Thomas Bosilevac

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

MLRs are up. Budgets are tight. AI pressure is everywhere. But here’s the uncomfortable truth:   Most Medicare Advantage performance problems don’t start in the campaign. They start in the data.   Broken identities. Duplicates. Eligibility mismatches. Outdated contact records. Small errors that feel harmless — but quietly absorb 5–10%+ of acquisition spend before a single mail piece drops.   In a 90%+ MLR world, that margin matters. What I’m seeing across the industry right now is a shift: Growth isn’t about pushing more volume anymore. It’s about eliminating waste before it scales.   And here’s the part that surprises people most: When you fix identity, qualification, and contactability first — performance often improves without increasing spend. Not by spending more. By leaking less.   AI absolutely has a role to play. But AI multiplies what already exists — good or bad. If the foundation is unstable, it industrializes inefficiency. The smartest teams I’m talking to aren’t anti-AI.   They’re sequencing correctly: 1. Tighten identity 2. Lock qualification 3. Improve contactability 4. Then layer innovation   That’s disciplined growth. That’s how you protect budget in a high-MLR environment. That’s how marketing earns credibility with finance.   I wrote a deeper piece on this shift and what it means for Medicare Advantage leaders right now. If you’re feeling the pressure to “do more with less,” this may resonate.   👉 Read the full article here: https://lnkd.in/geRPXtYn   I would genuinely love to hear what others are seeing in the market right now. Is your team talking about AI first — or fundamentals first? — David Schneider #MedicareMarketing #MedicareAdvantage #HealthcareMarketing #HealthcareData #Medicare

11

There’s been an explosion of ‘off-the-shelf’ analytics platforms promising capabilities like MMM, attribution/MTA, segmentation, LTV modeling, , and more. I come from the more traditional approach - working closely with statisticians to build custom models from the ground up These new platforms promise speed, accessibility, and scale. The traditional approach offers deeper customization and often more methodological rigor. But now AI is entering the equation - raising a bigger question: Will AI augment analysts… or eventually replace much of this work altogether? We’re planning a bit of a 'bake-off' internally to compare approaches, but it feels like the bigger shift is structural: how analytics actually gets done may be changing. Curious how others are thinking about this: - Are off-the-shelf platforms “good enough” for most companies? - When does a custom statistical model still matter? - And does AI ultimately reshape—or disrupt—both approaches? Feels like we’re at a real inflection point in the analytics industry. Would love your thoughts and experiences here.

20

7 Comments

Almost everyday I hear about MMMs that fail because they haven’t been set up properly, don’t go into enough detail or just spit out numbers with no thought behind what they mean for the business. I like this graphic below, sums up nicely what an MMM should be. A strategic tool to help guide investment decisions ✅

29

3 Comments

Market research generates huge volumes of data. But the real value lies in how effectively those insights are communicated. That’s where data visualization comes in simplifying complex datasets, highlighting patterns, and enabling faster decision-making for businesses. In our latest blog, we share best practices to create impactful data visualizations that make research insights easier to understand and act upon. Read the full blog here👇 https://lnkd.in/djAprnqw #MarketResearch #DataVisualization #Insights #BusinessStrategy #Blog #GlobalSurvey

2

Some exciting news today from Newton Research with the announcement of our integration with Snowflake Cortex AI. Proud to be partnering with Snowflake and our joint customer Horizon Media in driving secure, in-platform marketing analytics - Scaled through agentic AI

20

Great overview of the AI powered research tech landscape - I find it amazing how quickly this space has become crowded, and it seems to be getting harder to establish a clear differentiation/USP, I guess because the barriers to development keep getting lower.

3

The team at Knowledge Excel is always cooking up amazing survey experiences! Today Saurabh Aggarwal Rashmi Sharma and Tarun Khanna introduced AI Avatars in surveys at the Sawtooth conference. Yep—it’s not sci-fi anymore! They showed real potential to enhance engagement and improve data quality—and the early results are worth paying attention to. In their tests, respondents were surprisingly open to the experience: - 9 in 10 said they’d take another survey with an AI avatar - Almost all described the avatar as friendly, relatable, and engaging - And overall, the experience was seen as more enjoyable than a standard survey But like any new method, impact depends on how it’s implemented. They recommended the following best practices: 1️⃣ Reassure respondents about privacy 2️⃣ Let them choose their avatar for a sense of agency 3️⃣ And use them strategically… aka only where “the interruption adds value” What’s next? 🤖 Leveraging LLMs to power avatars so they can engage in real-time, two-way dialogue. Imagine avatars that can guide respondents, probe deeper where needed, and adjust the experience on the fly. Love love love this idea. What do you think? Are AI avatars in surveys a game changer or a mere gimmick? Comment below! 👇🏽

38

2 Comments

Qualtrics just showed that when you train LLMs on real historical survey data, it improves replication accuracy significantly. Here's the issue - LLMs learn from SINGLE INSTANCES in the training data. That's because these models are now big enough that they can effectively store this information in their model weights. How big? Qualtrics trained the model on "hundreds of thousands of questions". ChatGPT 4.5 (now ancient) has 12.8 trillion parameters. That's about 50 million parameters per question in the training model. So yeah, the model can EASILY recover specific data. Don't believe me? The latest ChatGPT models have started solving some of the Erdos math problems. It turns out that several of those proofs merely replicated obscure work published by amateur mathematicians in hidden corners of the internet. More proof? LLM models can effectively replicate the entire text of novels - not just summaries, but text themselves. See comments for citations. This is a humdinger. I re-checked Qualtrics' terms, and they are allowed to update them just by publishing them. Their terms currently explicitly state: "To train our artificial intelligence tools. We may process personal information using artificial intelligence (AI) technologies including generative AI, and we use this information for training machine learning models, natural language processing, large-language models and other relevant AI use-cases." So yeah, they can do it legally. They say they are anonymizing and aggregating training data. Anonymizing offers no protection - it just means removing any userID from the data. It's still an actual answer to a question that a client paid to gather on Qualtrics. Aggregating does not specify at what level they are aggregating. I could aggregate, for instance, by grouping people into sets of 5, or specific subgroups. Even if I'm aggregating at the survey level, it's still giving the LLM the answer to a question that is highly confidential business information. What does this mean in practice? It means I could use this tool to write a question like: "What are the top innovations for my competitor brand that you find appealing?" or "What are the most important sources of dissatisfaction for my competitor brand?" I then "simulate" respondents answers, which are partly based on my competitors' actual survey or customer experience data. That's some really nice competitive intelligence. [Edit Feb 13] Callout to thoughtful comment by David Thor below noting that the language for AI training in the T&Cs was added November 2023.

310

100 Comments

One of the biggest unspoken realities in market research right now - fragmentation is quietly killing agility. Mid-size agencies and consultancies are still juggling separate tools and vendors for survey programming, sampling, data collection, cleaning, analysis, and reporting. Every handoff introduces delays, errors, and rework. Timelines stretch from weeks to months, costs balloon from unexpected fixes, and data quality becomes a constant gamble, especially with rising fraud and inconsistent validation. The result? Teams spend more time firefighting operations than delivering strategic insights. Clients notice the slowness and unreliability, and trust erodes. What's your biggest operational bottleneck in 2026? Fragmented workflows, data quality risks, or something else? #MarketResearch #InsightsIndustry #ResearchOps #DataQuality #BioBrainInsights Aditya Sonal Edan Prateek Dilip Rajat Navneet

21

4 Comments

If Your Intent Data Provider Cannot Explain Their NCOA Cadence, Ask Why When was the last time your dataset was refreshed? If the answer is “annually,” you are operating on decaying data. If the answer is unclear, that is worse. In high-spend environments, unverified match infrastructure is not neutral. It is expensive.

1

🧠 𝗬𝗼𝘂 𝗱𝗼𝗻'𝘁 𝗵𝗮𝘃𝗲 𝗮 𝗺𝗮𝗿𝗸𝗲𝘁𝗶𝗻𝗴 𝗽𝗿𝗼𝗯𝗹𝗲𝗺.  𝗬𝗼𝘂 𝗵𝗮𝘃𝗲 𝗮 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝗽𝗿𝗼𝗯𝗹𝗲𝗺. In a world drowning in data, here's what most businesses still get wrong about loyalty, retention, and experience: 🚨 Customer acquisition costs are up nearly 60% in 5 years. Worse? You're losing $29 per new customer, on average. If retention isn't your top priority, you're scaling a sinking ship. 🚨 81% of people sign up for loyalty programmes... but only 49% use them. It's not about points. It's about psychology. Mass personalisation is over. Precision engagement wins. 🚨 80% of customers will abandon your brand after one bad experience. Today's consumers aren't loyal to products. They're loyal to how you make them feel. 💡 𝘐 𝘤𝘢𝘯 𝘵𝘶𝘳𝘯 𝘺𝘰𝘶𝘳 𝘣𝘦𝘩𝘢𝘷𝘪𝘰𝘶𝘳𝘢𝘭 𝘴𝘪𝘨𝘯𝘢𝘭𝘴 𝘪𝘯𝘵𝘰 𝘨𝘳𝘰𝘸𝘵𝘩. Because knowing your customer isn't enough – you need to understand what drives them, what repels them, and what keeps them coming back. 👇 What's the one customer stat that changed how you think? #customerexperience #retention #behaviouralscience #customerloyalty #marketingtruths #businessgrowth #datapsychology

12

8 Comments

Does your CRM know “R. Smith” and “Rob Smith” are the same person? If not, you’re losing time and money every day. Duplicate records lead to manual reviews, regulatory headaches, and member frustration. The stakes are even higher when verifying eligibility or processing loans—errors cost you real dollars and trust. But this isn’t just an IT issue. It’s an operational roadblock affecting every facet of how you serve members. More data sources and rising speed expectations mean the problem will only get harder to ignore. So what’s the fix? Master entity resolution. Clean, connected data allows teams to make smart, compliant decisions the first time, every time. Ready to turn fragmented identity headaches into legendary member experiences? https://hubs.li/Q03R1JLk0

1