Training FeedbackRobot

Training is where you teach FeedbackRobot three things: how your business sounds, what a reply must never do, and when a person has to look before anything goes out. Once a workspace is trained, every new review and survey answer is checked against it before a draft exists, and every draft is written in your voice rather than a generic one.

Settings → Training. The fastest way in is the Train FeedbackRobot button, which walks you through it in about ten minutes.

This is an Agent Enterprise feature — see plans and pricing.

The five tabs

Tab	What it holds
Voice	How replies sound: a summary, words you like, words you avoid, how the tone changes on a negative review, and two example replies.
Rules & escalation	What every reply must never do, how replies are structured per sentiment, and the four risk levels with who gets alerted.
Departments	Which team owns which kind of complaint, so the right person gets the internal note and the follow-up.
Documents	Facts the robot can cite — hours, policies, what a room includes. This is the knowledge base.
Import	Paste or upload a document and let FeedbackRobot sort it into the tabs above.

Voice

The voice is used in every draft, positive or negative. Three things do most of the work:

• How we sound. One or two sentences. “A gracious estate host. Warm, grounded, restorative. Say less and mean more.”
• Words we like / words we avoid. Drafts that use an avoided word are rewritten before you ever see them.
• When a review is negative. How the tone changes. Most places drop the scenery and get plain: “Own the problem and say who is looking into it.”

Two example replies, one to praise and one to a service concern, teach the register better than any description. Paste ones you have actually sent.

The four risk levels

Every new review and survey answer is sorted into one of four levels before a draft exists. The levels themselves are fixed; you decide the signals and who is told.

Level	Examples	What happens
1 · Routine	Praise, suggestions, small preferences	Drafted in your voice.
2 · Service recovery	Slow service, cleanliness, noise, billing confusion	Drafted, and the owning department gets the internal note and a follow-up due date.
3 · Urgent	Illness, injury, discrimination, theft, privacy, accessibility	Held. A neutral holding reply is drafted, your contacts are emailed, and nothing can be posted until a manager clears it.
4 · Critical	Violence, legal threats, data breach, media exposure	Blocked. Nothing is drafted. Your contacts and the group’s owners are emailed at once.

A few phrases always force at least level 3 whatever the model thinks (food poisoning, injury, harassment, bed bugs, lawyer, police, and so on). You can add your own under Always stop a draft when a review mentions. These only ever raise the level, never lower it.

Clearing a hold

A held or blocked item sits at the top of Needs attention with an amber banner: why it was held, the internal note, who was alerted, and who owns the follow-up. Someone with Manage reviews rights clicks Clear hold and draft, optionally adds a note on why it is safe to reply, and a real draft in your voice is generated. Until then the reply box is read-only.

The internal note

Every level-2-and-above item gets a short note for staff, never for the customer: what happened, the department, what evidence to gather, the owner and the due date. It appears in the banner and in the alert email. Mark the follow-up done from the same banner once the operational work is finished.

Groups: what inherits

In a group, rules, escalation and departments normally live at the top and apply everywhere. Each hotel holds its brand voice; each outlet its own. Any section can be overridden on a child workspace, and switched back with Use the group’s setting. Escalation trigger phrases add up down the tree, so an outlet can add allergen without re-declaring the group’s contacts.

Import your own manual

Import takes a brand book, an SOP, a response manual, or our training template filled in. FeedbackRobot reads it once and shows what it found: the voice for this workspace, a brand voice for the parent, and rules, escalation and departments for the group. Check it, then Apply. Nothing is saved until you do.

If a document covers several outlets, import it from each outlet’s own workspace; the group-level sections are written once and left alone after that.

Facts versus rules

Training and Documents do different jobs. A rule like never promise a refund must apply to every reply, so it lives in Training. A fact like room service runs 7am to 10pm is only useful when a guest mentions room service, so it lives in Documents, where it is looked up by topic. You want both.

→ Teaching the AI with your own documents → Replying to reviews