Apple Podcasts  ·  Spotify  ·  Buzzsprout

Paper A

Can Large Language Models Identify Meaningful Touchpoints in Conversion Attribution?

Jinqi Wu, Sishuo Chen, Zhangming Chan, Yong Bai et al. — 2026 — arXiv (Cornell University) / CIKM '26

preprint  ·   ·  deep dive

Key findings

• Almost 15% of customer touchpoints that led to purchases were 'hidden' connections — for example, someone browsed a camera, then bought a memory card. Existing systems completely missed these because the two products aren't in the same category and don't share obvious buying history patterns.
• Top AI models (like GPT 5.5 and Gemini 3.1 Pro) are good at finding the obvious touchpoints, and decent at finding hidden ones, but there's still a lot of room to improve on the hidden ones specifically.
• Asking the AI to compare touchpoints one-by-one (pairwise) works better than showing it a whole list at once (listwise) — but this gap mostly disappears when you use a very capable model.
• Using AI-identified touchpoints to train a purchase-prediction model improved that model's accuracy by 0.35 percentage points over the standard production system and 0.15 points over a competing method — meaningful gains at the scale of hundreds of millions of users.

Marketing implications

• If you run attribution for an e-commerce store, your current model is probably missing about 1 in 7 meaningful customer touchpoints — specifically the 'complementary product' browsing (e.g., a customer looked at camping tents before buying a sleeping bag). Ask your data team to audit how many cross-category touchpoints your attribution model currently counts.
• If you're building or buying a CVR prediction or attribution tool, test whether adding LLM-based semantic matching on top of your existing rules improves predictions — this paper shows even a modest improvement at scale translates to real money.
• When prompting AI to analyze customer journeys or touchpoints, have it compare two options at a time rather than evaluating a whole list — you'll get more accurate results.

Paper B

Human–Generative AI Collaboration in Digital Marketing: Its Impact on Consumer Trust, Purchase Intentions, and Financial Decision-Making

Ratna Oza, P. William, P. Tamilselvan, Malvika Kandpal et al. — 2026 — Journal of Intelligent Decision Making and Information Science

peer reviewed journal article  ·   ·  read now

Key findings

• When AI and humans work together in digital marketing (rather than AI acting alone), consumers trust the brand or platform more. This trust was the strongest effect the researchers measured.
• Human–AI collaboration directly increased how likely consumers were to buy a product and how confidently they made financial decisions — even before accounting for trust.
• Trust acted as an important connector: part of why human–AI collaboration leads to more purchases and better financial decisions is because it first builds trust, which then drives those outcomes.
• The model explained 64% of the differences in people's purchase intentions and 55% of the differences in financial decision-making — meaning human–AI collaboration and trust together are powerful predictors of what consumers actually decide to do.

Marketing implications

• If your brand uses AI chatbots or recommendation tools, tell customers a human reviewed or oversees the AI's suggestions. Even a small disclosure like 'Reviewed by our team' may increase trust and purchase rates.
• For high-stakes purchases (finance, insurance, big-ticket items), make sure your AI-assisted customer experience includes visible human accountability — a named advisor, a human sign-off, or a live chat escalation — rather than pure automation.
• Test your AI marketing flows with a 'human-in-the-loop' version against a fully automated version to see if trust signals (review requests, return rates, repeat purchase) improve.

Paper C

Pengaruh Generative AI Marketing terhadap Purchase Intention melalui Customer Trust dan Perceived Personalization pada Pengguna E-Commerce

Destiyana Yuwanti, Imam Yuwono, R. Saputra — 2026 — Jurnal Ragam Pengabdian

peer reviewed journal article  ·   ·  use cautiously

Key findings

• When shoppers encounter AI-powered marketing features (like personalized product recommendations or AI chatbots), they are more likely to say they intend to buy something on that platform.
• The main reason AI marketing lifts purchase intent is a two-step chain: AI features make people feel the platform 'gets them personally,' which then makes them trust the platform more, which makes them more likely to buy.
• Feeling like recommendations are tailored to you (personalization) matters on its own too — it directly increases both trust and purchase intent, not just through the trust step.
• The study suggests that for AI marketing to work, the technology needs to do more than generate recommendations; it needs to make users feel understood and safe — without that, the purchase-intent boost may not follow.

Marketing implications

• If you run an e-commerce store or marketplace, make sure your AI recommendation features clearly signal *why* they are recommending something ('Because you viewed X…'). Shoppers who feel understood are more likely to trust you and buy — so the explanation matters, not just the recommendation itself.
• Before adding more AI features, check whether your current ones feel personal to users. Run a quick survey asking: 'Do our recommendations feel relevant to you?' If the answer is no, more AI won't fix it — you need better personalization signals first.
• Add trust-building elements alongside your AI features: show data-privacy badges, explain how your AI works in plain language, and make it easy to adjust preferences. This study suggests trust is a key middle step between 'AI feels personal' and 'I actually buy.'

Apple Podcasts  ·  Spotify  ·  Buzzsprout