Help Scout AI Playbook: Scout, AI Drafts and What They Cost

Help Scout's AI comes with two bills. The agent-side tools sit inside the per-user price: AI Assist is unlimited from Standard, while AI Drafts and AI Summarize are unlimited on Plus and Pro with no usage fee. The customer-facing agent, Scout (formerly AI Answers), is metered through the Envoy add-on at $0.75 per resolution on Standard, Plus and Pro, billed monthly in arrears on top of the per-user price ^[1]. That split is the whole playbook. Switch on the Copilot tools your plan already pays for. Treat Scout as a variable cost, cap it, and measure it against a human seat.

Which AI feature unlocks on which plan

None of the three Copilot tools is on the Free plan. The step that matters is Standard to Plus. Standard costs $25 per user per month billed annually, Plus costs $45 per user per month billed annually, and the $20 difference per user, by our arithmetic, buys AI Drafts, AI Summarize and Topics along with custom fields, Teams and routing. Every login is a billable user, the Account Owner included, and monthly billing runs about 20% higher, as our pricing guide shows. Accounts on legacy contact-based plans keep AI Drafts and AI Assist but get no AI Topics.

Copilot: the agent-side tools in daily use

AI Drafts writes a reply draft from the team's past replies and Docs articles, and per Inbox it can also draw on Scout's sources. Help Scout's help center says it runs on OpenAI GPT-5.1. Admins switch it on per Inbox and choose greeting and closing styles, and agents need the Reply to Conversation permission. A workflow can also generate drafts automatically for conversations that match tags or subject lines. Nothing goes out without a human. The agent edits, accepts or ignores every draft.

AI Summarize turns a long thread into bullet points in the conversation header. Three limits shape its use. The summary is always written in English, it works only in the web app, and it never refreshes itself, so an agent clicks Update Summary after new replies arrive. AI Assist works on text the agent already wrote, with six actions: fix spelling and grammar, make it longer, shorter, friendlier or more professional, and translate. On Standard it is the only Copilot tool. There it polishes replies rather than writing them.

Help Scout's customer story says King Arthur Flour cut email response times by about a third with AI Drafts alone, saving roughly 2.5 FTEs a year. Treat that as one customer's result. The Copilot page also describes asking questions and triage suggestions for tags and priority, but marks both Coming Soon, so plan without them. If a team wants out, the Account Owner can ask support to switch AI Summarize off, and removing all AI from an account also goes through support.

Setting up Scout: channels, sources, roles and exclusions

Scout answers customers in Beacon chat and, since October 2026, by email. It is enabled per Beacon for chat and per Inbox for email ^[2]. Once it is on in an Inbox, it replies to every new conversation unless conditions limit it. Decide that scope first. Scout does not work on Messenger, Instagram, WhatsApp or SMS. It only answers: it cannot issue a refund, edit an order or update a record. With Scout on, a Beacon offers only the Self Service and Neutral modes.

Feed it in order of trust. Docs sites, public and private collections alike, sync immediately. Public web pages, plus the PDF, Word and Excel files found on them, need a manual re-sync after you change them. Google Docs and Sheets are added one at a time and must be shared with anyone who has the link. Short text snippets fill small gaps fast. Each Beacon can use only the sources that fit its context, so a storefront Beacon and an in-app Beacon can answer from different material.

Roles change how Scout replies, not what it knows. A role has a name of up to 100 characters and a Character field of up to 2,000 characters, and it can be assigned to specific Beacons and Inboxes, so one Scout can sound different on the website and in the app. Exclusions are the real safety rail. Help Scout advises an exclusion, not just role instructions, for medical, legal or financial questions. An excluded email skips Scout and lands in the Inbox, while in chat the customer sees Email or Chat buttons.

Before launch, run conversations under Scout, then Test, in the chat interface or with email conversations. Build the test set from your real inbox: the questions you answer most often, a few that should trip an exclusion, and one Scout cannot know. Watch how it hands off. Help Scout says Scout answers in more than 50 languages and is always labeled as AI, so test in the languages your customers actually write in.

After launch: reports and Improvements

Scout reports split conversations into Resolved, Unresolved and Human Escalation, and show the last 30 days by default. Plus and Pro keep all-time history, while Standard's history is limited by plan. Resolved means Scout answered and the customer did not signal a need for more help. Read the Unresolved and Human Escalation lists weekly during the first month. They show the gaps in your sources. They also show which topics deserve an exclusion.

Improvements let the team correct a Scout answer by giving it a better response, without editing a Docs article, and the system can draft suggested improvements from the team's own replies in the Inbox. Use a correction for a one-off fix. Rewrite the Docs article when the same gap appears twice, because AI Drafts reads Docs too and one fix then helps both tools. Developers can log every conversation Scout creates through a dedicated webhook event, available from Standard.

What a resolution costs, and when Scout beats a seat

A resolution is one conversation Scout handles without human help: the customer got an AI answer and did not escalate, search the knowledge base, ask more or press the button asking for more help. It is charged once per conversation, however many questions it covered, and a greeting or follow-up question without a full answer costs nothing ^[3]. Charges land in arrears on the next billing date. New accounts get three months of free, unlimited resolutions counted from the day the account is created. Existing customers get no trial.

The Account Owner or an Administrator can set a monthly cap counted in resolutions, with email warnings before it. At the cap, Scout is disabled until the next cycle and Beacon falls back to its other options. Size the cap from a budget by dividing your dollars by 0.75: a monthly budget of $150 allows 200 resolutions, by our arithmetic. Help Scout says prepaid resolutions save up to 33%. Its Terms add that unused prepaid bundles expire annually, so prepay only after a few billed months.

The useful yardstick is a seat. About 33 resolutions a month cost as much as one Standard user at $25 per user per month billed annually. Sixty resolutions match one Plus user at $45 per user per month billed annually. One hundred match one Pro user at $75 per user per month billed annually. At 500 resolutions a month the meter reaches $375, and at 1,000 it reaches $750. Those sums buy no judgment. The real test is whether each resolution replaced a reply your team would otherwise write.

Help Scout says Scout resolves 73% of requests on average. Its customer stories spread wide: eCatholic says it handles over 15% of conversations automatically, while EmailOctopus says AI covers 70% of its support volume. Your number depends on your Docs. Run the trial or a capped first month, then read the Resolved share in Scout reports before you raise the cap or buy a prepaid bundle.

Data use, HIPAA and the MCP connector

Help Scout's AI Transparency page says it builds on OpenAI foundation models with lightweight LoRA adapters for voice, tone and formatting, does not send account data back to train OpenAI's global model, and may use limited samples to evaluate feature quality. Its help center adds that OpenAI stores the data for up to 30 days. The Terms go further: Help Scout may aggregate de-identified customer data into Training Datasets for its AI offerings, and you opt out by emailing support ^[4]. Accounts with a signed BAA are opted out automatically. OpenAI and Anthropic both sit on the sub-processor list for machine learning.

HIPAA needs Pro, or a paid add-on on Plus with no published price, plus a signed BAA. AI features stay off in those accounts until you also sign the AI Feature Healthcare Addendum. Docs sits outside Help Scout's HIPAA coverage, so the Docs sources Scout reads should hold non-sensitive content only, and exclusions should catch anything that touches patient data. Our healthcare guide covers inbox design for protected health information.

The Help Scout MCP server at mcp.helpscout.net/mcp gives Claude, ChatGPT, Cursor or an in-house agent read-only access to conversations, customers, Inboxes, saved replies, workflows, reports and Docs. It works on any plan with API access, so Standard and up, while report tools need Plus or Pro and Docs tools need a separate Docs API key. Each user connects with their own OAuth credentials and sees only what they may see. Light users cannot connect. There is no account-wide off switch, and Help Scout's BAA does not cover your own AI provider. Our API guide has the setup.

The bottom line

Help Scout's AI is two separate decisions. Copilot is a plan decision: AI Assist from Standard, AI Drafts and AI Summarize on Plus and Pro, with no usage fees. Scout is a budget decision at $0.75 per resolution, worth it while each resolution replaces a reply your team would have written and while the monthly count stays below what another seat costs. Feed it Docs first, add exclusions for risky topics, test it, cap it, and read the Unresolved list weekly. Check the Terms' training opt-out before you switch anything on.