Shipping Meta's first agentic helper
From overlooked extension to funded AI platform
Despite significant investment over the years, Meta's commerce platforms have seen low engagement, stalled adoption, billions in potential revenue left untouched.
Meanwhile, a forgotten Chrome extension with no designer, no PM, and no roadmap has become one of the most-used advertiser surfaces outside Ads Manager. Nobody had noticed.
I came across it on an adjacent project, saw what it could become, and turned it into a funded AI platform now live and being significantly expanded by multiple organizations.
In plain sight
Meta Catalogs power every product ad across Facebook, Instagram, and WhatsApp, including Marketplace, Shops, and Business Messaging. It's a multi-billion-dollar system. And yet, it faces an enormous challenge: nearly 4 in 5 new Catalogs never activate.
The initiatives and tools designed and built to fix this have only had single-digit engagement rates. The surfaces Meta has invested in for years — Ads Manager, Commerce Manager, and Events Manager — have been reaching only a fraction of the product advertiser base, leaving a staggering addressable market on the table.
Meanwhile, a Chrome extension called Meta Pixel Helper, originally built for the simplest debugging, has quietly accumulated hundreds of thousands of daily active users with millions of installs. It has more daily activity than Events Manager and Commerce Manager combined, by a significant factor. And it's been on nobody's radar.
The insight
I was investigating why Catalog adoption was so low, mapping every surface where advertisers encounter product data. Commerce Manager, the primary surface for Catalog management, has many thousands of daily active users. Events Manager has roughly 5x that. Neither comes close to the advertiser base these tools are supposed to serve.
Then I noticed the Pixel Helper's numbers: 7x Events Manager, 30x Commerce Manager. It sits in the browser, exactly where catalog problems manifest, often preventing activation. More technical advertisers have organically adopted it as their primary tool for understanding what's happening on their websites. They trust it, use it daily, and yet it existed in a Meta blind spot: strategically owned by no one.
But the extension was broken, and hasn't been meaningfully updated in years. It lacks authentication, which raises some privacy questions. The UI is outdated, the information architecture is confusing, and it generates a disproportionately high support escalation rate because it often marks broken Catalog setups as healthy, and vice-versa. Advertisers use it despite its quality, not because of it.
The data was starkest with WooCommerce. Advertisers who had the Meta plugin connected outperformed those who didn't by roughly 25% on weekly spend performance, with substantially lower events coverage and selection liquidity among the non-connected. But roughly 80% of the WooCommerce addressable market wasn't connected or was running outdated plugins, which can silently break authentication, prevent product data from syncing, and leave ads serving stale prices and stock levels. The revenue gap I had anticipated wasn't theoretical anymore, but instead measurable, and growing.
That gap between organic adoption and product quality wasn't the problem. It's the opportunity that had gone unnoticed.
Toward a funded program
This didn’t start with a brief. There was no product spec, no engineering allocation, no executive mandate. It started with me noticing a pattern in usage data and making a big bet on it, with the enthusiastic support of my manager.
Scalars vs vectors
The obvious framing was: fix the broken extension. Rebuild it. Make it accurate. But a more accurate status indicator is still just a status indicator.
So I reframed it: every tool in Meta’s advertiser ecosystem has been a scalar, telling you what’s happening. Your Pixel is firing. Your Catalog has 1,247 products. Yet none of them tell you what to do about it.
The extension, properly built and leveraging Meta AI, could be a vector: a signal with direction attached. Not just “your Pixel is firing” but “your Pixel is firing, and it’s missing these three Content IDs that would improve your Catalog match rate. Here’s how to automate this fix now.”
This reframe became the strategic and conceptual backbone of everything that followed.
Your Pixel is firing
Your Pixel is firing, but it's missing 3 Content IDs. Here's how to automate the fix.
1,247 products in Catalog
1,247 products, 312 missing price. Fix now with Meta AI to unlock 25% more reach.
3 events detected
3 events detected, but PageView fires twice. Deduplicate here.
Feed connected
Feed connected, but 68% of items fail validation. Enable an ongoing auto-fix here.
From validation to investment
I pulled the usage data myself, cross-referenced it with Catalog adoption metrics, and sketched intentionally low-fidelity mocks to invite conversation. From there I wrote the foundational strategy doc and a three-phased roadmap proposal: Fix, Expand, Transform.
But a roadmap doesn’t convince anyone, visuals do. I presented across the Monetization organization, adapting the narrative for each audience: technical leverage for engineering leadership, the business case for product leadership, a new interaction paradigm for design leadership. The reaction from one engineering director mid-presentation: “When can we have it, and can it be yesterday?”
The result: a funded, cross-organizational program spanning the Catalog and Signals organizations, with 20+ engineers, dedicated PMs, UXR, and many other cross-functional partners. The extension was renamed Meta Signal Helper, heading toward Meta Business Helper, reflecting how far the scope has grown.
The vision in four scenarios
Strategy alone doesn’t move organizations. I needed people to engage with the opportunity as a narrative. I built four high-fidelity prototype scenarios using a fictional UK advertiser brand (Aerius) to make the problems and solutions concrete and emotionally real. Each scenario was deliberately directional: polished enough to be credible, not so finished that feedback stayed at the pixel level. These scenarios became the shared language of the program. Referenced in alignment sessions, roadmap planning, and cross-functional decision-making.
Expanding awareness
Previously: False confidence. Green checkmarks on broken setups.
The extension used to mark everything as healthy, even broken setups. Now it surfaces contextualised diagnostics specific to the advertiser’s Catalog: which signals are missing, why they matter, and what to do next.
Extending coverage
Previously: Toggle between three dashboards. Or call support.
Diagnosing signal issues used to mean toggling across Events Manager, Commerce Manager, and agency dashboards simultaneously. Now the extension detects the advertiser’s platform and offers insights at the right time and the right place.
Evolving diagnostics
Previously: Nothing. No resolution paths existed.
Issue resolution used to require developer access or paid support. Now the extension analyzes connected data sources, explains what’s happening in plain language, and offers recommended actions with its reasoning visible.
Expanding adoption
Previously: Nothing. Catalog setup lived in a surface nobody visited.
Catalog setup lived in Commerce Manager, which almost nobody visits. Now the extension detects a business’s product inventory automatically, generates a Catalog draft, and offers a one-click path direct to publishing an ad campaign.
Designing for AI in the browser
Scenarios 3 and 4 introduced a design challenge with no clear precedent at Meta: an AI agent that takes real actions inside an advertiser’s browser. Not a chatbot or a recommendation, but an agent that acts.
The first shipped automation flows handle Shopify and WooCommerce end-to-end. The WooCommerce flows go deepest: installing the Meta for WooCommerce plugin, installing the WooCommerce plugin itself, and detecting and updating outdated versions of either that silently break product sync.
I developed the design approach and saw it through to execution. Working with a UXR partner, I helped shape research to validate in parallel. A 7-day live study surfaced core trust barriers: advertisers couldn’t see what the AI was doing, didn’t know where on screen to look, and had no way to recover when something went wrong. These insights informed the final experience as shipped.
As other teams across organizations build their own automation "recipes", the design principles I set up are now the standard for agentic experiences across the platform: a Milestone/Task/Action scroll structure, an always-visible island showing current progress, a guided cursor directing attention to where the AI is working, an audit trail with screenshots, user-controlled pacing, and graceful degradation to step-by-step guidance on failure. This was a team effort, led by me but informed by many.
Impact
The program reached full rollout in March 2026 and is now tracking through phased Alpha, Beta, and GA for Catalog diagnostics and agentic automation flows, including the shipped Shopify and WooCommerce install and update recipes.
Reflections
Opportunities live between teams
Every team in Catalog and Signals was focused on improving the surfaces they owned. The extension wasn’t on anyone’s roadmap because it wasn’t in anyone’s org chart. The insight didn’t require genius, just looking at data without organizational blinders, and a bit of luck.
Socialization is a design skill
Having the insight was necessary but not enough. Converting it into a funded program required months of deliberate work: adapting the narrative for each audience, finding champions in each organization, building conviction without formal authority, articulating approaches to solving major technical gaps or requirements. The narratives and visual mocks weren’t just design artefacts, but key persuasion tools.
What made this hard
A toggle in the extension can represent months of authentication infrastructure, ML-powered diagnostics, and cross-platform data integration. Design leadership in a domain this complex is strategic: defining what to build, why it matters, and how to bring an organization along. The broad impact is in the thinking and the ability to connect dots that others hadn’t connected.
