Mustafa Ibrahim

What about #Predictive #Segmentation? In my last post, I discussed the power of historical data in segmenting customers using methods like RFM. But what if we could look ahead, not just at what customers have done, but what they’re likely to do next? That’s where Predictive Segmentation steps in. Let’s clear up a common misconception first: predictions are often labeled as the latest “AI” breakthrough, but in reality, predictive modeling is rooted in Machine Learning (ML) techniques that have been around long before the AI boom. While predictive segmentation does leverage AI, it’s not a new technology in our domain. Here’s how predictive segmentation can transform your CRM strategy: 1. Predicting New User Behavior Predictive models can analyze user actions — even those who haven’t made a purchase yet — to forecast future behavior. By examining non-purchase actions like browsing patterns or content engagement, businesses can segment users who are likely to convert into buyers, long before they do. This insight allows us to create proactive engagement strategies that are personalized and timely. 2. Predicting Existing Customer Behavior For your existing customers, predictions can help identify those who are at risk of churn. For example, a Champion customer today could be on a path to becoming a lapsed customer based on recent actions like lower interaction frequency or reduced spend. Without predictive models, detecting these shifts would be much harder, if not impossible. By forecasting future behaviors, businesses can intervene with retention strategies before it’s too late. 3. Predictive Customer Profiling: Finding Your Ideal Customers Beyond segmenting based on behavior, predictive models can help businesses define their ideal customer profiles (ICPs) by analyzing demographic data. These models use data such as age, location, income, and other characteristics to find patterns that reveal your best customers — those most likely to engage, convert, or become repeat buyers. This helps businesses tailor their marketing strategies toward the right audience, ensuring efforts are targeted at customers who fit their ideal profile. Predictive profiling also helps identify new customer types, expanding your reach to similar profiles you might not have initially considered. In essence, predictive segmentation isn’t just about understanding past behavior but anticipating the future, allowing businesses to stay ahead of their customers’ needs. Stay tuned for my next post, where I’ll dive into how to implement predictive models into your everyday customer-centric strategies. #PredictiveSegmentation #CRM #MachineLearning #CustomerInsights #AI #SupervisedLearning #UnsupervisedLearning #CustomerProfiling #CustomerExperience

37

4 Kommentare

“How do we know if the model is right?” That’s one of the most important questions in marketing mix modeling. In MMM, we don’t just care about predictive accuracy. We care about getting a correct read on the true incrementality of our marketing dollars. If we spend an extra $10,000 on a channel like Meta, how many additional conversions will that drive regardless of tracking? Unfortunately, it’s impossible to know for certain what the true value is. No one can go look up that number. That is the core problem we’re up against. So when we think about how to validate an MMM, here’s how we do it: 1 - Start with a causal model based on marketing science When doing causal inference, you always need to start with theory. We start with a causal diagram outlining how marketing scientists believe marketing works in the real world, and we use that to determine our statistical model. 2 - Parameter recovery If we generate a data set where we know the true Facebook marketing performance (because we created the data), can we run the model and get results back that match the truth when we run simulated data through the model? 3 - Compare model results against the truth where we have it. If we ran a lift test or did some other high-quality experiment, we know the actual true parameters. At a certain point in time, we can compare those parameters with our model results to ensure the model is close to those parameters. 4 - Look at out-of-sample forecasting accuracy. The theory here is that if we have truly captured the underlying causal model we should be able to make accurate forecasts on data we haven’t observed yet. In fact, we should be able to make these accurate forecasts even when the marketing budget and channel mix change. So, when we predict the next 30 days, can we make that prediction accurately? Can we do that over and over again for the last 6 months? This is a necessary but not sufficient condition for having a good model. Just because you can do out-of-sample prediction doesn't mean that you have good reads on incrementality, which is the thing that we care about and why we do the other steps in this process.

67

2 Kommentare

Useful post by Christoph Molnar. In the Peakspan Ibbaka NRR Survey for 2024 we found a cluster (found using Leiden) that had a lot more variation than the other clusters. We called it Emergent. It seems to reflect the way that B2B SaaS companies are being disrupted by generative AI. Perhaps that variation is part of how the companies, and the Vertical AI space as a whole, is learning what business and pricing models will be effective. Michael Mansard Edward Wong

2

🚨Controversial opinion : vanity metrics… are actually great. There’s a holy war in Data that has been waged against vanity metrics - usually big numbers like unique visitors, page views or GMV, in general very hard to move and not necessarily correlated with business success. Sales teams once asked me: give me the biggest number you can get to show to clients. I used to laugh at them, but now I get it 💡. There’s a reason why people want to hear big numbers: if you have a big number, you must be a big deal. It is easy mental shortcut that most people will do, mostly unconsciously. Big vanity metrics will grab attention and make people care about what you have to say. It won’t make them act - you still need actionable, operational metrics to drive experimentation and build a KPI tree that reflects your business engine. But it will lure them to pay attention to your harder truths. So to Data teams out there: be more like sales teams, and use vanity metrics!

5

2 Kommentare

AI expenses grew 293% this past year. Companies are shifting budgets and rapidly scaling their AI spend 🤖, according to Ramp's latest business spending report. Tomorrow, join our conversation with Maciej Mylik of ElevenLabs, and Emily Jorgens of Perplexity as we dive into: ✅ AI spending benchmarks by company size and industry ✅ Rising costs to watch out for and keep under control ✅ The fastest-growing vendors on Ramp - and why ⏰ May 22 @ 12pm ET / 9 am PT https://lnkd.in/evN8xtUD

32

1 Kommentar

agreed on using #econometrics instead of #voodoo #attribution, and 9% explanatory power of #googleads is consistent with my findings

13

10 Kommentare

Correlation isn't causation. The goal of attribution is to answer the question "how much business performance is my advertising creating?", not "how much performance is it taking credit for?". Interestingly in my experience this is a major unlock for high-quality, creative, consumer-centric marketing. Win-win.

7

I love Olga Berkovich Lopategui's posts. They always get me excited when her insights correlate to the data I study at Thanx. Here she asks "What examples have you seen of capturing less-obvious impact of loyalty programs?" One shift I've seen is that raw value of the data itself has shifted the value prop of loyalty. Post-covid, restaurants have a lot more data (because of the shift to digital). On Thanx capture rates are now far, far higher because of it. When you can measure the impact of 50% of your business at the user level suddenly the data itself starts providing value, regardless of how provable the ROI of the loyalty program or marketing automations are. I'll give you an example. We had a customer who was struggling with Activation Rate (% of guests who try food and convert to habitual customers). Their data showed that 1x guests were worth $32, but guests who made at least 3 purchases were worth $312, 10x. This is a pretty common multiple. Get em hooked! The problem was the rate (which we show in our dashboard) was stubbornly stuck at 20%. They tried a lot of campaigns and marketing to move it but couldn't budge it. They worked with our team to analyze customer feedback and found that order accuracy was a problem – 20% of consumers expressed something wrong with their order. They had turnover in ops leadership and the new team they hired focused on improving accuracy (and leadership tied financial bonuses to it). Over 9 months they went from 80% accuracy to 99%. The activation rate went from 20% to 30% – 50% improvement! YOY same store sales went from -7% to +7%. Engaged loyalty members (activated customers - churned customers) grew 40% (where it had previously been flat). The point here is that *none* of this was driven by the loyalty program – it was *enabled* by it. They couldn't have diagnosed the issue with out it, they couldn't have measured the impact of their efforts without it. Attaching value to loyalty in 2024 is not only about proving incremental lift of discounts, it's about applying massive leverage to the entire business through insight.

4

2 Kommentare

🚀 Discover the G.R.O.W. Report 2024: Experimentation Maturity Insights At our GMS2023 we introduced the G.R.O.W. Model® for Experimentation Maturity, shedding light on where companies across Germany, Austria, and Switzerland stand in their experimentation journeys. Through extensive interviews and surveys, we collected data from 98 companies and 148 study participants, offering a comprehensive view of experimentation maturity in these regions. 📊 This September, we made the konversionsKRAFT AG G.R.O.W. report 2024 available for download, and now, for our international community, I’ve used Google NotebookLM to summarize the report in English. I’m thrilled with how effectively the study’s findings have been conveyed, offering valuable insights into experimentation maturity. 🎧 The virtual interview lasts around 15 minutes, but feel free to listen at your preferred speed. 🔗 Enjoy exploring the G.R.O.W. Report 2024! 👉 In the comments: • English self-assessment • Link to download the German study PS: please don't be surprised about the black video - it's only about the audio content #Experimentation #Growth #Research #Innovation #GROWModel

10

4 Kommentare

The AI-first contact center workforce is different Less junior agents. More senior agents. And a whole bunch of new roles. And much, Much, MUCH, less churn 🙌 10th in the series, link to the others in the comments 👇 below #AiGetIt #IVR #chatbot #customerservice #GenerativeAI

15

2 Kommentare

GA4 is... not good. I wonder how the introduction of GA4 has affected customer acquisition for Heap | by Contentsquare, Mixpanel, and the like? It's certainly pushed us to look for other solutions. #ga4 #googleanalytics #heap #mixpanel #saas #analytics

2

2 Kommentare

𝐎𝐩𝐞𝐧𝐀𝐈 𝐏𝐫𝐢𝐜𝐢𝐧𝐠 𝐌𝐨𝐝𝐞𝐥. 𝐖𝐡𝐲 𝐬𝐡𝐨𝐮𝐥𝐝 𝐲𝐨𝐮 𝐜𝐚𝐫𝐞? One of the most common pitfalls I see companies face when deploying GenAI at scale is making a good business case. Understanding how pricing models work for GenAI SaaS provider is crucial for that. But there is even a more crucial question to be asked once you understand these models: 𝐈𝐬 𝐲𝐨𝐮𝐫 𝐛𝐮𝐬𝐢𝐧𝐞𝐬𝐬 𝐦𝐨𝐝𝐞𝐥 𝐬𝐮𝐬𝐭𝐚𝐢𝐧𝐚𝐛𝐥𝐞 𝐢𝐧 𝐭𝐡𝐞 𝐞𝐫𝐚 𝐨𝐟 𝐆𝐞𝐧𝐀𝐈? This is an exercise I would recommend software and in general services companies taking very seriously, asking themselves if their model is sustainable, and 𝐦𝐨𝐫𝐞 𝐢𝐦𝐩𝐨𝐫𝐭𝐚𝐧𝐭𝐥𝐲, 𝐛𝐞𝐢𝐧𝐠 𝐚𝐛𝐥𝐞 𝐭𝐨 𝐜𝐥𝐞𝐚𝐫𝐥𝐲 𝐞𝐱𝐩𝐥𝐚𝐢𝐧 𝐭𝐡𝐞𝐢𝐫 𝐜𝐮𝐬𝐭𝐨𝐦𝐞𝐫𝐬 𝐰𝐡𝐲. Let me put you an example. OpenAI's states that for his 𝐧𝐞𝐰 𝐚𝐝𝐯𝐚𝐧𝐜𝐞𝐝 𝐯𝐨𝐢𝐜𝐞 𝐦𝐨𝐝𝐞𝐥 "Audio input costs are approximately 6¢ per minute; Audio output costs approximately 24¢ per minute". This means that companies could now "hire" AI call center digital assistants for less than 18$/ hour. That, when added to the new announcement of prompt caching will make this pricing lower in the coming months/ years. If you are a call center company, what does this mean for your business model? What is the differential value you provide your customers that makes you stronger than these pure GenAI-based offerings? Definitely food for thought for business leaders. #generativeAI #businessmodel All OpenAI pricings can be found here: https://lnkd.in/dNwHsbeJ

17

IMHO any B2B company should use Agentic AI Graphs to generate Next Best Actions in #Sales. It is not that hard. First, understand the data model underneath the CRM. What data points are available? What data points are missing? Second, extract relevant objects (company, person, purchase history, etc.). Enrich these objects with external signals using tools like Tavily. Third, build the agentic graph for each use case (e.g., with LangGraph from LangChain). Some cases it is quite straightforward, while others may be more complex (due to high process complexity). Fourth, loop the next best actions back into the CRM. Anyone has built & deployed Agentic AI Graphs for Sales? Let's have a chat.

108

21 Kommentare

Good McKinsey & Company article on how and why to rewire CPG companies around digital and #AI. 💰 Understand where the value can be created with existing technologies. Consumer insights and channel & customer management have biggest ROI in a lot of categories (see diagram). 👀 One beverage company that embarked on a digital and AI transformation unlocked 18 percent EBITDA uplift over two years 🚪 The importance of creating an innovation culture with cross-functional pods and phase gate processes for innovation funding 🎓 Upskilling your people through tailored learning curricula including on-the-job learning, formal training sessions, and individual coaching. 🔢 The importance of standardizing data pipelines, create reusable data platform products, elevating data product owners https://lnkd.in/ekWTUCHg

33

11 Kommentare

If you run ads only on one channel - Meta, Amazon, Google Search, TikTok - is it OK to rely on last-click attribution? How wrong can the results be? The answer: Potentially 100%. The issue is that the rules of causal inference and math do not change for smaller companies or if you only use 1 or 2 media channels. Selection effects can completely mislead you, even for a single channel, as shown in my IAG case study example shared recently: https://lnkd.in/geQ9ysFi Now, the risk of being 100% wrong is much smaller if you are a lesser-known brand and only operate online. However, the issue is - incrementality still matters. Customers may convert anyway if they know your website after their first purchase, heard about your great product from a friend (offline word of mouth), or remember you and then Google you later and convert. As a result, last-click or last-touch attribution inflates allocated attributions and suggests a very inaccurate CPA. Maybe you're only wrong by 40-70% in terms of absolute CPA, but this still matters when every dollar counts for SMEs or startups. This becomes even more critical as soon as you start comparing two campaigns for the same platform - maybe one prospecting and one retargeting campaign, both running on Facebook. Last-click/ touch will typically tell you to invest more and more in retargeting as the CPA will always look better here. Even when this primarily shows ads to anyway-converters. So, are there situations where last-click attribution can be correct? Yes, but these are rare and usually for a short period of time: when you sell to your very first customers. What can you do instead? Well, experimentation is free (and fun). The main costs are opportunity costs, which are just learning costs and an important investment in the future. For example, you can always vary your single channel or tactic and simply compare sales: 🔁Switchback experiments: we turn a tactic on and off every second day. 🌍Geo-tests: select some regions where you turn off one tactic. 📊A/B tests: Most platforms also have A/B test functions (some randomized, others optimized ones) that you can use. See: Meta - Conversion Lift and A/B Testing https://lnkd.in/gr6nyg7m and https://lnkd.in/gTip8uJ9 TikTok - Split Test https://lnkd.in/gaCiZ-G7 Amazon - A/B Test and Brand Lift https://lnkd.in/gF2YBmsN and https://lnkd.in/g5KDzgek Google - Conversion Lift https://lnkd.in/gWCYnwsu Trade Desk - Conversion Lift https://lnkd.in/gWBaaGdS While it’s tempting to rely on last-touch and last-click attribution as a default starting point for a company’s measurement journey, let’s not encourage people to trust a broken clock. With all the free alternatives available, there’s no reason not to move on to more reliable marketing effectiveness measurement as soon as possible.

179

51 Kommentare

We have been doing some really exciting work at LLMate to redefine data analysis and insight generation. Traditional BI is going to change completely with the help of LLMs. Researches point out that 92% of data workers report time wastage by operational tasks outside of their role. 30 years ago, when dashboards came into the picture, they solved a lot of problems that traditional rows and column views could not. But with time and technological advancements dashboards have started to become obsolete and less useful. Some of the major drawbacks are - 1.) Real time access of data not available. By the time data is translated to dashboards, the moment to act already passes (avg. time to create a dashboard is 4.5 days) 2.) Granularity - Teams need a freestyle approach to data discovery, rather than static dashboards. 3.) Missing collaboration - Dashboard development is usually done in silos. 4.) Discoverability - With varying access and permission controls, tremendous overheads are there to make data accessible. Apart from the above issues, dashboards fail to empower the frontline workers and can't match the modern pace of data analysis. But, with the help of AI, these problems could be solved. Keeping this in mind we started our journey of creating LLMate - A Notebook styled tool for data analysis using plain natural english as the interaction language. LLMate solves 3 major problems for data teams - 1.) Making LLMs understand your data and its semantics 2.) Creating AI data analysis co-pilots to power your internal teams and push them into production. 3.) Managing and improving the AI agents in production. We are launching our new "Notebook" feature, that helps you create analysis notebooks that can be published and shared with stakeholders. Imagine Notion, but with AI agents that can be inserted as custom blocks. Here are some glimpses of our "Notebook" feature. More exciting updates to be announced soon! If you feel LLMate can upgrade your data teams, do reach out to us for an early access. Nikhil A. Abhas Sinha Parth Jaimini #DataAnalysis #CopilotForDataTeams #Notebook #LLM #ChatWithData #AIDataAnalysis #Collaboration

48

4 Kommentare

https://lnkd.in/eVvZx_4J Complete list of all Kroger and competitors with geocoded address, phone number, open hours, etc for instant download. This dataset contains Kroger, Walmart, Whole Foods Market, Costco, Target locations.

2