Every marketing team in the country is being asked a version of the same question: how are you going to do more next year, with the same people?
The usual answers are working longer or buying another tool. Optimizely's answer is Opal and before writing it off as more AI noise, it is worth looking at what happened to the teams who actually rolled it out.
| What changed | By how much |
|---|---|
| Campaign completion time | Cut by 54% |
| Experiments run | Up 79% |
| Personalisation campaigns launched | Up 24% |
| Experiment win rate | Up 9% |
| Experiment win rate | Up 17% |
Those figures come from Optimizely's 2025 Opal benchmark study, drawn from 47,000 Opal interactions across nearly 900 organisations. The methodology matters: it measures each team against its own performance before Opal, rather than asking people how productive they feel. Teams adopting Opal broadly across content, experimentation and delivery reported close to 50% more output. Australia was the third-largest adoption market in that dataset, behind only the United States and the United Kingdom.
Adoption in APAC is well ahead of where most people assume.
Opal is Optimizely's AI agent orchestration platform.
Ask most AI tools to plan a campaign and you get an outline. You still have to build the thing. Ask Opal and it can generate the brief, build the campaign structure inside Optimizely CMP, draft the content for each channel, and route it for approval. That is the difference between AI that drafts and AI that does and it is why the time savings show up in campaign completion rather than in how fast someone writes a first draft.
You work with it in plain language through Opal Chat. Behind that chat window, Opal interprets what you are trying to achieve, enriches it with your own content and data so the output sounds like your brand, asks for anything it is missing rather than guessing, and then does the work. Results land in chat or on the Opal Canvas for review.
| The part | What it is | Think of it as |
|---|---|---|
| Tool | A single capability. Creating a campaign, updating page metadata, running a web search, generating an image. Hundreds ship out of the box. | What your team can do |
| Skill | Your instructions. Brand guidelines, tone of voice, and the rules for when and how tools should be used. Written in plain English. | Your ways of working |
| Agent | Tools and skills packaged into a specialist that completes a task on its own. | A team member who knows both |
| Workflow agent | Several agents chained together with logic and conditions to run a multi-step process end to end. | A process that runs across the team |
The order matters. Tools give Opal capability, skills give it judgement, and agents combine both into something that can be handed a job and left to it. Skills are the part teams most often skip, and they are what separate output that sounds like your brand from output that sounds like everyone's.
Four things, in about the time it takes to read this paragraph:
Opal runs on Google Gemini and Anthropic's Claude models through business accounts, so your data is not used to train the models or shared with other customers.
Strip away the vocabulary and the day job changes in fairly specific ways:
| Role | What Opal takes off their plate |
|---|---|
| Content marketer | Campaign briefs from a rough goal, on-brand copy for blogs, emails and social, and complete campaign structures built directly in CMP rather than assembled by hand. |
| CRO / experimentation | Test ideas generated from a page URL or screenshot and a stated objective, complete with problem statement and hypothesis; variation development with automatic element detection; and real-time analysis with recommended next tests. |
| Digital lead / web editor | Site audits that flag duplicate and outdated content, structured content models built from an existing URL, and metadata and schema markup applied directly in the CMS. |
| Ecommerce manager | Promotions built and configured across products and variants, catalogue translation, and bulk updates across thousands of items in a single action. |
The smaller things stack up too. Each person can teach Opal their own shortcuts and preferred style. Opal produces finished Word documents and PowerPoint decks directly, so a brief becomes a deliverable without a formatting detour. And agents can run on a schedule or trigger off an event, which means some of the work happens without anyone opening a laptop.
Optimizely's own marketing team is a reasonable proof point here. They have logged more than 32,000 hours of work through Opal, which at a blended team rate lands somewhere north of $3 million in time back.
Take AI search visibility. It is on every marketing team's list, and nobody has time for it.
Opal has three agents for exactly this. One reviews a page and tells you what is weak. One writes clear question-and-answer pairs so AI tools can quote you properly. One adds the behind-the-scenes code search engines read, straight into your CMS once you approve it.
Useful on their own. Link them together, point them at your site, and it stops being a project you keep postponing; it becomes a job that runs across hundreds of pages.
You can now reuse a workflow inside another one, so bigger automations get built from pieces that already work.
Until recently, going beyond the pre-built agents meant understanding how agents are constructed. Agent Builder and Skill Builder now sit inside Opal Chat so you describe what you want, refine it conversationally, and keep improving it as you learn what works.
So the starting sequence is simple:
A fair question about handing work to AI is knowing what happened.
Every agent workflow keeps a full version history. When something fails, the interface names the reason instead of shrugging. Editing or deleting an agent warns you which downstream workflows it affects. Any URL an agent visits is screened against Google Web Risk. And a usage dashboard shows administrators which agents are running, who is using them and how activity is trending.
Switch is a Silver Optimizely partner. We help marketing and digital teams work out which processes are genuinely worth automating, write the skills that keep output on-brand, and build the agents and workflows.
Get in touch and we will walk you through Opal and what it can do for your teams: hello@switchit.com
A skill is a set of instructions; your brand guidelines, tone of voice, and the rules governing when and how Opal should use its tools. An agent is a packaged specialist that combines tools with those skills to complete a task on its own. Put simply: skills shape how Opal behaves, agents are what carry out the work. You write skills once and every agent inherits them, which is why Opal provides a separate Skill Builder and Agent Builder.
What is a workflow agent?A workflow agent connects several specialised agents with logic and conditions so a multi-step process runs end to end, passing the output of each stage into the next. Workflows can also be nested, meaning one workflow runs inside another — so a process you have already built and tested becomes a component of something larger.
Do I need a developer to set up Opal?No. Skills are written in plain English, pre-built agents launch from the Agent Library in a click, and Agent Builder and Skill Builder let you create and refine your own directly in Opal Chat through conversation.
Can Opal connect to systems outside Optimizely?Yes. Connectors for Google Search Console, Gmail, HubSpot and ZoomInfo are available out of the box, and organisations can register their own remote MCP servers to connect agents to other systems in their stack. This is what allows a single agent run to draw on data from several platforms.
Which AI models does Opal use?Opal runs on Google Gemini and, Anthropic's Claude family of models accessed through Google Vertex AI. Both are used via business accounts, so customer data is not used to train the underlying models or shared between customers.
What can Opal do that a general AI assistant cannot?Two things. It has context your content i.e. your data, your brand rules and your history inside Optimizely. And it has reach i.e. it can execute inside the platform and across connected systems, creating campaigns, updating content and applying changes, rather than producing text for someone else to action.
Subscribe for the latest news and events from our talented team and community network.