Front AI: Complete Guide to Features, Pricing & Limitations (2026)

Front markets its AI with a simple promise: Autopilot can "resolve up to 70% of requests". It's a good line. It's also the one number Front has never actually published, and once you go looking, I keep finding a few other gaps too.

I've spent the last few years on calls with support leaders weighing exactly this decision, so I know the questions that actually get asked. Most people land here in one of three spots:

1. You're already on Front and deciding whether to switch on Copilot or Autopilot, and what it'll really cost once the add-ons stack up.
2. You're shopping helpdesks and weighing Front's AI against Intercom Fin, Zendesk AI, or an agent that plugs into whatever you already run.
3. You're trying to model the cost of Autopilot and hitting a "Talk to Sales" button where the real number should be.

This guide breaks down what Front AI is, how it works, the channels and languages it covers, its real limitations, what it costs, the resolution rate you can actually expect, and who it's right (and wrong) for. Let's get into it.

What is Front AI?

Front is a customer-communication platform built around a shared inbox. Email, chat, SMS, WhatsApp, Slack and social all land in one collaborative workspace that sits somewhere between a team mailbox and a full help desk. It was founded in 2013 by Mathilde Collin and Laurent Perrin, has raised around $204M, and says it now serves 9,000+ customers at $100M+ ARR.

"Front AI" is the umbrella brand for everything AI inside that platform, organized around three loops. Automate is Autopilot, Front's autonomous agent, plus Playbooks (multi-step workflows) and Resolve (in-app self-service).

Assist is the agent-productivity layer: Copilot for suggested replies, Compose for drafting, Summarize and Translate. Analyze is Topics (AI clustering of conversations), Smart QA (auto-scored quality) and Smart CSAT (satisfaction scoring without a survey).

The one-line version: Front AI is assistive-first. The productivity tools (Copilot, Summarize, Smart QA) are mature and genuinely useful. Autopilot, the agent that resolves tickets on its own, is newer and more cautious by design.

Front's rollout tells that story. AI Compose and Summarize shipped in March 2023, the "Front AI" brand was unified in July 2025, and Autopilot only reached general availability, with Playbooks and Resolve, in late 2025 into 2026. Front also acquired Idiomatic in November 2024, a voice-of-customer analytics company (not a voice/telephony one, which trips a few people up), and there's no first-party Front voice agent.

Its own users rate the platform highly. Front holds a 4.7/5 across 2,407 reviews on G2, mostly from small-business and mid-market teams.

G2: Front scores 4.7/5 from 2,407 reviews on G2, strong for the platform overall, though AI-specific feedback runs more mixed. One reviewer notes "the AI chatbot says it's applying rules but doesn't actually do so."

How easy is it to set up Front AI?

Setting up Front AI is admin-driven, and only company or workspace admins can enable it. The documented Autopilot setup flow runs roughly like this:

1. Build Topics first. Front AI scans up to 10,000 conversations from the past 30 days and auto-suggests Topics; you review, rename and merge duplicates. Front puts Topic accuracy at around 85%.
2. Enable Autopilot in workspace settings and pick which shared inboxes it runs on.
3. Connect a knowledge source (the Front knowledge base, a public URL, or Notion/Google Drive/SharePoint).
4. Review the "Ready for auto-replies" Topics, compare the AI's sample drafts against real teammate replies, and switch auto-replies on one Topic at a time.
5. Set the AI disclosure text appended to every AI reply.

Two constraints matter here. Autopilot and Copilot only run on shared inboxes, so anyone working out of an individual inbox is out of scope.

There's also a warm-up. Reviewers report the AI takes roughly a day after you connect knowledge before its answers get genuinely useful, and word-on-the-street is that the real setup work (cleaning Topics, writing rules) is heavier than a one-click toggle implies.

What channels does Front AI work in?

At the platform level Front is genuinely omnichannel: email, Front Chat, SMS, WhatsApp, Slack, Facebook, Instagram, and voice through integrations. The AI coverage across those channels is where it gets uneven (and easy to miss on a feature list).

Autopilot is the most complete, working across email, chat, SMS, WhatsApp, Slack (including Slack Connect) and custom channels. Copilot covers email, chat, portal, Slack, SMS and WhatsApp, and Compose's "Generate" is email-only. Front's Copilot and Smart CSAT help articles don't list Facebook or Instagram as supported, which suggests AI features are limited on Meta's social channels even though those channels exist as inboxes.

Voice is the clearest gap. Front integrates with Dialpad, Aircall and RingCentral, and partners provide call summaries, but there's no first-party Front voice AI agent. If native voice deflection is on your roadmap, Front won't do it for you.

What are the limitations of Front AI?

This is the section that decides things, so I'll be specific. There are eight limitations I'd weigh before committing.

1. You can't buy it standalone. Front AI only exists inside Front, so adopting it means moving your support onto Front. If your team already lives in Zendesk, Intercom or Jira, that's a full platform migration to take on.
2. Knowledge sources are capped. Front crawls up to ~3,000 pages from a public website, re-syncs manually at most once every 24 hours, and can't reach password-protected internal wikis. There's no dedicated way to upload arbitrary files, so third-party content comes through the Notion, Google Drive or SharePoint connectors.
3. The AI is officially English-only. Front is direct about it: "Only English is officially supported at this time. While it is possible to use this feature with other languages, unexpected results may occur." That covers Autopilot, Copilot, Summarize, Topics and knowledge sources.
4. The real Autopilot price still needs a sales call. Front's pricing page now shows Autopilot "starting at $0.05/conversation", an improvement on the old "contact us," but that's a floor and it routes to "Talk to Sales" for anything real. The assistive AI also stacks per seat on top: Copilot at $20, Smart QA at $20, Smart CSAT at $10.
5. There's no simulation or sandbox. You can't replay past tickets to forecast how the AI would have handled them before go-live. Front AI has to be switched on directly against live customer conversations, which is a real evaluation risk if you've been burned by an AI rollout before.
6. It's a black box you can't correct. Front says outright that it "does not train any models" itself, and that discarding a bad answer doesn't train the AI. Improvements come only from updated knowledge and the messages your team sends, so there's no direct "teach it the right answer" lever.
7. Shared inboxes only. Copilot and Autopilot don't run in individual inboxes, so anyone whose work happens in a personal inbox is outside the AI's reach.
8. Reviewers are lukewarm on the AI specifically. A Capterra reviewer sums up a common note: "AI features are pretty limited right now." The platform scores well; the AI layer draws more caveats.

What knowledge sources can I train Front AI on?

Front says its AI features are "only as powerful as the information they can access". I read that as the polite version of "garbage in, garbage out." Front supports four main knowledge types:

• The Front knowledge base, its recommended "most reliable source," with updates flowing into the AI within minutes.
• A public website or external URL, crawled up to the ~3,000-page cap, with manual re-sync and no support for protected sites.
• Third-party connectors: Notion, Google Drive and Microsoft SharePoint.
• Historical conversations, always on, so Front learns tone and common resolutions from past tickets.

There are supporting pieces too: admin-written Notes to patch gaps, attachment reading for PDFs and images, and fact-invalidation to trash a specific fact so it stops being used. What's missing is a dedicated file-upload type and any per-agent knowledge scoping, since sources are configured per workspace.

One thing to flag for anyone earlier in their AI journey: Front's model assumes you already have a solid knowledge base to point it at. If you don't, you're stuck at the starting line.

What features does Front AI have?

Front AI breaks into three layers, and here's what each one actually does (I'll skip the marketing gloss).

• Autopilot, the AI agent. Front's headline agent auto-resolves conversations on the Topics you enable, drawing on past conversations and connected knowledge, and hands off to a teammate when it can't answer. It supports mid-conversation "step-in," matches your team's tone, and reads PDF and image attachments. Autopilot Playbooks extend it into plain-language multi-step workflows, and Autopilot Resolve embeds self-service inside your own app or website.
• Copilot, Compose, Summarize, Translate, the assist layer. Copilot suggests replies drawn from similar past conversations (Front reports one beta customer, Fathom, used over 65% of the suggestions). Compose drafts and adjusts tone, and translates into 15 languages, rate-limited to 200 requests per user per day. Summarize auto-generates a conversation summary once there are four or more messages (and keeps updating it as the thread grows).

• Topics, Smart QA, Smart CSAT, the analyze layer. Topics cluster conversations for both reporting and Autopilot routing. Smart QA auto-scores every closed ticket against up to 15 custom criteria, and Smart CSAT infers satisfaction on every meaningful conversation without a survey.

Two things are worth knowing (both easy to miss on the marketing pages). The older AI Answers and AI Tagging products are legacy, deprecated in favor of Autopilot and Topics.

Front also runs a single tenant-level agent shaped by Topics and rules, where some competitors market a set of separate bots. Tone-matching is automatic, and Front is explicit that you "cannot explicitly train Copilot or Autopilot to sound like you" beyond that.

How do I improve Front AI responses?

Front's improvement model is supervised curation. The levers are keeping Topics clean (merge duplicates, review the similarity score before auto-enabling), keeping connected knowledge fresh, using fact-invalidation to remove bad facts, enabling auto-replies selectively Topic by Topic, and leaning on Smart QA as a feedback signal.

What you won't find is a fine-tuning pipeline or a labeled-feedback loop. Front works through model vendors rather than training anything itself, so "improving" the AI really comes down to better Topics and better knowledge.

What resolution rate can I expect from Front AI?

Here's the part I always flag: Front has never published a headline resolution or deflection rate. Intercom claims Fin resolves around 56-62% of queries; Decagon, Ada and Fini routinely cite 70-80%. Front's marketing instead argues that "deflection is cheap, churn isn't", and that chasing a high deflection number is the wrong goal.

The closest Front comes to a number is operational. A "resolution" is a billing event: an AI reply followed by no customer follow-up within 24 to 72 hours (so a wrong answer the customer simply gives up on still counts). That's a billing definition, and accuracy never comes into it.

Its case studies lean on time saved instead: Branch Insurance reports 482% ROI and 75% faster responses after switching from Zendesk, Boundless Immigration saved 10,000+ hours a quarter, and Reed & Mackay hit 97% CSAT.

For context on where that leaves Front, our own 2026 vendor scorecard put Front in the 41-56% band, against 84% for My AskAI and 74% for eesel. Across the wider field, industry resolution rates cluster around a median of 70%, though I'd treat that as a directional aggregate (every vendor defines resolution differently, and the label alone can move the number by more than ten points).

So Front AI today is more assistive than autonomous. If you need a defensible "X% of tickets handled fully by AI" figure for a board, Front can't give you one.

Front frames that as responsible restraint, and I'd push back gently: my own view is that near-total automation is achievable and better for customers when the AI has the right access and a system that flags what it struggles with. Not publishing a number is a choice, and buyers who want a measurable one have options.

What AI model does Front AI use?

Front is specific here. Its AI features FAQ names the providers directly: OpenAI and Azure OpenAI's GPT models, plus Mistral models hosted on Amazon AWS.

There's no Anthropic in the mix, and Front doesn't disclose specific GPT versions. Front accesses these through vendor APIs, trains no foundation models itself, applies zero data retention at the API layer, and confirms customer data isn't used to train the vendors' models.

For what it's worth, "what model do you use?" is a question I'd gently push back on for any AI agent. A real agent fires ten or more orchestrated model calls between question and answer (one to summarize the conversation, one to detect language, one to apply guidance and tone, one to decide whether a workflow should run), each tuned for a different job.

What languages does Front AI work in?

I went through each AI feature's help doc, and the answer comes back the same: officially English only. Front repeats the line across Autopilot, Copilot, Summarize and Topics: only English is officially supported, and while the underlying vendor models handle 40+ languages, results outside English are unsupported. Topics can interpret conversations in other languages but generate the Topics themselves in English.

The exception is Translate, which auto-detects and translates across email and chat channels. Translation still isn't the same as an agent that natively understands and answers in the customer's language, and for that, Front's own documentation sets the ceiling at English.

How secure is Front?

Security is one of Front's genuine strengths. I'd put it among the most mature platforms in this category. Front holds SOC 2 Type II, ISO 27001, GDPR, CCPA and HIPAA compliance (HIPAA with a signed BAA), offers US or EU data residency, and encrypts data with AES-256 at rest and TLS 1.2 in transit.

It runs SAML SSO, 2FA, IP allowlisting, annual penetration tests and a HackerOne bug bounty, and applies zero data retention at the AI model layer. If your security review is strict, Front will pass it comfortably.

Who is using Front AI?

Front says it has 9,000+ customers, and its review base skews small-business and mid-market, with a thinner slice of enterprise. The customers it spotlights specifically for AI usage sit in a clear set of verticals: logistics, travel, immigration, insurance and B2B SaaS.

• Boundless Immigration: Autopilot, Topics and Copilot; 10,000+ hours saved per quarter.
• Uber Freight: organic adoption of AI Summarize across a large logistics operation.
• BentoBox: knowledge base plus Copilot, with a 50% improvement in handling time.
• Central Storage & Warehouse: Autopilot Playbooks cut scheduling time by 90%.
• Reed & Mackay: 97% CSAT and a 43% productivity gain.
• Branch Insurance: 482% ROI after switching from Zendesk.

One pattern I'd point out: even on Front's own customer page, most testimonials lead with collaboration, routing or Copilot summaries over Autopilot autonomously resolving tickets. Boundless, the headline Autopilot reference, still frames it as AI surfacing answers under human control.

How much does Front AI cost?

Front's pricing is hybrid: per-seat for the platform and assistive AI, per-conversation for the autonomous agent. The base plans (billed annually) are:

• Starter, $25/seat/mo (up to 10 seats): Topics, Compose, Translate and Summarize included; Copilot, Smart QA and Smart CSAT are paid add-ons.
• Professional, $65/seat/mo (up to 50 seats): same AI inclusions, plus omnichannel, more rules and SSO/SCIM.
• Enterprise, $105/seat/mo: Copilot, Smart QA and Smart CSAT included; Autopilot still an add-on.

The AI add-ons stack on top: Copilot $20/seat/mo, Smart QA $20/seat/mo, Smart CSAT $10/seat/mo (the last two bundle at $25), and Autopilot from $0.05/conversation with the real rate behind "Talk to Sales."

Two things catch people out. The add-ons compound: a 10-seat Professional team adding Copilot and Smart QA pays 10 × ($65 + $20 + $20) = $1,050/mo, a 61% premium over the $650 base. And the full AI suite on Professional works out to $115/seat, more than the Enterprise base of $105.

Because Autopilot is priced per conversation and gated behind sales, you also can't fully model your bill from the page. Here's an illustrative monthly cost I'd sketch for a 20-agent team on Professional handling 10,000 conversations:

Twenty seats on Professional with Copilot runs 20 × $85 = $1,700/mo. Autopilot handling all 10,000 conversations at the $0.05 floor adds roughly $500/mo. So the illustrative total is around $2,200/mo, before you add Smart QA or Smart CSAT.

Does Front AI have a free trial?

Yes. Front offers a 14-day trial with no credit card required, and it mirrors the Professional plan. There's a real catch for AI evaluation, though: the trial includes Topics, Compose, Translate, Summarize and a throttled Copilot, but leaves out full Copilot, Smart QA, Smart CSAT and Autopilot.

Play video

So you can't self-serve trial Front's flagship AI agent. To pilot Autopilot, you go through sales, and there's no free tier either (the lowest paid entry is Starter at $25/seat).

Is Front AI worth it?

It depends entirely on where you're starting from.

Front AI is worth turning on if you're already on Front and you want assistive AI (Copilot suggestions, Summarize, Smart QA) with a human firmly in the loop. The sensible path is to start with Copilot before paying for Autopilot. Front's control-first philosophy genuinely fits high-stakes B2B work in immigration, freight and financial services, where blindly deflecting tickets is the bigger risk.

It's a weaker buy if you're shopping greenfield for best-in-class autonomous AI, if you support customers in more than one language, if you're cost-sensitive and scaling fast, or if you're running high-volume ecommerce support where deflection economics dominate. In those cases the assistive-first design, the English-only ceiling and the stacked add-on pricing all work against you.

Here's how Front and My AskAI stack up at a glance:

What are the Pros and Cons of Front AI?

Pros

• A genuinely good shared inbox with useful assistive AI: Copilot, Summarize and Smart QA are mature and land well for agent productivity.
• A control-first design that suits high-stakes B2B: for immigration, freight and finance, the human-in-the-loop default is a feature.
• Strong, mature security: SOC 2 Type II, ISO 27001, HIPAA with a BAA, and US/EU residency put Front among the best in the category on compliance.

Cons

• No published resolution rate and an assistive-first agent: Front leans on hours-saved case studies rather than a deflection number (see the resolution-rate section).
• Officially English-only AI: a hard ceiling for any multilingual support team (see the languages section).
• Autopilot's real price is still sales-gated, and the add-ons stack: the $0.05/conversation floor plus per-seat Copilot, Smart QA and Smart CSAT make the true cost hard to model (see the pricing section).
• No simulation sandbox: you enable the AI on live conversations, and the trial excludes Autopilot (see the limitations and free-trial sections).

If you want a wider view of the field and how My AskAI compares, our Front alternatives and vendor comparison is worth a read alongside this guide.

FAQs