AI Marketing Research: CRM Performance, Agent Loyalty & AI Forecasting

When your AI says it's predicting the future, is it actually forecasting — or just remembering? And what happens to brand loyalty when an AI agent, not a human, is the one making the purchase? Three papers this week examine whether we are actually getting what we think we are getting from AI in marketing. In this Research Radar Brief, Dr. Eva Wolf reviews 3 recent AI marketing research papers covering AI-driven CRM and financial performance in digital banking, AI agent loyalty loops in autonomous commerce, and a method for evaluating whether AI forecasting tools are genuinely predicting or retrieving memorized answers. Note: Evita is an AI-generated research briefing avatar trained on the research framework of Dr. Eva Wolf. Every Friday, Dr. Wolf records a live weekly roundup with her own analysis. What you'll learn: - Why AI-powered CRM outperformed both personalization and chatbots as a financial performance driver in a Nigerian digital banking study, and how to use that R-squared value in a budget conversation - How to tell whether an AI forecasting tool your vendor sells is actually predicting anything, or retrieving answers it already knows from training data - Why emotional brand loyalty may stop mattering once an AI agent takes over purchasing decisions — and what marketers need to put in its place - What machine-readable brand signals means in practice, and why loyalty programs need to be legible to algorithms, not just humans - Why noisy or speculative retrieval sources make AI predictions worse, not merely less precise Papers covered: 1. The Influence of AI-Driven Marketing on the Financial Performance of Digital Banks in Nigeria Source type: Peer-reviewed journal article (likely peer-reviewed, published via Zenodo) Access: Full text reviewed DOI: 10.5281/zenodo.21277512 2. The Dynamic Verifiable Multi-Agent Human Agentic Loyalty Loop (DVM-HALL) Model and the Net Human-Agent Score (NHAS) in Autonomous Commerce Source type: Preprint (not yet peer-reviewed — treat findings as early-stage and theoretical) Access: Full text reviewed Source: https://arxiv.org/abs/2607.13998v1 3. Hindcast: Replaying Prediction Markets to Evaluate LLM Forecasters Source type: Preprint (not yet peer-reviewed) Access: Full text reviewed Source: https://arxiv.org/abs/2607.14051v1 Full show notes, transcript, and citations: https://bigplans.media/episodes/ai-marketing-crm-performance-agent-loyalty-forecasting-2026-07-16 Disclaimer: This is a first-pass research briefing produced by an AI-generated avatar trained on Dr. Eva Wolf's research framework. It is not a substitute for reading the original papers. Preprints have not been peer-reviewed and findings may change. Correlation findings do not establish causation. Always consult the original source before citing or acting on any research discussed here. -- This is a first-pass research briefing, not a final academic review. Read the original papers before making major marketing or business decisions. AI & Marketing Research Radar is produced by BigPlans Media. Subscribe wherever you listen to podcasts.

Thanks for listening to AI & Marketing Research Radar by Big Plans Media.

I’m Dr. Eva Wolf, and I help marketers, educators, consultants, and business owners turn AI marketing research into practical strategy, smarter workflows, and real business opportunities.

More episodes: https://bigplans.media/ai-marketing-research-radar/
Consulting: https://bigplans.media/ai-marketing-consulting/

Big Plans Media — Where Big Ideas Meet Smart Marketing.

Transcript

Copy Transcript Copied

SPEAKER_00 0:07

You're listening to Evita, an AI-generated research briefing avatar trained on the research framework and methodology of Dr. Eva Wolfe, marketing professor, AI researcher, and founder of Big Plans Media. Every day, Evita scans emerging research in AI, marketing, consumer behavior, psychographics, and business strategy to identify the most relevant developments, opportunities, and risks worth watching. These daily radar reports are designed to help busy professionals stay informed without having to read hundreds of research papers themselves. And every Friday join Dr. Eva Wolfe live for her personally recorded weekly AI Marketing Radar Roundup, where she breaks down the biggest stories, explains what actually matters, and shares practical insights and strategic implications for marketers, educators, entrepreneurs, and business leaders. Now here's today's radar report. Here's the signal I can't ignore today. What if the AI tools you're using to predict, personalize, and build loyalty are doing something much simpler than you think. Not forecasting, not optimizing, just remembering. And what if the customers those tools are supposed to win over aren't even human anymore? Today's papers point to the same uncomfortable pattern. AI and marketing is moving faster than our ability to measure whether it's actually working or working on the right audience. We screened 394 papers. Three cleared the full text bar and made today's radar. Quick caveat: this is a first pass research briefing, not a final academic review. Every paper covered today has full text access. I'll tell you what the papers suggest, what they don't prove, and which ones deserve a deeper read. Okay, let's get into it. Paper one. Here's the business question. If your digital bank or your fintech client is about to invest in AI marketing tools, which one actually moves the revenue needle? Researchers surveyed 236 employees at digital banks in Nigeria. They tested three AI marketing tools, personalization engines, AI-powered CRM, and chatbot. Then they ran regression analysis to see which one best predicted financial performance. So here's what they found. All three tools showed a statistically significant positive relationship with financial performance. But they weren't equal. Personalization explained about 61% of the variation. Chatbots came in around 57%, and AI-powered CRM, that was the winner. 66%. Let me translate that. CRM. The tool that tracks customer behavior, flags the right moment to reach out, manages the relationship over time, that's the one most closely linked to actual bank revenue. Not the chatbot, not the fancy recommendation engine, the CRM. Here's where this gets interesting. Most fintech marketing budgets right now are going toward chatbots and personalization because they're visible. You can demo them. CRM is invisible to the customer, but apparently it's where the money is. Not the flashy feature, the infrastructure. Now, and I want to be honest about this. The data came from employees, not customers, and not audited financial statements. Employees reporting on their own bank's performance is not the same as looking at actual revenue figures. That's a real limitation. And this is cross-sectional, one snapshot in time. So we can't say AI CRM caused better performance, only that banks using it more appeared to perform better. That's the part that bothers me a little. Because those R-squared values are high enough that teams will treat this as proof. It's not proof. It's a signal. A strong one, but still a signal. Plain English payoff. In digital banking, AI-powered CRM appears to be a stronger predictor of financial performance than chatbots or personalization. So if you're prioritizing your AI marketing stack, start there. Okay, here's where this becomes commercially interesting. Money move. Build an AI CRM audit service for African fintech companies. This research signals strong demand and a clear ROI narrative in that market, and the consulting gap is wide open right now. Action step. Before your next budget review, pull your current AI marketing spend and categorize it, personalization, chat bot, or CRM. If CRM is the smallest line item, you have a reallocation conversation to start today. Evidence check, employee reported survey, Nigeria only cross-sectional design. Don't treat those R-squared numbers as universal benchmarks, directional, not definitive. Radar verdicts. Read now. The core finding AICRM outperforms chatbots and personalization as a financial predictor is immediately useful for fintech marketers building an investment case, even with the methodological limits. This next one looks abstract. Long academic name, zero empirical data. Stay with me because the business implication is genuinely big, and I think most loyalty teams are going to be caught completely off guard. Paper two. Here's the business question. When an AI agent is the one actually clicking buy, not your customer, the AI shopping for them, does your loyalty program still work? This is a theoretical paper, no experiments yet. What the researchers did is synthesize existing work on human-machine behavior, consumer psychology, and algorithmic trust to build a new formal model for how AI agents make brand choices. And here's the finding that stopped me cold. Traditional loyalty programs are built around human emotions, brand love, nostalgia, satisfaction. But an AI agent doesn't feel any of that. It picks based on rules and past performance data. So the researchers propose that when an AI agent is doing the buying, brand choice is driven by five things. How much the human emotionally values that brand, how well the AI has experienced dealing with that brand, how much the human trusts the AI, how much decision-making power they've given it, and how reliably transactions with that brand actually execute. In plain terms, your brand equity still matters, but only to the extent it's been encoded into the AI's criteria. If your brand isn't legible to a machine, it's invisible. They also propose a new metric called NHASC, net human agent score. Think of it like a trust rating that tracks whether the AI agent's purchasing behavior actually matches what the human user wanted, based on real transaction logs, not feelings. Now, here's the catch, and it's a big one. This is purely theoretical. No data, no experiments. The validation plan exists only on paper. The researchers know it, they say so. But here's what I keep coming back to. You don't need empirical validation to start asking whether your loyalty program is machine readable. That question is worth asking right now. Plain English payoff. As AI agents start making purchases on behalf of humans, your marketing needs to influence the AI's decision criteria, not just the human's emotions. Here's the business hiding inside this research. Money move. Assess whether a brand's rewards data, APIs, and signals are structured in ways that AI purchasing agents can actually parse and act on. This is basically SEO for AI agents, and almost nobody is doing it yet. Action step. Pull your loyalty program's core value props and ask, could an AI agent read these? Are your return policy, transaction speed, and reliability data machine accessible? If not, that's your gap. Evidence check. Preprint, no peer review, zero empirical data collected. The model is proposed, not proven. Use it as a strategic thinking frame, not a validated finding. Radar Verdict, watch list. The framework is genuinely novel and the problem it names is real. But without empirical validation, you're acting on theory. Track it. If the three-stage validation plan gets executed, this becomes important fast. Okay, paper three is the one I almost skipped. It's a technical AI systems paper about benchmarking. But give me 60 seconds, because what it reveals about how AI forecasting tools actually work should change how you evaluate every AI prediction product you're currently paying for. Paper three. Here's the business question. When your AI tool tells you it's predicting next quarter's trend or your campaign's likely performance, how do you know it's actually predicting and not just remembering what already happened? The researchers built a system called Hindcap. The idea? Take resolved prediction markets, things that already happened, real outcomes, and test AI models against a frozen archive of public information from before the event. No future data allowed. The AI only knows what was publicly available on Reddit up to a specific past date. Then you score how well the AI forecast matches the actual outcome, and you compare it to what the prediction market crowd thought at that same cutoff date. So here's what they found. Most current AI forecasting tests are broken. When you ask an AI today to predict something that already happened, it just looks it up in its training data. It's not predicting, it's recalling. And most benchmarks don't catch this. That is not a minor technical issue. That is a fundamental flaw in how we evaluate AI prediction products. Now when the researchers gave AI models access only to pre-event Reddit data, eight out of nine models still got better at forecasting. Retrieval genuinely helps. But, and this is the part that really matters. If the pre-event sources were just speculation and noise, giving the AI that information made it worse. Garbage in, garbage out. Except with AI, you might not notice for a while. The limitation here is real. This only tested open weight models. The big proprietary ones, the ones most marketing teams are actually using, weren't tested. So we don't know how bad the problem is for the tools you're running today. That's the part I'm genuinely bothered by. Because a lot of marketing teams are making budget and campaign decisions based on AI forecasts that may just be historical recall, dressed up as prediction. Plain English payoff. Before trusting any AI forecasting tool with real budget decisions, ask whether it's actually predicting or just remembering. And ask the vendor how they'd prove the difference. Action step. Next time you're evaluating an AI trend or demand forecasting tool, ask the vendor one question. How do you prevent the model from training on the outcomes it's supposed to be predicting? If they can't answer clearly, treat the output as historical analysis, not foresight. Evidence check. Preprint only, proprietary models untested, Reddit-based retrieval may not reflect all real-world information sources. The methodology is sound, but we're not done yet, and the most important models weren't in the study. Radar Verdict. The problem this paper identifies is real and significant, but it's an AI systems paper. Indirect marketing relevance, no marketing experiments, preprint status. Watch it. If the methodology gets adopted in marketing adjacent forecasting tools, it becomes directly important. Okay, let me pull back and tell you what I think these three papers are actually saying together. At first glance, these papers look separate. One's about fintech, one's about loyalty theory, one's about forecasting benchmark. But together they show a single pattern. We're deploying AI marketing tools faster than we're measuring whether they work, or whether they're working on the right thing entirely. The Nigeria study says AICRM predicts revenue better than chatbots, but the measurement is still employee perception, not audited results. The loyalty paper says brand strategy needs to work on AI agents, not just humans, but we haven't built a single validated way to measure that yet. Not more AI. Better measurement. That's the through line. Not more channels, more accountability for what those channels are actually doing. And here's the tension I keep sitting with. Two of today's three papers are preprints with no empirical validation. The one that has actual data is limited to employee surveys in one country. Hmm. So the research community is identifying the right problems: agentic commerce, AI forecasting integrity, AI marketing ROI, but the evidence base is still thin. Which means the practitioners who move now, who start auditing their AI tools against these questions today, are going to have a real head start by the time the validated research catches up. That's the frame I'd take into your next AI investment conversation. Here's the playbook from today. One, audit your FinTech or banking AI marketing staff. If CRM is underfunded relative to chatbots and personalization, reopen that conversation before your next budget cycle. Two, run a machine readability check on your loyalty program. Ask whether an AI purchasing agent could parse your core value props, return policy, reliability, transaction terms. If not, you have a gap that's going to get expensive. 3. The next time someone pitches you an AI forecasting tool, ask how they prevent the model from recalling past outcomes as predictions. Push for a real answer. Evidence check on all of that. One of today's papers is employee survey based with no causal claims, and two are unreviewed preprints with no empirical data. Use them to decide what to test, not what to blindly believe. Links to all three papers are in the show notes. Read the originals before making major decisions. Want the human expert take? Join Dr. Eva Wolf every Friday for the AI Marketing Radar Roundup, where she extracts no nonsense, money-making tips, practical strategy, and real business opportunities from the week's research. Subscribe on Apple Podcasts, Spotify, YouTube, and wherever you listen to podcasts. This is Evita for Big Plans Media, and I'll be back in the next radar brief.