There are so many AI support tools out there right now, and I've tried most of the big ones. They all pitch the same thing. A question comes in, the bot answers it, the ticket never happens.
What took me a while to figure out is that the thing that actually matters is where the bot runs. In your helpdesk inbox, on your website, or inside the product itself. Each one can see different things, so each one can fix different things. So I've categorized the seven tools below by where they run instead of ranking them, because that's what you're actually choosing between.
We build Frigade, which is the in-product one. I'll try to be fair to the others. The inbox agents in particular are good at what they do, and a lot of teams should probably be running two of these.
If what you're really after is getting users through the product rather than answering tickets, the product adoption platforms list is the better starting point. And if you want the mechanics of deflection itself, we wrote up how to deflect support tickets with an in-product assistant separately.
The inbox agent reads tickets that already exist and tries to close them. Intercom Fin and Zendesk AI are the two everyone has heard of. They live in your helpdesk, they've read your help center and your old conversations, and they hand off to a human when they get stuck. They do well on questions like "what's your refund policy" or "how do I change the billing email", where the answer is written down somewhere.
The website chatbot does the same job earlier, on your marketing site or help center, before there's a ticket. Ada is the big one. Most of the "train a bot on your docs in five minutes" startups also end up here. Same knowledge, different spot in the funnel.
The in-product assistant runs inside the app. It can see what screen the user is on and what state their account is in. When someone asks "how do I invite my team" it opens the invite flow and walks them through it, or just does it for them. That's what we build. If you want to know whether a tool belongs in this bucket, ask whether it can click things in your product. Most can't.
| Tool | Where it runs | What it's good at | Starts at |
|---|---|---|---|
| Intercom Fin | Inbox | High ticket volume where most answers are in the docs | $0.99 per resolution |
| Zendesk AI | Inbox | Same, if you're already on Zendesk | Add-on to Zendesk plans |
| Frigade | Inside the product | "How do I" questions that need product context or an action | Free trial, then $1K/mo |
| Ada | Website | Consumer-scale deflection before a ticket exists | Quote based |
| Decagon | Inbox | Enterprise orgs replacing a tier-one team | Quote based |
| Forethought | Inbox and triage | Routing, tagging, drafting replies across channels | Quote based |
| Plain | Support platform with AI | API-first teams that want support as code | $59 per seat/mo |
Prices are list prices as of October 2026. The quote-based ones won't give you a number until you're on a call.
Fin is the one I hear about most, and the quality is there. It runs in Intercom's inbox, reads your help center and past conversations, and answers before a human picks up the ticket. When it gets stuck it hands off to your team with the conversation attached.
Pricing is $0.99 per resolution. That sounds great until you're at volume, and then it's a real line item. Fin is the best pick when most of your tickets can be answered from docs you already have.
Zendesk AI used to be a step behind Fin and mostly isn't anymore. It answers tickets, suggests replies to agents, and routes things to the right queue. If you're already on Zendesk, try it before you buy anything else, because switching helpdesks to get a better bot is rarely worth it.
It's an add-on to Zendesk plans rather than its own product.
Frigade is the in-product one, and it's what we build. It doesn't learn your product from your docs. It learns it by using the product, the same way a new support hire would, and it keeps that knowledge current as you ship. So when a user asks how to invite their team, it can check whether they have the right permissions, open the right screen, and get them through it, or just do it for them.
It runs next to an inbox agent rather than replacing one. We have customers running Fin and Frigade together and that's the setup I'd suggest to most B2B teams. There's a free trial, then it's $1,000 per month. You can also book a free demo on frigade.com.
Ada is for consumer-scale volume on a website. It catches questions on your marketing site or help center before they turn into tickets, and it's built for companies with a lot of end users asking similar things. For a B2B SaaS app with a few thousand users it's usually more than you need.
Pricing is quote based.
Decagon is the enterprise version of the inbox agent. It's priced and sold for companies that are replacing a big chunk of a tier-one team with it, and the setup work reflects that. If procurement is involved in your support tooling, it belongs on the list.
Pricing is quote based.
Forethought is more about the plumbing than about answering. It routes and tags tickets, drafts replies for agents, and works across channels. It's a good fit when the problem is that tickets land in the wrong place or agents spend too long writing the same reply.
Pricing is quote based.
Plain is a support platform that added AI, not a bot you bolt on. It's a good fit if you'd rather wire support up over an API than click through a helpdesk, and a lot of developer-tool companies use it for that reason.
It starts at $59 per seat per month.
Pull your last 200 tickets and split them into two piles. Pile one is questions you could answer by pasting a doc link. Pile two is questions that are really "how do I do this thing in the app."
If pile one is bigger, buy an inbox agent first. Fin if you're not tied to a helpdesk, Zendesk AI if you are, Decagon if you're big enough that procurement is involved.
If pile two is bigger, an inbox agent will disappoint you. It answers those tickets with a link to the article about the workflow, the user is still stuck, and you get a follow-up ticket. For those you need something running inside the product.
I don't have a clean industry number for how the piles usually split. In the accounts I've looked at, pile two is a lot bigger than people expect. That's why I keep pushing back when someone asks which tool to buy. Usually it's two.
Depends where your tickets come from. Intercom Fin and Zendesk AI are the best inbox agents for tickets the docs can answer. Ada is the pick for high-volume deflection on a website. Frigade is the pick for the in-product category, where the assistant can see what the user is doing and take the action for them. Plenty of teams run an inbox agent and an in-product assistant side by side.
Answer them inside the product, where the user is already stuck. A bot trained on your docs replies to "how do I invite my team" with a link. An in-product assistant opens the invite screen and walks them through it, and that's the version where the ticket stops coming back.
Yes. That's the in-product assistant category, and it's what Frigade does. It learns your app by using it, answers questions with the user's actual screen and account in front of it, and can complete the steps instead of describing them. The easy check is whether the tool can click anything in your product. If it can't, it's a chatbot.
For automating the inbox, Intercom Fin, Zendesk AI, and Forethought cover routing, drafting, and resolution. For automating the "how do I" questions before they become tickets, you need something inside the product, which is where Frigade fits. Plain is the option if you want the support platform itself to be the automation layer.
Two levers. Deflection answers the question from your docs before or after it becomes a ticket, and an inbox agent or website bot does that well. Resolution finishes the thing the user was trying to do so the question doesn't come back, and that takes an assistant running inside the product. Be careful judging tools on deflection rate alone. A bot can end a conversation without solving anything.
In the same category, Zendesk AI, Decagon, and Forethought. If your problem with Fin is that it answers but users stay stuck, a different inbox agent won't help much. You need something in the product. If your problem is per-resolution pricing, Zendesk AI and Plain price differently.
The chatbot is only as good as the knowledge base it reads. Fin, Zendesk AI, and Ada all answer from your help center articles, so if an article is out of date the bot gives the out-of-date answer with full confidence. Keep the articles that cover your top ticket reasons current first, then turn the bot on. An in-product assistant like Frigade works differently, because it learns from the product itself and doesn't depend on articles being right.
Most AI support tools understand your docs, not your product. They can tell a user what an article says about a workflow. To actually understand the workflow, the tool has to run inside the app and see the screen and account the user is on. That's the in-product category, and Frigade is built for it.
