Apple Podcasts  ·  Spotify  ·  Buzzsprout

Paper A

Recommendation with Generative Models

Yashar Deldjoo, Zhankui He, Julian McAuley, Anton Korikov et al. — 2026 — Foundations and Trends® in Information Retrieval

preprint  ·   ·  test this week

Key findings

• AI recommendation systems that can generate content (not just rank existing items) are better at handling new users or new products with no purchase history — a problem that has long plagued traditional recommendation engines.
• Chatbot-style recommendation (where a system has a real back-and-forth conversation with a user to refine suggestions) is now practical using large language models, and it outperforms older systems that couldn't maintain a real dialogue.
• Generative AI can create personalized visual content — like showing a shopper what a piece of clothing would look like on them (virtual try-on) or generating a custom product ad — going far beyond just listing 'you might also like' items.
• Deploying these advanced AI recommendation systems at scale is expensive and risky: they can unintentionally reinforce existing biases, trap users in narrow content bubbles, spread misinformation, or be manipulated to push users toward certain products. The authors argue current evaluation methods aren't sufficient to catch these problems.

Marketing implications

• If your e-commerce platform suffers from low conversion on new product launches (because the algorithm has no history to work from), look into LLM-based recommendation tools — this paper explains why they specifically help with the 'no data yet' problem.
• If you run a high-consideration purchase category (fashion, furniture, electronics), the virtual try-on and personalized visual ad generation described here are now real technologies — worth piloting as a premium product experience.
• Before deploying any AI recommendation system, build in explicit checks for filter bubbles (are you only showing users the same narrow slice of your catalog?) and bias (are certain customer groups getting systematically worse recommendations?). The authors say these problems are easy to miss with standard metrics.

Paper B

Artificial Intelligence in Marketing: A Bibliometric Analysis and Integrated AI Marketing Knowledge Framework

Husni Rifqi, Zaidan Mufaddhal, Rishab Manocha, Husna Putri Pertiwi — 2026 — Journal of AI & Immersive Marketing

peer reviewed journal article  ·  open access  ·  test this week

Key findings

• AI marketing research has been growing at roughly 27% per year since 2020, meaning the field is nearly doubling every three years — it is not a fringe topic anymore.
• The study found ten distinct research clusters in the AI marketing literature, which can be grouped into three big buckets: (1) Strategic and Service AI (using AI for business strategy and customer service), (2) Consumer AI (how AI affects consumer behavior and decisions), and (3) Conversational AI (chatbots, virtual assistants, and AI-driven dialogue).
• The authors built a unified 'knowledge map' showing that how much AI can do (capability level), how consumers react to it (response mechanisms), and what marketing job it is doing (application domain) all work together to create business value — no single piece works in isolation.
• Generative AI (tools like ChatGPT that create text, images, and video) has opened a brand-new research stream within AI marketing, but it also brings fresh risks around bias, misinformation, data privacy, and ethical governance that the field has not yet fully worked out.

Marketing implications

• If you are trying to figure out where AI fits in your marketing org, this paper gives you a simple three-part map: use AI for strategy and service automation, for understanding and influencing consumer behavior, and for chat and conversation. Pick one bucket to start, not all three at once.
• Generative AI (ChatGPT-style tools) is officially its own fast-growing research area — if you are not already experimenting with AI content creation or AI chat, you are behind the curve of both research and practice.
• Before rolling out any AI marketing tool, build a short checklist covering data privacy, algorithmic fairness, and what happens when the AI gets things wrong — this paper confirms those risks are real and under-studied.

Paper C

Integrating AI to Improve Customer Experience and Marketing in Zambia's Insurance Sector: A Case Study of Selected Insurance Firms

Oscar Mulungu, Austin Mwange — 2026 — African Journal of Commercial Studies

peer reviewed journal article  ·   ·  test this week

Key findings

• Most Zambian insurers are barely using AI: 63% of respondents said their firm uses some form of AI, but only 31% confirmed having a chatbot — meaning most 'AI use' is basic and not customer-facing.
• More AI use was linked to better marketing results: firms that used AI more tended to score higher on marketing effectiveness, with a moderate positive relationship (correlation of 0.566 out of a maximum of 1.0). This is a correlation, not proof that AI caused the improvement.
• In interviews, staff said AI helped with things like faster customer targeting and smoother internal operations, but confirmed that full enterprise-wide AI rollout is rare — most firms mix digital and human service delivery rather than automating end-to-end.
• Three main barriers are holding adoption back: (1) not enough staff with the right digital skills, (2) poor or incomplete customer data, and (3) unclear government rules about how AI can be used in insurance.

Marketing implications

• If you're a consultant or tech vendor targeting emerging-market insurers, this paper signals that most Zambian insurance firms are at step one of AI adoption — basic digitization, not intelligent automation. Lead with simple, easy-to-explain tools (e.g., chatbots for FAQs) rather than complex AI pitches.
• The biggest blockers aren't budget or interest — they're skill gaps and messy data. If you're selling AI tools to insurers in frontier markets, bundle in training and data-cleaning support or you'll lose the sale at implementation.
• Hybrid models (some AI, some human agents) are what's actually working on the ground. Don't design or pitch fully automated customer journeys for this market — keep humans in the loop for complex queries.

Apple Podcasts  ·  Spotify  ·  Buzzsprout