“I was able to paint out the full user journey for the agent, helping us to quickly understand if there are any alarming drop-offs within the funnel.”
Try searching for
As performance teams plan for 2027, there’s a question worth asking: which tasks still need your team’s attention, and which could run without it?
Reviewing networks, checking performance against benchmarks, spotting anomalies and finding opportunities take up many hours every week. That leaves less time for what actually moves the business forward, and it slows the time to action.
AI agents can take on much of that work, from the first check to a recommended next step. To be useful, they need to understand your business: your KPIs, your conversion events, and what counts as a meaningful change.
In Agent Hub, you can create an agent tailored to your own business needs, based on your own data. Write the job the way you’d brief an analyst. You write the brief and the agent runs it, on your schedule, with results in Slack or email. It recommends, you decide, and every run can be traced in the Activity tab. AppsFlyer’s AI management system is certified to ISO/IEC 42001, the standard for governance, accountability, and risk management across the AI lifecycle.
Both Agent Hub and MCP draw on AppsFlyer data and are open to anyone on the team. The difference is where the work lives, who maintains it, and how deep each can dig into the data.
In MCP…. In Agent Hub, you can start without an LLM. The data is already clean, structured, and ready for AI, allowing you to skip the prep work most teams do first. Agents work with both raw, event-level data, and aggregated data, so they can follow the full user journey, not just An aggregated view summary.
Your data stays in AppsFlyer, under your existing permissions and privacy controls. Agents work across your data (including raw data), and run multi-step workflows that end in a recommendation. AppsFlyer maintains it for you. We keep it running and improving, so there are no pipelines or prompts to fix when things change, while tokens are included.
With MCP, you connect the AI assistant you already use, like Claude, to your AppsFlyer data and build agents and workflows there, alongside your other tools. The trade-off is ownership. Once you delegate the work to an LLM, AppsFlyer becomes just the place the data is fetched from.
The model does the interpreting, and you maintain the prompts, schedules, and connections when things change. That works for one agent. At scale, the question changes from whether one agent works to whether all of them do: who owns each one, which are still running, and whether every team defines a KPI the same way. A platform like Agent Hub makes that visible across teams and keeps every agent on the same governed data, which turns individual automations into shared infrastructure the whole organization can build on and trust. Without one, you add automations faster than anyone can verify them.
Start with the tasks your team can’t realistically check every day. The following six are examples of what you can create, and they apply to almost any performance team. For work every team shares, Agent Hub already has ready-to-use agents, such as Custom executive summaries for weekly leadership reporting and Spot creative opportunities for finding top-performing creatives (requires Creative optimization).
| Agent | The job | Why it matters |
|---|---|---|
| Retargeting performance | Compare the last 14 days of retargeting campaigns with the prior 14, by media source and campaign: spend, re-engagements, revenue and ROAS. | Spot retargeting spend that does not generate any revenue before the next review. |
| Retention watch | Flag D1 or D7 retention drops against your own baseline, by source and geo. | Catch a quality problem before it becomes a budget problem. |
| Spend check | Compare daily spend by media source with the previous 7 days and flag significant moves. | Catch a source that stopped spending or scaled unexpectedly before the next review. |
| Site ID audit | Find sub-publishers sending installs but no registrations or purchases. | Cut low-quality traffic at the publisher level instead of the whole source. |
| High-value customer finder | Compare your top 20% of customers by revenue with the median in their first 14 days, by campaign. | Shift budget toward the campaigns that bring paying customers. |
A useful agent works with the KPIs your business actually uses. Here’s one agent per industry, each with a sample task you can paste in and adjust. Swap in your own event names.
For gaming teams managing multiple titles and markets, daily performance reviews add up quickly. This agent compares a recent day with the previous seven and flags what moved, so a weekend dip doesn’t go unseen.
Sample task
Compare installs, D1 retention, and sessions per user by geo and media source for the day before yesterday (the most recent day with mature D1 data) against the average of the previous 7 days. Flag any move above 15% and name the campaign behind it.
High registration volume doesn’t always translate into first deposits. This agent shows which sources bring both, and where the funnel breaks.
Sample task
For the last 7 days, show installs, registrations, and first deposits by geo and media source. Calculate the install to first deposit rate and the registration to first deposit rate. Flag sources where registrations rose but first deposits did not.
ROAS alone doesn’t show which partners bring buyers who come back. This agent follows install to first purchase to repeat purchase, partner by partner, using cohorts of consumers who have had time to buy twice. Retargeting is just as easy to measure: a second agent below follows spend, re-engagements, and ROAS across every retargeting campaign.
Sample task
For installs between the last 14 and 28 days by media source, show installs, add-to-cart, first purchases, and repeat purchases within 14 days of install. Calculate install to first purchase and first purchase to repeat purchase rates. Flag sources where installs grew against the previous period but first purchases did not.
Sample task
You are a retargeting manager measuring overall retargeting performance for this app. Include only retargeting campaigns.
If the data has a re-engagement or re-attribution conversion type, use it to identify them. Otherwise, include campaigns whose name contains rtg, ret, rte, or reattr or contains retarget, reatter, or reeng anywhere in the name.
Compare the most recent 14 days against the prior 14 days, grouped by media source with a campaign-level breakdown. For each, report spend, re-engagements, re-attributions, cost per re-engagement, paying users, revenue, and ROAS.
Flag any retargeting campaign that is spending with no measurable revenue.
List the campaigns you included and the rule that matched each, so I can confirm the filter caught the right set.Data should be up until yesterday, don’t include data from today.
An increase in installs doesn’t always mean more customers. This agent measures conversion from install to first transaction and flags partners where performance is slipping.
Sample task
Compare installs, registrations, and first transactions by media source and agency over the last 14 days versus the previous 14 days. Flag sources where the first transaction conversion rate dropped by more than 10%, and identify whether specific geos, app versions or site IDs are contributing. Use mature cohorts only.
Retention shows whether the users you acquire keep coming back. This agent compares weekly install cohorts, flags the ones where retention dropped, and points to the media sources and geos behind the decline.
Sample task
Analyze Day 1 and Day 7 retention for weekly install cohorts from 8 weeks ago to 2 weeks ago, comparing each week to the previous one. Flag any cohort where retention dropped by more than 10% versus the previous cohort. Show which media sources and geos contribute most to the decline, and which cohorts need attention first. Only include cohorts with at least 1,000 installs.
GCash, the Philippines’ leading mobile wallet, used to review only a handful of ad partners each week. With Agent Hub covering all of its active networks, the team now catches anomalies across primary and secondary KPIs. Issues that took weeks to identify surface within days, and the team reports roughly two working days saved every week across teams.
The team also created their own agent that maps their full conversion funnel. In one case, a partner that looked weak on primary KPIs turned out to drive valuable conversions further down the funnel. Rather than pause it, the team scaled it.
You don’t need to automate everything at once. Start with one task your team repeats every week, whether that’s checking campaign spend, measuring conversion rates or catching performance changes.
Now in open beta, Agent Hub lets your team create a customized agent based on what they need, built on your own data and scheduled to run automatically.