As an Optimizely Platinum Partner and Agent Platform specialist, we hear these questions a lot: what does Agent Platform include? And where do you start?
Here are the answers.
What is Agent Platform?
Agent Platform (Optimizely renamed it from Opal in September 2026) gives the AI its own identity: Mark. You'll see Mark as the chat interface and the in-product name wherever the platform shows up across Optimizely's products.
Mark Models are post-trained models built specifically for marketing tasks rather than general-purpose reasoning. Optimizely strips out the parameters and context a marketing agent never needs, which is how it gets to roughly ten times the cost efficiency of a frontier model doing the same job.
Mark-IQ is the context layer underneath those models. It draws on your experimentation history and web analytics, so an agent already understands your brand and your data, without a manual prompt rebuild every time.
Mark-Bench is the benchmark Optimizely built to test the models. It runs 285 tasks across 15 marketing functions, more than 6,000 evaluation criteria, on real outputs like press releases, social copy and email. On that benchmark, Optimizely's models hit a 67% pass rate at around half the cost of a comparable general-purpose model.
Optimizely groups what Agent Platform can do into three areas.
The marketer experience covers Virtual Teammates and agentic content operations, aimed at increasing output without adding headcount.
The customer experience covers personalisation, autonomous experimentation and dynamic content, aimed at turning more interactions into revenue.
The agent experience is about making sure your own content is visible and readable to AI systems, not just human visitors.
Agent visibility works differently to the other two. Optimizely's Agent Visibility Analytics uses log-level data, not estimates, to show which AI systems are visiting your site, what they're requesting and why. Combined with Optimizely's partnership with Conductor, that visibility data sits alongside search intelligence, so you can see how you're being cited, or not cited, in ChatGPT, Perplexity, Gemini and Copilot, and act on the gap.
Optimizely's Mark Eting
Where do you start?
Agent Platform isn't locked to Optimizely's own CMS. It connects to other platforms, including tools like Contentful, through a standard MCP-based connector, so it can work with a stack you've already built.
It also isn't the same experience everywhere yet. On Optimizely CMS 12, you get access to Mark's chat window, but not the tools that let an agent take action inside your content or your experimentation work. Those deeper capabilities need CMS 13 or Optimizely's SaaS platform. If you're still on 12, that's the constraint to plan around before committing budget, and it's the question we get asked most often.
Usage runs on a credit-based model, so cost scales with how much you ask it to do. Mapping that against your own use cases early, rather than after you've started building, is where a specialist partner earns their keep.
Worth the investment
Agent Platform earns its place fastest where you already have a foundation to build on: experimentation running that it can help scale, a content team producing more than it can measure, or a multi-market brand paying heavily for translation and localisation. If none of those exist yet, the agent visibility side is often the easier place to start, because it tells you something concrete about your current AI exposure before you commit to building anything on top of it.
How we help
Onboarding Agent Platform starts with knowing which of these pieces to switch on first for your set-up, and that's the judgement call we bring to every engagement.
Our agentic readiness assessment shows you where your organisation stands today, and what to tackle first.
________________________________________
Sources:
Optimizely Introduces Purpose-Built AI Models for Marketing
How ready are you for an agentic world?
Don't forget to share this post!
