Agentic Ads and Small AI Models Are Reshaping MarTech Stacks

The advertising industry spent the last three years convincing itself that bigger AI is always better AI. Two developments this week suggest that assumption deserves a hard review.

Agentic Ads Are Rewriting the Paid vs. Organic Playbook

Amazon’s Alexa+ now serves ads directly inside AI-generated shopping conversations — a format Digiday describes as a glimpse of advertising’s “agentic future.” The mechanic is straightforward in principle and complicated in practice: when a user asks Alexa+ for a product recommendation, sponsored results can surface alongside organic ones, with the AI narrating both as part of a fluid dialogue.

For brands selling on Lazada or Shopee, this should feel familiar — those platforms have long blurred the line between discovery and paid placement. What’s different with agentic ads is that the AI mediates the decision, not just the display. The customer isn’t scrolling past a banner; they’re receiving a recommendation from something that sounds authoritative.

The immediate implication for MarTech teams: your attribution models weren’t built for this. Agentic interactions collapse the funnel into a single conversational moment, and most current measurement infrastructure — click-through tracking, last-touch attribution, even incrementality tests — struggles to account for influence that happens inside a language model’s response. Teams running paid search and retail media need to start pressure-testing their measurement stack against this scenario now, not after budget has been committed.

Small Language Models: The Stack Rationalisation You Didn’t Know You Needed

Meanwhile, AdExchanger reports that AI company ZeroGPU is among a wave of providers pushing small language models (SLMs) as a cost-efficient alternative to GPT-4-class tools for routine marketing tasks. The timing isn’t accidental — OpenAI has reportedly discussed lowering token costs to retain customers who are actively cutting AI usage due to expense.

Here’s the operational reality most MarTech audits reveal: brands are running LLMs for tasks that have no business requiring one. Generating subject line variants, classifying customer intent tags, populating product description templates, routing support tickets — these are pattern-matching and text-generation tasks that a fine-tuned SLM handles competently at 10–20% of the compute cost.

The smarter architecture isn’t “pick one AI.” It’s a tiered model: SLMs handle high-volume, low-complexity automation in your CRM and content ops layer; LLMs get reserved for genuine reasoning tasks — campaign strategy synthesis, audience insight generation, creative concept development. If your current setup routes every marketing API call through a frontier model, you’re not being sophisticated, you’re being wasteful.

For Southeast Asian brands managing multilingual content across Thai, Bahasa, Vietnamese, and Filipino, SLMs fine-tuned on regional language corpora can outperform general-purpose LLMs on localisation tasks and cost less. That’s a meaningful operational advantage in markets where content velocity is high and translation budgets are perpetually squeezed.

---

Consolidation at the Top Changes Negotiating Dynamics Everywhere

Omnicom Media’s recognition by COMvergence as the world’s largest media management organisation — with $75.6 billion in total billings following the OMG–Mediabrands integration — isn’t just an agency holding company story. It reshapes the negotiating environment for every brand that buys media independently or through smaller regional players.

When a single network controls that volume across platforms, the pricing and data access it secures from Google, Meta, and programmatic exchanges creates a structural gap. Independent brands and regional agencies face a harder climb to equivalent CPM rates, priority beta access, and algorithmic preference in auction dynamics. COMvergence’s data makes the scale of that gap concrete rather than theoretical.

The practical response for mid-market brands in Southeast Asia isn’t to panic — it’s to be deliberate about where consolidation actually buys you leverage versus where it creates dependency. Retail media networks like Lazada’s LazMall Sponsored Solutions or Shopee Ads operate on inventory that Omnicom’s scale doesn’t automatically unlock. Platform-native buying, category-specific PMPs, and direct publisher relationships remain genuine differentiators for brands willing to invest the operational effort.

Your AdTech stack decisions — DSP selection, data clean room partnerships, identity resolution vendors — now need to factor in this consolidation dynamic explicitly. Who owns the supply relationships your campaigns depend on, and what happens to your pricing if that intermediary’s incentives shift?

What This Means for Your Stack Right Now

Three developments, one consistent signal: the marketing technology environment is bifurcating into infrastructure that rewards scale and infrastructure that rewards precision. Brands caught in the middle — over-invested in enterprise tools they’ve under-configured, running frontier AI on commodity tasks, buying media through layers of intermediaries — are paying a compounding inefficiency tax.

The audit question isn’t “do we have enough technology?” Most teams have too much. It’s “which parts of our stack are doing work that justifies their cost, and which parts are just running because no one scheduled the conversation to turn them off?”

---

Key Takeaways

• Audit your AI tier architecture: Map every marketing automation workflow against task complexity — anything that’s pattern-matching or template-based is a candidate for SLM migration and immediate cost reduction.
• Pressure-test attribution before agentic ad formats reach your markets: Conversational AI placements will break last-touch models; build measurement frameworks for influence-without-click now.
• Treat media consolidation as a stack variable: Factor holding company scale into your DSP and programmatic partnerships — independent buying strategies on platform-native retail media networks are a structural hedge worth maintaining.

The question worth sitting with: as AI compresses the distance between customer intent and purchase decision, does the MarTech stack you’ve assembled actually move faster than the buying moment it’s trying to intercept — or is it still optimised for a funnel that no longer exists?

---

At grzzly, we spend a lot of time inside stacks exactly like the ones described here — auditing what’s earning its seat at the table and what’s quietly running up costs in the background. If the consolidation and AI shifts above are prompting a rethink of your media and MarTech architecture in Southeast Asia, we’re a useful conversation to have. Let’s talk