Reporting on Agentic Campaigns

The top of the Overview tab summarises how the campaign was set up and where the agent is in its learning:

Below this row sits a plain-language description of what the current phase means, followed by a progress bar across the five phases: New → Broad Exploration → Narrowing In → Optimisation → Always On. These phases are based on volume of traffic in the first instance. In the absence of sufficient volume the phase will become based on length of time the campaign has been live.

The Performance section opens with a toolbar:

Below the toolbar are headline cards comparing experience versus control:

The two primary goal cards change to match the primary goal - e.g. Orders / Conversion Rate for a conversion goal.

Below the cards is the Performance Metric trend chart - control vs experience, with Learning Milestone markers showing the points at which the agent crossed a meaningful learning threshold. anchored either on time live (e.g. after the first 7 days of learning, after the first 14 days) or on volume reached (e.g. after the first 25,000 users, after the first 250,000 users). You can switch the reporting to scope to any of these milestones - for example, measuring performance only from the moment the agent reached 25k users, or only across the last 14 days. Use them to strip out the noisier exploration period early in a campaign, or to compare like-for-like windows across campaigns at different stages of maturity.

Click View Performance Breakdown under the trend chart to expand a fuller view of the selected analysis window. The breakdown is split into two columns.

Performance for the selected window:

Alongside, Annual Projected Uplift turns the extra revenue already measured into a projection for the next 12 months. It stays hidden until the selected window is more likely than not to beat control. Once visible, it walks through three assumptions used to estimate extra revenue over the next 12 months.

These can be edited manually to show how revenue projections vary in line with changes.

Allocations are the proportion of eligible traffic directed to each experience. Unlike fixed split testing, these percentages adjust continuously as the agent gathers evidence.

The Cumulative Allocation % by Day chart shows each experience's share of traffic over time. The current allocation percentage for each experience is displayed alongside its name above the chart. For relevant strategies you can also toggle between experience and timing.

Seeing allocations shift is a normal, healthy sign that the agent is learning:

Focus on trajectory, not isolated single-day movements. Expect more fluctuation during exploration, and more stability as the campaign matures into Optimisation.

Secondary Goal Effects shows how the campaign is impacting other goal metrics beyond the primary goal.

At the top of the section are filter tabs by goal type: All, Transactional, Custom, Engagement, Funnel, Intent, Affinities.

Below the tabs is the table, with columns for Metric, Control, Experience, Relative Uplift, Probability to Beat Control, Certainty.

At the bottom of the table is the Show Credible Intervals toggle - switch this on to show the credible interval range alongside each metric instead of just the point estimate. This is the range within which the true value of a metric most likely falls, based on current data. Early in a campaign the range will be wide, reflecting limited certainty. As more data is collected, the range narrows.

This is also where transactional metrics like Conversion Rate and Revenue Per User appear when you're optimising for something else - they continue to be tracked here automatically. Each section of the report also has its own CSV download button at the top right.

The Insights tab is where you go to understand why the agent is making the decisions it's making. The Performance section on the Overview tab tells you whether the campaign is working; Insights tells you who it's working for and which signals the agent is leaning on.

Use it when you want to:

The Insights tab also surfaces the groupings the agent has chosen for itself. For example, if the agent has identified "Top Product Price Band Affinity" as a signal worth using, you'll see how it has bucketed users inside that signal (e.g. budget, mid-range), these groupings come from the agent, not from segments you've set up.

At the top of the tab is a filter icon - every panel below scopes to whatever audience you set. The total user count for the current filter appears on the right.

Three quick-question cards. Pick one to drive the panel below:

Spotlight an experience and compare its allocations against the campaign baseline across each user context. The table shows Context, % Users, Best Experience, Allocation, and the difference vs Baseline.

For each experience served, how confident the model was that it was the best choice for that user. Distribution is bucketed across Low, Moderate, and High.

A transparent log of what the agent has done within the campaign.

Three summary cards at the top:

A detailed breakdown of every activity the agent has run, listed most-recent first. Each entry shows the activity name, a short description, an optional "time saved" estimate, and a timestamp.