Meet Autopilot: The AI agent built for control, not chaos

Hi! When is Copilot and Autopilot expected to come to the Slack integration?

This is on our roadmap and expected in the next few months, but we don’t have a target release date yet. Stay tuned! 

How does Autopilot integrate with tools we’re already using? Can it pull live info such as from third-party systems (CRM, ERP, etc.)?

Autopilot can work with external knowledge sources as long as they are supported by Front, and workflows can leverage connectors and dynamic variables. API-triggered replies are possible via rule actions and branching logic, but direct live pull from all external databases (e.g., Oracle/SQL) may require custom connector setup. 

Can you connect external documents for knowledge sources (Google Docs, Notion, Coda, Salesforce, Confluence)? 

Yes, Autopilot supports connecting external knowledge sources, including Google Docs, Salesforce, Confluence, and others (max ~3,000 pages). Integration works best with English content. See this article for more details. 

Is there a way to set a blacklist so AI avoids certain Topics?

While there isn’t a formal blacklist, you can decide, Topic-by-Topic what should be automated and what shouldn’t. So if there are more sensitive Topics you do not want Autopilot to reply to, you can just skip them and not turn on auto-replies for them. 

If a reply is inaccurate, will it learn from my corrections? Or if you discard an AI reply, does AI still learn?

Autopilot continually learns from new interactions and updated conversations, improving automation accuracy over time. Direct discarding does not explicitly “train” the AI, but improvements come from updated knowledge and messages you send.

Can I edit Topics if AI was incorrect? 

To an extent – you can edit items such as the title, merge topics or remove duplicates. You cannot manually create your own topics or choose specific conversations to add to it. However, you can improve Autopilot by adding knowledge sources (via Front or external) in addition to Topics so that Front’s AI can create the best response for you. We plan to continue expanding our Topic list editing in the future. 

What if our team’s edits are wrong/outdated? How do we update knowledge?

Simply update your connected knowledge source or Topics; Autopilot will sync changes. 

If our company supports 300+ companies with unique Q&As, is setup complex?

Setup depends primarily on how well your knowledge sources are organized and how up to date they are in addition to ensuring you add all inboxes that contain these Q&As. Once your knowledge is ready the Topics discovery job is optimized for large data sets but may require some management to fine tune the list for the specific use cases you care about. After that, enabling is a simple as reviewing some sample autopilot responses and clicking enable. ​​​​​​​

Can Autopilot help different teams beyond support?

Absolutely – you can configure Topics, rules, and workflows per team and inbox to handle any questions that are repetitive. This could easily apply to Relationship Management, Sales, Billing, HR, and more. The great part is that Front will identify the Topics for you based on your conversations and you can choose which ones you want to automate. 

How do you train your language models—only on historical conversations?

We do not train any models ourselves; we work through vendors for LLM use. However, for specific features like Topics and Copilot/ Autopilot we do consider historical conversations for additional context, but these are never used in any training by us or our vendors, we have do not train clauses for our enterprise access. 

Can a reply trigger an API call to fetch more context or a document?

Yes, for every inbound we attempt either an Autopilot response or a Copilot suggested reply that considers the context of those inbounds and matches that against added knowledge sources. 

Are there use cases for Autopilot expected outside of repeatable conversations?

Autopilot can also automate repeated actions, kind of like automating a macro. So you could, for example, use Autopilot playbooks to move or escalate a message to your team, and then auto-reply to others. 

What is pricing? Do drafts in Autopilot count toward quota, or only successful resolutions?

Autopilot is billed at $0.89 per Resolution, defined as when the AI delivers a complete response without human intervention. Drafts do not count; only successful Resolutions are billed. 

Does each Topic have a single standard reply, or does Autopilot adjust per situation?

Autopilot adjusts replies based on the context of that specific interaction and the related knowledge needed to respond. If you do wish to have more consistency, setting up workflows with templates is a good option. 

Can Copilot help create templates for use in shared mailboxes?

Copilot can’t create templates directly at this time, though in the future it might be able to! For now, you can manually create templates from the responses Copilot generates. 

Can we remove the “Generated by AI” note at the end of Autopilot emails?

Disclosure messaging can be customized or turned off from the Autopilot settings, though it’s recommended to stay on to maintain compliance for the specific region you operate out of. 

If customer responses differ by customer, how granular can Autopilot replies be?

Autopilot looks at previous messages to understand how to reply, so they need to be similar enough to assess that, but we focus more on the factual information in those messages rather than the additional personalizations that certain agents may use. 

Does Autopilot auto-archive messages it interacts with?

Yes, the replies Autopilot sends will be in Resolved status after sending and no longer open. 

Can Autopilot send links automatically (e.g., password resets)?

Yes, Autopilot can include links within its replies based on connected knowledge, your responses, or your instructions in the Autopilot rules.

How is Autopilot’s accuracy or effectiveness measured?

We’ll be releasing an AI and automation dashboard that will help you measure the effectiveness of both Copilot and Autopilot soon. We expect to release it later this month, but you can subscribe to product updates to stay up to date on when this dashboard and other features are released.

How does AI decide when to escalate to a human, and is this customizable?

Autopilot automatically hands off when it cannot generate a reply, or hands off based on your branching rules to use Front AI to evaluate the message further. 

Can we create our own Topics?

No, our Topic discovery is an automated process that uses a special blend of advanced algorithms and LLMs to find conversation groupings that are turned into Topics. We are exploring the possibility of adding this functionality but we are still in the exploration phase so no guarantees this will make it onto our roadmap. 

Can we see an audit log of everything Autopilot does?

Conversations resolved by Autopilot will be visible in the “Resolved” section of the inbox, so you can always see what actions were taken and what messages were sent. In the future, we plan to add Autopilot actions to the audit log as well. 

What languages are supported?

Only English is officially supported currently, though other languages may still work as our AI vendors do support 40+ languages. 

AI drafts don’t match my tone—can we train Copilot and Autopilot to sound like us?

You cannot explicitly train Copilot or Autopilot to sound like you; however, if you keep a close eye on Front next quarter, you might see some new functionality related to this ;) 

To what extent do you think an accounting firm could make use of Front AI?

As long as you have thorough documentation you include as knowledge sources then our copilot and autopilot features will rely on this data for reply generation. Of course with autopilot we put the control in your hands to enable these and check them for accuracy. 

What are the security policies—will Autopilot access customer info? Is Front AI safe for sensitive info?

Autopilot and other Front AI tools are designed with security in mind, with data encrypted and not shared with third parties or used to train other companies’ AI. You have direct control over whether AI features are enabled, and Front’s policies aim to keep your sensitive information protected. This article goes into more detail with FAQs about Front’s AI. 

How is the risk of AI hallucinations addressed in Front?

All of our AI features operate via a complex pipeline that includes multiple evaluation steps that check for common hallucination traps and ensure any factual information only comes from knowledge sources you add. 

How can we prepare Topics so Autopilot succeeds?

Review and merge duplicate Topics, connect robust knowledge sources, and pilot with a few Topics to start. Review this article for more info and tips about Topics. 

How long does it typically take to set up and go live with Autopilot?

The setup time varies depending on the complexity of the processes you’d like to automate. You can quickly review a set of suggested replies by Autopilot compared with actual replies sent by teammates and if they are accurate, simply turn the toggle on to let Autopilot begin replying to conversations about that Topic going forward. This could take just a few minutes. However, you can also use Autopilot Playbooks to create much more tailored, customized workflows, which would take additional time that varies depending on the complexity of the workflow. 

How do we get team buy-in, especially from overwhelmed or resistant team members?

Resistant and overwhelmed team members are the best audience for this! Start with pilot use-cases one Topic at a time, highlight time saved, and offer transparency/training. We have a Front Academy course to onboard them onto the feature. 

If we operate many inboxes, can we enable Autopilot for only one or test with one?

Yes, Autopilot can be enabled for specific shared inboxes allowing targeted pilots or tests before broader deployment. See the instructions in our Topics article.