Agentic Campaigns are always-on personalisation campaigns that run continuously in the background. You set a strategy, choose a goal, and add your experiences - the agent handles the rest.
It groups visitors by behaviour, identifies the right moments to engage them, and continuously shifts traffic towards whichever experiences perform best. The longer it runs, the smarter it gets.
Read the guide below to find out how to set these up.
Agent Setup: Strategy
Agent Setup: Strategy
A strategy is the starting point for your Agentic Campaign. It sets the campaign’s shape - defining who the agent focuses on, how it groups visitors for personalisation, and whether it also optimises timing.
Choosing the right strategy means the agent starts with the right framework for your goal, rather than learning from scratch.
What changes by strategy?
Depending on your chosen strategy, your campaign will have certain settings auto-applied.
Each strategy applies different logic across three areas:
Eligibility; the starting criteria that determine which visitors are included, before any campaign-wide rules are applied
Segmentation; the contextual signals the agent uses to group visitors for personalisation
Timing; some strategies also optimise when to show an experience, not just what to show
What strategies are available?
Strategy | Description | Optimises | Example Response Types |
Welcome Visits | Targets very early-journey visits. Optimises first-impression and welcome experiences. | Timing & Experiences | Email Capture, Brand USP messaging, Discovery Functionality |
Returning Visits | Targets returning visitors at the start of a session. Optimises “welcome back” experiences. | Experiences | Recently Viewed with Intent Interests, Re-engagement Banner |
Browse Abandonment | Focuses on visitors who are browsing but haven’t added to cart yet. Optimises both timing and content for abandonment prevention. | Timing & Experiences | Email Capture, Discounting, Brand or Product Messaging |
Basket Abandonment | Focuses on visitors who have added to cart. Optimises timing and content for cart recovery. | Timing & Experiences | Email Capture, Discounting, Brand or Product Messaging |
Product Visits | Focuses on visitors viewing product pages. Optimises on-page content to improve engagement and conversion. | Experiences | Social Proof, USP banners, Over-product messaging, Tooltips, Discovery & Consideration Functionality |
Timing Optimisation | Focuses on finding the right moment to engage visitors, regardless of content type. Use when timing is the primary lever. | Timing & Experiences | Any experience type where delivery moment matters (e.g. Email Capture) |
Experience Optimisation | A broad, default strategy. Good for general content or functionality testing. | Experiences | Prompts, Recommendations, Live Chat |
Which strategy should I choose?
Pick the one that matches the part of the journey you’re trying to improve. If you’re not sure, Experience Optimisation or Timing Optimisation are safe starting points. You can always create a new campaign with a more specific strategy once you’ve seen how the agent works.
Strategy Defaults
When you select a strategy, campaign settings are pre-filled with recommended defaults for that strategy. You can adjust any setting after selection, but the defaults are set to match what each strategy actually needs.
Agent Setup: Goal
Agent Setup: Goal
The goal is the outcome you want to improve. It’s the metric the agent uses to judge which experiences are performing best, and to decide how to shift traffic over time.
How the agent uses your goal
Your goal becomes the campaign’s feedback loop. The agent uses it to:
Measure how the variant visitors perform versus control visitors
Calculate uplift and the probability that the difference is real
Continuously update traffic allocation across experiences based on those signals
All headline metrics in the Agentic Report — uplift, confidence, projections — are relative to your selected goal. However you're able to see how all goals are affected within 'Secondary Goals'
Goal Categories
Goals are grouped into the following categories:
Transactional - conversion and revenue events
Intent - behavioural signals that predict purchase, such as high-intent phase or add-to-cart rate
Funnel - steps within the purchase journey
Engagement - on-site interactions and content engagement
Affinities - preference and interest signals
Custom - business events you have defined yourself
Before committing to a goal, you can see real-time validation stats, when it was last tracked, and its reward rate. If a goal is rarely triggered or not currently being tracked, a warning will appear so you are not optimising against a signal the system has not seen.
Guardrail Metrics
By default, revenue and conversion are designated as guardrail metrics. A guardrail goal is monitored continuously throughout the campaign, even when you are optimising for a different goal.
Guardrail goals are marked with a shield icon throughout setup and reporting.
This is particularly useful when optimising for a non-transactional goal such as add-to-cart rate or intent phase, while still maintaining visibility over revenue performance.
Why goal choice matters
Changing the goal changes how the agent behaves. A campaign optimising for Revenue Per User may route traffic differently to one optimising for Conversion Rate, because the best-performing combination of experience and audience can differ depending on what you’re measuring.
Choose the goal that most closely reflects the outcome you actually care about.
Agent Setup: Customisation (Optional)
Agent Setup: Customisation (Optional)
You have the ability to further customise your Agent with the following options;
Custom Prompt
Add a custom prompt to steer how the agent reasons about your specific business context. For example: "This campaign is aimed to reduce discounting. Prioritise experiences that reinforce product value over discount-led messaging." or "Focus on customer affinities and price". The agent factors this in alongside the strategy and experience data when making allocation decisions.
Optimised Control - enabled by default
Optimised Control automatically adjusts the size of the control group over time as the agent's confidence in results grows. Rather than fixing the control at a set percentage, it starts at a higher share and moves down as certainty increases, minimising the cost of learning while maintaining statistical stability.
Within the set-up, you are able to set the minimum and maximum share boundaries, as well as disabling this option and setting the control traffic.
Optimisation Speed
Controls how quickly the agent commits to a winning experience. The default is Balanced, which is suitable for most campaigns.
Conservative: learns more cautiously. Closer to the previous default behaviour.
Balanced: recommended for most campaigns.
Aggressive: commits to winners faster. Intended for high-traffic campaigns with short durations, such as weekend flash sales. Note that faster commitment carries a higher risk of acting on an early signal that may not reflect long-term performance.
Adding Experiences
Adding Experiences
Once you've chosen your strategy and goal, it's time to set-up your experiences. These are the content variations the agent will test and optimise. You create them, the agent decides who sees what, and when.
Agentic Campaigns run as standard campaigns, so experiences are individual pieces of content shown to visitors based on the agent’s real-time decisions.
How to add experiences
You have three options of how to build experiences in the platform
Pre-Built Templates; choose a template and customise it with preset variables. The fastest way to get started.
Custom Code; write your own HTML, CSS, and JS for full creative control.
Third Party Triggers; allowing you to build experiences within your personalisation or testing tool of choice
How many should I add?
The minimum number of experiences required depends on the strategy you have chosen. Some strategies require at least two experiences to function correctly. The setup will prevent you from proceeding if the minimum is not met. The agent enforces a maximum of 5 experiences.
Multi-Experience toggle
Each campaign has a Multi-Experience setting that controls whether a single visitor can be shown different experiences across pages/sessions (On) or stays locked to the same one (Off).
Off is the right default for intrusive, once-per-journey moments (welcome pop-ups, basket abandonment overlays)
On is the right choice for always-on optimisation strategies (Product Visits, Experience Optimisation, Returning Visits)
Setting that are auto-applied can be found in the Strategies guide.
How do controls work?
Agentic Campaigns always include a control group by default, so the agent has a baseline to measure against, and you have the ability to make this a Custom Control.
Optimised Control will be enabled by default, meaning that the Agent will automatically adjusts the size of the control group over time as it's confidence in results grows. Rather than fixing the control at a set percentage, it starts at a higher share and moves down as certainty increases, minimising the cost of learning while maintaining statistical stability. You are also able to disable this and set the control as a certain traffic % in the Agent Set-Up.
We also suggest adding a control as an Agent Experience, which allows the Agent to route traffic to this if that is the right experience for a segment, or if the control outperforms all experiences. Click 'Add to Variants' in the set-up screen to do this.
Do I need to set up targeting or segments?
No - that’s the agent’s job.
Based on the strategy you’ve chosen and the content you provide, it learns which experience works best for which visitors.
You can still influence who is included in the campaign using campaign-wide global rules (see Campaign Rules & Targeting). The agent then forms smaller groups within that audience for personalisation.
Campaign Rules & Targeting
Campaign Rules & Targeting
Once you've set-up your campaign, you are then able to set Global Rules to define your global matching criteria.
Global matching criteria are your campaign-wide targeting rules. They define which visitors are eligible for the campaign, and which visitors the agent learns from. For example, you may want to exclude visitors on a certain page, or from a certain traffic source.
Think of them as the boundary you set. The agent then works within that boundary, forming smaller groups for personalisation.
Within this area, you can define;
Page Targeting; which pages your campaign will fire on, either by including pages, or excluding them. By default campaigns will fire across all pages
Segment; apply a segment for your campaign to match against
Device Targeting; which devices the campaign should fire on (All, Desktop or Mobile)
Schedule; whether the campaign should start running immediately, or from a specific date and time
Holdback Group; whether there should be a Holdback Group in the campaign, in order to help you report on its success. This is disabled by default for all agentic campaigns
Campaign Dependencies; rules that control whether a campaign can trigger, based on whether a visitor has or has not already been exposed to another campaign or experience.
Previewing a Campaign
Previewing a Campaign
To preview a campaign, go to the Campaign Manager, and find the campaign you wish to preview.
Then select the eye icon. If there are multiple experiences, or experiment variants, you will need to choose which you would like to preview.
The preview will then load onto the URL you specified in the campaign set-up.
Saving & Setting an Agentic Campaign live
Saving & Setting an Agentic Campaign live
As you build your campaign, you are able to save your progress at any time by clicking 'Save Campaign' in the top right of the editor.
Once you are ready to set your campaign live, select 'Start Campaign'. If your campaign is scheduled to start immediately it will begin to fire. If it is scheduled for a specific date and time, it will begin to fire from then.
Once your campaign goes live, the agent will then go through the following steps:
Data check; It will confirm there's enough recent data to learn from. If not, it will try again in 4 hours.
Please note, we require at least 1000 users and at least 10 'goal events' to match the strategies logic in order to go live
Experience check; The agent will understand the content, behaviour and aim of the experiences you've added.
Audience grouping; It will apply your global rules and create the segments to receive your experience.
Timing setup (if enabled based on chosen strategy); It will create moment-based groups to optimise when experiences are shown, not just what is shown.
Learning begins; The campaign will go live, and the agent will begin comparing performance across experiences and visitor groups. Shifting traffic automatically toward what works best.
What can I change mid-campaign without resetting the agent?
Safe to change anytime:
Control share percentage
Global rules (page targeting, device, schedule)
Pausing/resuming the campaign
Adding a new experience (the agent folds it into exploration)
Changing the goal metric
Will reset or heavily disrupt learning:
Changing the strategy (the agent re-segments from scratch)
Removing or substantially editing an existing experience (the agent loses its history on that arm)
If you need to change the goal or strategy, we recommend duplicating the campaign and either re-starting the campaign, or let them run together using a Campaign Traffic Allocation Group.
Running Agentic Campaigns with other Campaigns
Running Agentic Campaigns with other Campaigns
Can I run an Agentic Campaign alongside other campaigns?
Overlapping campaigns
Yes, but be mindful of audience overlap and how it may affect learning and performance. If this is a concern, create an Exclusion Group in Campaign Settings.
See Campaign Conflicts for full details on managing overlap between campaigns.
Side-by-side comparison
We also allow you to run an Agentic Campaign and a Custom Campaign side by side to compare performance. Use a Sample Share Group in Campaign Settings to split traffic between them for a more controlled comparison.