MoEngage hands AI the choice of message, channel and timing: how to measure a CRM that decides on its own

Section 1 · The announcement

"Agents act. Decisioning decides.": Aampe's decisioning AI joins MoEngage

MoEngage, a customer engagement platform (email, push notifications, SMS, in-app messages), presented its NEXT Autumn '26 releases on 30 October 2026, in a post by its co-founder and CEO, Raviteja Dodda.

The headline is Aampe, a San Francisco start-up founded in 2020 that MoEngage acquired in June 2026. Deal terms were not disclosed. Its decisioning technology is now fully integrated into MoEngage, and Aampe's founding team leads this area.

MoEngage sums up the split of roles in one line: "Agents act. Decisioning decides." AI agents carry out tasks (analysing, drafting, auditing). The decisioning layer chooses what each customer receives.

Agent versus decisioning An AI agent performs a task it is given: produce a report, draft a campaign, check a configuration. A decisioning AI makes the call instead of the team: for this specific customer, which message, on which channel, at what time. It is this second piece that changes how you measure.

Section 2 · How it works

From segment to customer: the AI picks content, channel and timing, person by person

MoEngage's starting point: segment-based marketing targets an "average" customer who does not exist. The post names three obstacles: thinking in segments, learnings that reset with every campaign, and calendars driving decisions.

The CRM team's job changes. MoEngage describes it as a shift from "segment, build, schedule" to "strategize and assign". According to the acquisition release, Aampe's technology relies on reinforcement learning (multi-armed bandit algorithms) and on causal inference at the level of each individual.

The release also includes the Merlin Agent Harness, a shared workspace for marketing teams and AI agents (agent builder, MCP tools, connectors, governance rules), a global agent that proactively flags metric drops, and new data and channel features.

Section 3 · Measurement

When every customer gets a different message, the campaign A/B test goes away

Until now, a CRM team tested campaigns: subject line A against subject line B, a 9 am send against a 6 pm send. The two versions went to two halves of the base, and you compared.

With a decisioning AI, there are no longer two versions but as many journeys as customers. Comparing one subject line to another stops making sense. The question becomes: does the AI beat our current rules? And does it beat sending nothing at all?

MoEngage announces causal measurement, crediting outcomes to the messages that drove them rather than to what would have happened anyway. That is the right instinct: a loyal customer would often have bought without the reminder they received.

Causal measurement, in plain terms Causal measurement looks for the message's own effect: the difference between what the customer did after receiving it and what they would have done without it. The only simple way to estimate it across a programme is to keep a randomly drawn share of customers who do not get the treatment.

The Aampe acquisition release itself quotes a customer, the tax app Taxfix, measuring its results against a global holdout. That is exactly the benchmark a brand should keep in house, independently of the measurement the tool provides.

Section 4 · Two benchmarks

Two comparison groups to keep on the brand side

The vendor's measurement is useful, but it is the tool grading itself. To keep an independent view, we recommend two groups, randomly drawn and stable over time:

• Size: large enough to detect a meaningful gap in revenue per customer. A few percent of the base is often enough on a large base; more on a small one.
• Duration: several months. An AI that learns customer by customer is judged over time, not over a week.
• Metric: revenue per customer across the whole group, not revenue per opened or clicked message.

“The first question before switching on automated decisions: what will we compare them against?”

Section 5 · Unsubscribes

Track unsubscribes as closely as purchases

An AI rewarded on clicks quickly learns to send more. In the short term, sales go up. In the medium term, part of the base unsubscribes, marks messages as spam or turns off notifications, and that audience does not come back.

1. Define negative signals from the start: unsubscribes, spam complaints, notifications turned off, app uninstalls.
2. Set frequency caps per channel and per customer, in the tool.
3. Track the share of the base still reachable each month, by channel, in the AI group and in both comparison groups.
4. Revisit the goals given to the AI if the reachable base shrinks faster than in the "current rules" group.

Section 6 · Where to start

Where to start before switching on a decisioning AI