These answers cover what the Decision Agent decides, how it treats your content, what you need to set up, how it learns, and how to get it. For the concept behind it, see AI Decisioning.
What it is
What does the Decision Agent decide?
Every time a journey is about to be sent, the Decision Agent decides four things for each user: which content variant, which channel, when, and whether to send at all. It chooses channel and time together as one decision. It learns from each user's own reactions and keeps adjusting.
Is the Decision Agent the same as Next Best Channel or Send Time Optimization?
No, and it doesn't replace them. Both stay as they are. The Decision Agent adds content variants, per-user learning from each user's reactions, joint channel and time decisions, and the no-send decision
Is the Decision Agent the same as Call an Agent?
No. Call an Agent is a separate element: you write an instruction and get one answer back per user. It doesn't learn, and it doesn't decide channel, time, or content. The Decision Agent learns per user and decides content, channel, and time.
No. Call an Agent is a separate element: you write an instruction, and it returns one answer per user that you can branch on, personalize messages with, or save to the user profile. It doesn't learn over time, and it doesn't pick between channels, send times, or content variants on its own, only when your instruction asks it to. The Decision Agent learns per user and decides the content variant, the channel, the time, and whether to send at all.
Is the Decision Agent a chatbot or a generative AI feature?
It isn't a chatbot. It uses a large language model (LLM) to label your content and write variants inside your guardrails, and machine learning and reinforcement learning to decide what each person gets. The LLM writes; the learning engine decides
Content
Do I have to tag my content by hand?
No. Insider One labels your base content automatically on six dimensions. You choose only which labels to generate variants for.
Who writes the variants?
The Decision Agent generates them from your base content, inside your Brand Voice and Content Rules. Nothing goes out until you approve it in Content Review. To set the rules, see Guardrails Setup Guide.
Why are there fewer variants than the label combinations suggest?
The banner shows the maximum number. The optional cap, the guardrail checks, and de-duplication reduce the count. Fewer, more distinct variants help the Decision Agent learn better than many near-identical ones.
Why does a variant carry no discount code?
The base content had none. A variant never invents a code, a gift, or a discount, and never changes an offer value. Put the code in the base content, and it appears in every variant that carries that offer.
Can the Decision Agent choose the offer or its size?
It decides whether a user sees the version with the offer, not which offer or what value. Per-user offer levels and offer budgets are on the roadmap.
Does the Decision Agent work in my languages?
Yes. Generation works in your languages, and variants stay in the language of the base content. Localized Content Sets follow the same flow.
Does the Decision Agent generate images?
No. Variants keep the image from your base content.
Can I edit a variant?
You can approve a variant, reject it with a reason, or regenerate a rejected variant for new copy on the same labels. To change the wording yourself, edit the base content and create a new set.
Setup
Which channels does the Decision Agent support?
Channel and time decisions cover Email, SMS, App Push, Web Push, and WhatsApp. Content variants cover Email, SMS, App Push, and Web Push. WhatsApp gets channel and time decisions only.
How many channels do I need?
You can start with one. The Decision Agent opens with two channel paths by default and supports one to five. With a single channel, it runs a content test on that channel. More channels give it more to decide between.
Can the Decision Agent run in transactional journeys?
No. It works in standard journeys only.
Does Send Time Optimization respect the frequency cap?
It behaves exactly as it does on a regular campaign: Send Time Optimization overrides the frequency cap. Silent hours and send windows apply as configured.
What happens to personalization tags?
They pass through unchanged, with the fallback values you wrote in the base content. The Decision Agent never invents a fallback.
Can I A/B test the Decision Agent against my current journey?
Yes. Put your existing flow on one path of an A/B Split and the Decision Agent on the other. Keep the split stable so a returning user stays in their group.
Can I use the Decision Agent on one message only?
Yes. One Decision Agent decides one send. For a series of messages, add one before each message.
How long does setup take?
About two weeks from kickoff to go-live. Results are readable about two weeks after go-live.
Learning and results
How long does the Decision Agent take to learn?
It doesn't start from zero. At onboarding, it reads back about three months of your account's message engagement, so most users already have a profile at the first send. It sharpens as reactions arrive. Read results about two weeks after go-live. Low volume or near-identical variants slow learning down.
Does the Decision Agent learn from messages I sent before it?
Yes. It reads back about three months of message engagement at onboarding, and every reaction after that feeds the user's profile of channels, times, and content qualities.
Is there a control group?
Yes. Run an A/B Split in the journey with your current flow on one path and the Decision Agent on the other. You read both on the goal you already track.
Can I see which user received which variant?
Yes. Your account team can provide a custom export today. A self-serve export from the InOne panel follows within weeks. Toward the end of Q4 2026, the variant each user received appears on the user profile in the panel and can be exported through the CDP API.
Why did a user get the base content?
The Decision Agent couldn't evaluate the user's data at that moment, an error occurred, or the path is on WhatsApp. The base content is always the fallback.
Why did a user get nothing at this step?
Either the user had no reachable channel among the paths, or Suppression was on and held the message back because sending wouldn't change anything for them. Held-back sends are visible in reporting.
Will the Decision Agent stop sending to people?
Only with Suppression switched on, and only for that one message. The user continues through the journey. An unsubscribe on any channel holds the send back on its own.
Pricing and access
Is the Decision Agent paid?
Yes. It consumes your account's AI Credits for each user who passes through a Decision Agent. Your account team sizes the balance from your audience size and send frequency.
How do I get the Decision Agent?
The Insider One team enables it per account. Contact your account team to get started.