increase. Incredible, no? Let’s learn how to activate this immense value. This blog post was originally published as Premium edition #492 of my newsletter. Weekly, I share actionable insights and hidden patterns to stay at the bleeding edge of Marketing, Analytics, and AI. Sign up for TMAI Premium to accelerate your career trajectory. Activating AI Power. AI is not yet AGI , and certainly not SGI . Today, activating the awesomeness above will take human grit, intelligence, and persistence. Things won’t be perfect. Your True North: Somewhat failing to activate my recommendations is 25x better than your present. And, as a bonus, you’ll be ready for AGI. I have ten specific implementation ideas for you to turn your digital analytics over to AI. I hope they’ll spark a dozen more in your team. 1. Predictive Analytics via Propensity Modeling. Impact Potential: Transformational. Human-powered digital analytics tells us who converted. AI can tell us who will convert! There are thousands to tens of thousands of humans on your site, using your apps today. Instead of spreading your budget, attention on all of them, you can focus on high-propensity humans. ML algorithms thrive on pattern recognition across hundreds of variables, and thus identify subtle combinations of behavior that signal conversion readiness . IMPORTANT: Unlike rule-based systems , AI models consider non-linear relationships and interaction effects between dozens/hundreds of variables for a more brilliant understanding of human intent and what will happen next. Framed simply: What is the exact probability THIS human will convert/upgrade/churn in the next N days? AI Approaches and Algorithms to explore, stress test, and embrace: Gradient Boosted Machines . Currently, the gold standard for tabular data prediction. These algorithms excel at conversion prediction by combining many weak predictive models into a highly accurate ensemble. Random Forests. I have loved using RF when I have a need to understand feature importance. Ex: Which behaviors most strongly predict conversion? Neural Networks. The grandpa of AI. For massive datasets with complex, nonlinear relationships, deep learning architectures can uncover patterns other models miss. Survival Analysis. Good old statistics. Predicts not just if, but when a user will convert, enabling perfectly timed interventions. Each business is unique; you might use a couple from above, just one, or all of them to solve different propensity modeling opportunities . One of them you should have activated in the next six months. Practical Example. From my experience: A propensity model using approx. 90 behavioral features . The model scored each user in real-time, allowing the ecom COE to: Serve dynamic offers to high propensity users. Adjust bid strategies for retargeting ads based on conversion probability. Identify “at risk” users who showed high intent but did not convert . Potential Outcomes For You. Looking across my work on three continents, focusing on ecommerce: A. 35% – 60% improvement in Conversion Rates for the targeted segments. B. 20% – 35% reduction in acquisition costs due to more efficient ad spend. C. Not easily quantified qualitative impact of shifting from reactive to proactive marketing. Over the last three years, Propensity Modeling has been my most monetized, highest now-potential, game-changing action in handing over digital analytics to AI. Every quarter you don’t activate it, you are falling two to three quarters behind. 2 Advanced Customer Segmentation. Impact potential: High. You should not be surprised that this is so important. My blog was born May 2006; this is from then: Excellent Analytics Tip#2: Segment Absolutely Everything. Most analytics teams segment users by demographics or broad behavioral categories . These segments are often too broad, and miss thousands of nuanced behavioral patterns. Creating more relevant, precise, sophisticated segments manually is extremely time-consuming and limited by human bias, human knowledge . Unsupervised learning algorithms specialize in finding natural clusters in data without predefined categories. They can process dozens of behavioral dimensions simultaneously to identify segments that are statistically distinct rather than intuitively appealing. They can get to the unknown unknowns – hidden well below the human capability surfaces. AI Approaches and Algorithms to explore, stress test, and embrace: K-means Clustering. Thousand-year-old workhorse algorithm for segmentation and grouping users based on behavioral similarity across multiple dimensions. DBSCAN. Full, cute, name: Density-based spatial clustering for applications with noise. Instant love, no? 😊 Specifically awesome for identifying outlier segments or detecting novel user behavior patterns. Gaussian Mixture Models. Few models handle ambiguity better than hard clustering approaches, when segments overlap probabilistically. Hierarchical Clustering. Fifteen times a day, Analysts have to drill from broad categories to highly specific micro-segments. Segmentation trees created by hierarchical clustering are a perfect solution. There is immense untapped potential to be extracted by applying unimaginable scale to segmentation via AI and algorithms. Practical Example. Tying this back to an example from my 2006 blog post, but applied recently to a B2B client. We applied clustering to session data across 28 behavioral dimensions. Instead of the standard free trial users segment, the algorithms identified: Segment A: Feature explorers, who try many features quickly. Segment B: Cautious adopters, who read documentation before acting. Segment C: Social validators, who always check testimonials first. Segment D: Price-sensitive evaluators, who immediately navigate to pricing. Tying these segments to outcomes, working backwards to influence them, allowed the team to customize onboarding of the trial experience in real time, the content and flow of the product, and, obviously, subsequent messaging to dramatically improve activation rates! Potential Outcomes For You. Reflecting on my clients and work: A. 25% – 50% improvement in Conversion Rates from behavioral targeting based on algorithmic segments. B. 60% to 75% reduction in analysis time dedicated to customer segmentation. C. Not easily quantified qualitative impact on customer joy and company revenue from a deeper understanding of potential customers and their behavior. Big picture: What took weeks of manual cohort analysis , now happens automatically . 3. Voice-of-Customer Integration with Behavioral Analytics. Impact Potential: High. Another one of my old web analytics dreams has come true. Early readers will remember my, at the time, revolutionary Trinity model for Analytics. 19-years later, I can AI it! Survey responses, support tickets, chat transcripts, call center voice recordings, and social media mentions live in separate systems from behavioral analytics. Analysts struggle to connect the why with the what . This leads to incomplete understanding of user motivations, frustrations, and unmet needs. Solving this at scale with humans is futile. One of the key leaps of modern AI is multi-modality – the ability to understand text, images, voice, and video at unimaginable scale and incredible precision. Multimodal AI systems can process both structured behavioral data and unstructured text/voice data simultaneously. Advanced embedding techniques allow algorithms to find connections between language patterns and behavioral patterns at scale. Sentiment analysis has evolved beyond simple positive/negative classification to detect specific emotions, urgency, and intent that, a blessing for us, correlate with behavioral outcomes. AI Approaches and Algorithms to explore, stress test, and embrace: Multimodal Transformers. Great for processing text and behavioral data in a unified model architecture, with cohesive, understandable outputs – at scale. Cross-modal Retrieval. Helpful in finding behavioral sequences that correspond to specific feedback themes, assessing their quantitative relationship. Advanced Sentiment Analysis. Taking our positive/negative past approach to significantly higher accuracy and detailed why patterns by detecting frustration, confusion, excitement, and uncertainty. Applications across every part of the business . Topic Modeling with Behavioral Correlation. A data setup problem to overcome, but then at scale drive discovery of which discussion topics correlate with specific actions or drop-offs . Emotion-Action Mapping. Above hinted reflective analysis, now you switch the view to predictive analysis. Connect expressed emotions with subsequent behavioral patterns – driving proactive actions by ensuring you don’t lose a logistics provider or an employee quitting or a massive B2B client not renewing their contract. Practical Example. This one’s from a pal, in Canada. An ecommerce platform integrated the site’s chatbot data with behavioral data and put in place an AI model to analyze it. Discoveries: Positive mentions of sustainability in chats correspond with 3.4x higher lifetime value. Discovered emerging complaint patterns about a new feature, three months before negative NPS scores. Users expressing “size uncertainty” in chats have an 82% higher return rate. Potential Outcomes For You. A. 8 – 12 points improvement in NPS scores. B. 20 – 25% reduction in fails from real-time interventions put in place. C. Not easily quantified qualitative impact on product development from 360-degree customer understanding from connecting the why with the what systematically. The Profitable AI-Analytics Journey Continues. In TMAI #493 and #494, I’d shared additional super exciting ideas to deliver transformative profits via AI-Powered Analytics. Additional activations included: 4. Behavior Targeting & Intelligence . Impact potential: Transformational. 5. Natural Language Processing for Unstructured Data. Impact potential: High. 6. Anomaly Detection and Automated Insight Generation. Impact potential: High. 7. Predictive Customer Lifetime Value Modeling. Impact potential: Transformational. 8. Real-Time Pricing and Offer Optimization. Impact potential: High. 9. Intelligent “Liquid” Merchandising. Impact potential: Medium. Give all of the above is true today, I predict that the current type Analyst role will cease to exist over the next 18 or so month. In TMAI #495, I laid out a framework that outlines what the Analyst role will be in Jan 2028, and how you need to get ready for it starting now: TMAI #495: Analyst 2028: S.H.I.F.T For Relevance. If you are a new TMAI Premium member, please email me for the series above. If you are not, grab an annual Premium subscription here. Bottom line. The integration of AI into Analytics represents the most significant shift in our field since its birth as a science. The organizations that will thrive in the coming years aren’t those with the most data, most Analysts, most spending on Analytics. They will be the ones who can extract the most insight from their data with the greatest speed. i.e., reduce insights latency and increase automation. AI and advanced algorithms provide the tools to make this possible, transforming analytics from a practice of historical reporting to one of predictive intelligence and prescriptive optimization. Carpe diem. PS: It is only appropriate that I share with you an AI-generated summary visual of this blog post! For your slides… The post Bye, Bye Human-Powered Marketing Analytics. appeared first on Occam's Razor by Avinash Kaushik.