Apple Podcasts  ·  Spotify  ·  Buzzsprout

Paper A

HOBA: Hierarchical On-Policy Bidding Agents for Adaptive Online Advertising

Ji Wu, Yunshan Peng, Wentao Bai, Y. Bai et al. — 2026 — KDD '26 (ACM SIGKDD Conference on Knowledge Discovery and Data Mining)

preprint  ·   ·  read now

Key findings

• The HOBA system increased advertiser 'target cost achievement' by +3.6% compared to the existing bidding system in a real-world A/B test on a major ad platform — meaning advertisers got closer to spending their full budgets efficiently without violating cost-per-action limits.
• Using an LLM to automatically set bidding system settings (like bid caps and pacing speed) every hour reduced the need for humans to manually tune these settings per campaign — a task that currently costs engineering time and is slow to react to market changes.
• Limiting the AI's live learning to choosing between 5–10 pre-validated bidding algorithms (rather than freely adjusting bids directly) made the system much safer — it avoided the risk of a runaway AI exhausting an entire daily ad budget in minutes, which is a known failure mode of simpler reinforcement learning systems.
• Combining strategy (LLM), selection (SARSA agent), and execution (specialist models) in separate layers outperformed systems that treat bidding as one big single optimization problem, suggesting that breaking the problem into pieces each handled by the right tool works better than a one-size-fits-all approach.

Marketing implications

• If you manage large ad budgets on programmatic platforms, push your ad tech vendor or DSP to show you whether their auto-bidding system adapts in real time to market shifts — or whether it was trained offline months ago and is now running stale. This paper shows that real-time adaptation matters and is achievable.
• If you're building or evaluating ad tech, the core insight is: don't let AI explore freely with live budget. Constrain live learning to choosing between a small set of pre-tested strategies rather than letting it freely adjust bids. This keeps campaigns safe while still adapting.
• If you run campaigns on platforms with manual pacing or bid cap settings, ask whether those settings are being updated more than once a day — the paper shows that updating them every hour using market signals improves performance meaningfully.

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  ·  open access  ·  read now

Key findings

• When consumers believed a human was actively involved in overseeing the AI (rather than the AI acting alone), they trusted the recommendations significantly more. This trust was the strongest effect measured in the entire model.
• Trust in the human–AI combination directly made people more willing to buy — and more comfortable making financially significant decisions — not just trust the information in theory.
• Trust was not the only path to better outcomes: human–AI collaboration also directly boosted purchase intentions and financial decision confidence even before trust was considered. Trust amplified the effect but didn't create it from scratch.
• The model explained about 64% of why people said they'd buy and 55% of why they felt confident in financial choices — meaning human–AI collaboration and trust together account for a large share of those consumer decisions.

Marketing implications

• If you use AI chatbots or AI-powered product recommendations on your site, make it visible that a human reviews or validates the AI's output. Something as simple as 'Reviewed by our team' or a human specialist badge may increase the number of people who trust and follow through on AI recommendations.
• For higher-stakes products (financial services, insurance, big-ticket items), the trust gap between AI-only and human-plus-AI is especially relevant. Consider having a human advisor visible in the AI flow — even asynchronously — rather than fully automating the experience.
• When pitching AI marketing tools internally or to clients, lead with the trust story: showing that human oversight is built in is not just an ethics argument — this study suggests it meaningfully moves purchase intent numbers.

Paper C

Generative AI In Marketing: Productivity Gains and Work Automation

Joel Gastman, Marco Bastos Toledobastos — 2026 — AoIR Selected Papers of Internet Research

peer reviewed journal article  ·   ·  read now

Key findings

• ChatGPT could generate complete, coherent social media marketing strategies quickly — faster than a human professional would take — and the strategies were judged to be tailored to each company's brand identity and specific context.
• The AI was especially good at brainstorming concrete, measurable ideas and building in feedback loops for adjusting campaigns over time. Think of it as a very fast first-draft machine.
• ChatGPT had clear blind spots: its budget estimates and suggested posting frequencies were often unrealistic, and its answers sometimes felt generic or 'cookie-cutter' — fine as a starting point, but not ready to use without human editing.
• Professionals did not believe AI would immediately replace their jobs, but they did flag that a wide range of tasks — including strategy creation — could be partly automated, raising longer-term employment concerns. Larger companies were already building private, in-house AI tools to avoid data privacy issues with public tools like ChatGPT.

Marketing implications

• If you're a social media manager who needs to build a strategy from scratch, use ChatGPT to generate a first draft fast — it will give you a reasonable structure and some concrete ideas in minutes. Then edit it: fix the budget numbers, adjust the posting frequency to what's actually realistic for your team, and add anything specific to your audience that the AI missed.
• Don't feed confidential client data or proprietary campaign details into public tools like ChatGPT. As the paper notes, larger companies are already building private AI tools for exactly this reason. If you work at an agency, flag this risk to clients before using AI in their strategy work.
• Use AI for the parts of strategy work that feel repetitive or slow — brainstorming, structuring a content calendar, drafting objectives — and save your human time for the parts that actually require judgment: knowing the client's real budget, understanding the audience's cultural context, and making creative calls.

Apple Podcasts  ·  Spotify  ·  Buzzsprout