Workflows vs. Fin AI: When a Simple Bot Beats an AI Agent (and How to Combine Them)

Most businesses setting up Intercom eventually run into the same confusing moment: there are two different tools that can automate a conversation, and it is not obvious which one to reach for. Workflows (what used to be called Custom Bots) is Intercom's rule based automation builder. Fin AI is the AI agent that reads and reasons over your help content. They are not competing products, and they are not interchangeable either. Knowing where one stops and the other starts is one of the more practical things you can learn if you want your support setup to actually hold up under real traffic.

What Workflows actually is

Workflows is a visual, no-code canvas where you build a sequence of steps for a conversation to follow. You drag in message blocks, ask questions, branch based on a customer's answer or attributes, tag and assign conversations, snooze them, or close them out automatically. Every path is something you explicitly designed. If a customer picks "billing" from a menu, you decided exactly what happens next. Nothing about it guesses or interprets language on its own, it just follows the tree you built.

That predictability is the whole point. You can preview a workflow before it goes live, keep a draft version separate from what customers are currently experiencing, and roll back to an earlier version if a change breaks something. For anything where the business needs a guaranteed, repeatable outcome, that structure matters more than flexibility does.

What Fin AI actually does

Fin works differently. It reads your help center and other connected content, and generates an answer in natural language rather than following a fixed script. It can handle open-ended phrasing, understand images like screenshots or error messages, personalize a response using customer data, and work across 45+ languages. When a request needs more than a single answer, Fin can also carry out multi-step actions through connected tools and business logic, rather than just pointing someone to an article.

Fin is also built to know when it should not answer. It is designed to hand off to a human when it detects certain high-risk situations, when a query calls for medical, legal, or financial judgment, or when your own configured rules say the conversation should escalate. In other words, part of what you are buying with Fin is not just answer generation, it is a decision layer about when AI should stay out of the way.

Where a rule based bot still wins

Despite all the attention Fin AI gets, plenty of situations are better served by a plain Workflow with no AI involved at all:

Lead qualification forms where you need specific fields captured in a specific order. Routing a conversation to the right team based on a dropdown selection rather than free text. Pre-chat collection of account or order details before a human ever sees the conversation. Simple FAQ menus where the "questions" are really just navigation, not genuine ambiguity. Any process where a wrong or unpredictable answer is a bigger risk than a slower one.

In these cases, the deterministic nature of Workflows is a feature, not a limitation. You are not asking the system to interpret intent, you are asking it to execute a known process reliably.

Where Fin AI earns its keep

Fin is worth the setup effort once the incoming questions stop being predictable. Troubleshooting steps that vary based on what a customer already tried. Product questions phrased a hundred different ways that all mean the same thing. Support volume spread across time zones where a human team cannot realistically cover every hour. Any place where your help center already has good answers, but customers are not finding them because they are not searching in the right words.

This is also where the resolution based pricing model matters for planning. Fin AI usage is billed on resolutions, meaning you are charged specifically when it resolves a customer's question rather than for every message it touches. That makes it worth being deliberate about where Fin is actually turned on, rather than switching it on everywhere by default.

How to combine them instead of choosing one

The two tools are meant to sit inside the same automation, not compete for the same conversation. In practice, that usually looks like this:

Use a Workflow as the front door. Collect basic context, route by department or plan tier, and handle anything that is genuinely just navigation, before AI ever gets involved.

Hand ambiguous or open-ended questions to Fin with a "let Fin answer" step inside that same Workflow. This can sit anywhere in the tree, not just at the start.

Use what Fin detects about the conversation, sometimes called Fin attributes, such as the likely topic or intent, to branch the rest of the workflow. Note that Fin has to actually be engaged in the conversation for this detection to happen, it cannot classify a conversation it was never part of.

Layer in escalation rules that combine what Fin detected with your own business data, like a customer's plan, tags, or account value, so specific segments or issue types skip AI entirely and go straight to a person.

Scope where and when Fin runs. You can attach it to specific Workflows targeting specific audiences, or limit it to certain hours, which gives you a direct lever on both cost and risk without touching your entire support setup at once.

The mistake to avoid

The most common setup mistake is treating this as an either-or decision made once at the account level. Businesses either turn Fin on everywhere and get frustrated when it mishandles a process that should have been deterministic, or they build everything as a rigid Workflow and miss the entire benefit of AI resolving the long tail of odd phrasing. The better setups treat Workflows and Fin as two tools with different strengths, wired together deliberately, conversation type by conversation type.

Getting that split right, and tuning the escalation rules and Fin attributes underneath it, is most of what separates a support setup that quietly resolves tickets from one that quietly frustrates customers. If you are trying to figure out where that line should sit for your own business, or want a second pair of eyes on a Workflow and Fin setup you already built, reach out and we can look at it together.

Cover photo by Mohamed Nohassi on Unsplash.