Meta’s new agentic economy report reveals a 75/15 gap in ai ROI. Learn what it means for search visibility, GEO, and how SMEs can get ai-ready fast.
I watched a client’s WhatsApp inbox light up with 123,000 conversations in a single week, and not one of them was answered by a human. That is not a hypothetical. It is a real number from a real retailer, and it is sitting inside Meta’s newly released report on the agentic economy, a document I have been picking apart for the past few days because it lands directly in Ongito’s lane: helping small and mid-sized businesses stay visible as discovery shifts from search engines to ai agents.
If you only read one paragraph, read this one. The agentic economy is the shift from ai that answers questions to ai that takes action on a buyer’s behalf: discovering products, comparing options, negotiating, purchasing, and following up, all without a human clicking through each step. Meta’s report found 75% of enterprise leaders have already adopted agentic AI, but only 15% are seeing real returns. That 60-point gap is an infrastructure problem, not a technology problem, and it is exactly where SMEs can either get left behind or leapfrog the incumbents.
What Is the Agentic Economy, and Why Is It Forming Right Now?
The agentic economy is what happens when ai stops being a chatbot that waits for a question and becomes an agent that acts. A chatbot answers and stops. An agent answers, makes a personalized recommendation, closes the sale, processes a return, and reorders the product next month, all inside a single conversation thread.
Meta’s report, released this year, frames this as a structural reconfiguration of how businesses operate, not an incremental UX upgrade. Leading strategy firms project the global economic impact of agentic commerce will reach 3 to 5 trillion dollars by 2030, and Gartner has separately forecast that 60% of brands will rely on agentic AI for one-to-one customer interactions by 2028, replacing channel-based marketing as we know it. Consumers are already migrating: 55% are starting product discovery inside AI platforms instead of traditional search engines, according to figures cited in Meta’s report.
I have been saying this to Ongito clients for over a year, and it is a little surreal watching a company the size of Meta publish a report that says the same thing in bigger font. The customer journey (discovery, consideration, purchase, support, re-engagement) is becoming a single continuous loop that an AI agent manages on the customer’s behalf. If your business is not structured to be found and trusted inside that loop, you are not losing a little traffic. You are losing the customer relationship entirely.
Why Do 75% of Companies Adopt Agentic AI but Only 15% See Returns?
This is the number that should make every founder and marketing lead sit up: only 12% of CEOs report that AI has delivered both cost and revenue benefits, according to PwC’s 29th Annual Global CEO Survey, which polled 4,454 CEOs across 95 countries. Forty percent of enterprises are stuck running pilots that never scale, producing reports nobody acts on. Only 15% have reached what Forrester calls “agentish territory” in its “Mind the Agentic Action Gap” research, the zone where agentic AI actually generates measurable ROI.
Here is the pattern behind that gap, and it is not a mystery: the businesses stuck in proof-of-concept purgatory have agents that generate insights but still require a human to manually bridge the gap between recommendation and execution. The distance between “the AI suggested this” and “the AI did this” is exactly where ROI either gets built or gets eroded. It is the digital equivalent of hiring a brilliant strategist and then never letting them touch the actual project.
The fix is not a smarter model. Meta’s report is blunt about this, and I agree with it completely: your constraint is not the model, it is the infrastructure underneath it. Orchestration, unified data, and governance determine whether an agent can act or just advise.
What Five Infrastructure Layers Does the Agentic Economy Actually Require?
Meta’s report breaks this down into five capabilities that have to work together before an agent can move from answering to acting:
- Identity. When an agent represents a business or commits to something on a customer’s behalf, both sides need a verified identity to trust. This is the foundation everything else sits on.
- Relationships. Discovery runs on trust and networks. Just as people find businesses through word of mouth, agents need a directory to find other agents and businesses, which is why structured, machine-readable data matters so much (more on that below).
- Messaging. Meta’s data shows more than 1 billion messages a day move across Messenger, Instagram, and WhatsApp. People already talk to ai the way they talk to each other, through messaging, not through a search bar.
- Commerce. Catalogs, marketplaces, and checkout flows are the transaction layer. When an agent negotiates pricing and compares options in seconds, it needs infrastructure that keeps pace.
- Models and protocols. The AI models themselves are already capable and improving fast. What’s missing is a marketplace of tools and specialist agents they can call on, plus the protocols that let agents transact with each other directly.
Notice what is not on that list: a bigger, smarter chatbot. The bottleneck was never the model. It has always been whether a business gives that model something structured, current, and trustworthy to work with.
How Do AI Agents Actually Find and Choose Businesses?
This is the question every Ongito client eventually asks me, usually right after they read a headline like Meta’s, and it is the right question. When a buyer’s agent evaluates options, it queries structured data. A business without a machine-readable catalog, verified pricing, and clean schema markup, the kind documented in Google’s own structured data guidelines, simply does not appear in the results the agent hands back to the customer. Discoverability to agents is now a competitive surface, the same way page-one rankings were in 2015.
Practically, that means three things matter more than almost anything else on your site right now:
- Structured data that is accurate and not duplicated. If your schema markup contradicts itself or conflicts with what your CMS auto-generates, an ai system either ignores it or, worse, cites the wrong version of your business to a prospective customer.
- Content that answers a real question in the first sentence. Agents extract answers, they don’t read your whole homepage looking for one. Lead with the answer, then explain it.
- Named, verifiable specifics. “A photography studio in Tokyo” is invisible. “Soredenacho, a Tokyo-based photographer offering 1-hour and 2-hour sessions in Shibuya and Shinjuku” gives an agent something concrete to attach to.
(And yes, I recognize the irony of writing a keyword-optimized article about how AI agents hate keyword stuffing. The joke writes itself, and honestly, so does most of the content that still ranks like it’s 2019.)
What Are Early Movers Already Doing With Agentic AI?
Meta’s report includes case studies that are worth reading in full, but the pattern across all of them is the same: connect the agent to real backend systems, and the results compound fast.
A retail and ecommerce brand reached over 123,000 customers via WhatsApp in a single week, with 13% of conversations converting into personalized product recommendations delivered as shoppable carousels, and a median response time under seven seconds. A financial services company built for LATAM saw a 70% payment completion rate for customers who initiated payment mid-conversation. An automotive company reported an 85% resolution rate handled entirely by the agent, with no human assistance needed, and a 54% lift in conversion compared to its previous best-ever rate.
The common thread is not the industry. It is that each business connected the agent to its actual reservation, pricing, and payment systems instead of treating the agent as a glorified FAQ page. Meta’s report notes that results like these are self-reported and won’t be identical for every business, but the direction is consistent enough to take seriously.
What Should an SME Do in the Next 90 Days?
You do not need Meta’s enterprise budget to start closing the 75/15 gap. Here is where I tell clients to actually begin:
Audit your discoverability first. Search your own business by name and by category on ChatGPT, Perplexity, and google ai mode. If you don’t show up, or you show up with outdated pricing, that is your starting line, not a footnote.
Fix your schema before you write another word of content. A duplicate or conflicting schema block is worse than no schema at all, because it actively confuses the system trying to trust you.
Write content in the format an agent extracts, not the format a human skims. Answer the question in the first sentence of every section. Use headings phrased as the actual questions your customers ask.
Connect at least one workflow end to end. Pick your highest-volume, lowest-complexity interaction (booking, FAQ, order status) and make sure the agent can actually complete it, not just describe it.
Measure citation share, not just rankings. Track how often you get cited or summarized in ai-generated answers for your core topics, the same way you’d track keyword rankings. If a competitor is being cited and you’re not, that is your content gap.
What Are the Biggest Mistakes Businesses Make With Agentic AI?
I see the same four mistakes repeated across almost every business that stays stuck in the 40% “trapped” bucket Meta’s report describes:
- Treating messaging channels as one-way broadcast. Push notifications with no real conversation behind them get an initial bump, then declining engagement. The advantage is two-way, persistent context.
- Publishing content that hedges everything. “Studies suggest it might help” loses to a specific, sourced number every time an ai system is deciding what to cite.
- Skipping the boring infrastructure work. Identity, unified data, and governance are not exciting to build, but they are the actual bottleneck, not the model.
- Confusing adoption with results. Buying an ai tool is not a strategy. Only 12% of CEOs report both cost and revenue benefits from ai, according to PwC, which tells you adoption alone means almost nothing.
Frequently Asked Questions
Question: What is the agentic AI economy?
Answer: It’s the shift from ai systems that answer questions to ai agents that take autonomous action on a customer’s behalf, from discovery through purchase, support, and re-engagement, all inside a single ongoing relationship rather than isolated sessions.
Question: Why do most companies adopt agentic AI but fail to see returns?
Answer: Most stop at proof-of-concept. The agent generates a recommendation, but a human still has to manually execute it, which is exactly where ROI erodes. Meta’s report calls this the gap between recommendation and execution, and closing it requires real infrastructure, not a smarter model.
Question: How do AI agents find and choose which businesses to recommend?
Answer: They query structured, machine-readable data: schema markup, catalogs, verified pricing, and named specifics. A business without clean structured data is effectively invisible to an agent making a decision on someone’s behalf.
Question: What should an SME do first to prepare for agentic AI?
Answer: Start by checking whether you already show up when someone asks ChatGPT, Perplexity, or google ai mode about your category. Then fix schema conflicts, rewrite content to answer questions directly, and connect at least one real workflow end to end.
Question: Is GEO different from traditional SEO?
Answer: GEO builds on SEO rather than replacing it. You still need to rank. GEO adds a layer on top: writing definitive, well-sourced, clearly structured content that an ai system can confidently extract and cite in a generated answer.
Ready to Make Sure Your Business Shows Up When AI Agents Are the Ones Choosing?
The gap between the 75% adopting AI and the 15% actually seeing returns comes down to one thing: infrastructure most businesses haven’t built yet. Ongito helps SMEs close that gap, auditing your AI visibility, fixing the schema and structured data that AI agents actually query, and building content that gets cited instead of skipped.
Contact Ongito to find out where your business stands today, and what it takes to be in the 15% instead of the 75%.