Intercom Certification flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

Deflection rate

The share of conversations Fin handles without escalating to a human, calculated as Fin-resolved conversations divided by all conversations Fin handled. It is tracked separately from customer satisfaction scores.

Launch sequence

Connect content sources, configure Fin's tone and escalation rules, test in a sandbox, then enable one channel at a time, starting with the highest-volume repetitive channel.

Fin knowledge scope

Fin answers only from content sources you explicitly connect, such as articles, snippets, and external sources. Anything absent from those sources is treated as unknown rather than guessed.

Ramp period

The interval after launch during which Fin's deflection and resolution metrics are still stabilizing as it encounters real conversation volume. Early numbers should not be treated as final performance.

AI-first helpdesk

A support model where Fin is the default first responder and human teammates handle escalations. Routing, escalation triggers, and knowledge content are designed around that division of labor.

Snippets

Short reusable answers Fin can draw on, best suited to compact policy facts such as shipping windows, pricing, or return rules that do not justify a full help center article.

Workflows versus Fin

Workflows execute deterministic rules like tagging, assignment, and timed actions. Fin handles open-ended natural language questions. Choosing the wrong tool produces brittle automation or unpredictable answers.

Escalation triggers

Explicitly defined topics or conditions that must always route to a human, such as billing disputes, legal requests, or churn risk. They are encoded in Fin instructions or workflow rules.

Content hygiene rule

Duplicate or contradictory articles degrade Fin's answer accuracy, so overlapping sources must be consolidated and conflicts resolved before a source is connected or republished.

First response time

The elapsed time between a customer's first message and the first human reply. On AI-first helpdesks it is measured separately from Fin's instant initial response.

Backlog by inbox

Unresolved conversation counts grouped by inbox rather than aggregated overall. This view reveals which team or queue is actually understaffed and where escalations are accumulating.

Quality sampling cadence

The scheduled rate at which closed conversations are reviewed against a rubric. Fin-handled conversations should be sampled on the same cadence as human-handled ones.

Staffing forecast rule

Forecast conversation volume by channel and day of week, then map it to coverage. Fin deflection lowers but does not eliminate peak load, so coverage planning still applies.

Escalation ownership

A manager decision defining which topics Fin must always hand off and which team inbox receives each type of escalation. Agents follow it rather than choosing ad hoc.

Notes versus replies

Notes are internal and visible only to teammates, while replies are delivered to the customer. Confusing the two leaks internal commentary or hides information from the customer.

Snooze versus close

Snoozing sets a time for the conversation to reopen, keeping it pending. Closing ends the conversation and, where enabled, can trigger a customer satisfaction survey.

Handoff transcript

The conversation history passed to a human when Fin escalates. Agents should read it before replying so customers are not asked to repeat information they already provided.

Fixing wrong Fin answers

When Fin answers incorrectly, the correction is made in the underlying article or snippet content, not by retraining a model or editing the conversation transcript.

Optimization loop

Review Fin's unanswered and low-confidence questions, add or correct the relevant content, then re-test the same questions to confirm the gap is actually closed.

Content quality over volume

Short, single-topic articles with clear titles resolve better than long pages covering many topics, because Fin can match a specific question to a specific source.

Regression question list

A maintained set of real customer questions that previously failed. Re-testing it after each content change confirms fixes hold and no earlier answer has broken.

Baseline measurement

Recording deflection and resolution figures before a content edit, so any later change can be attributed to the edit rather than to seasonality or volume shifts.

Snippet promotion rule

Use a snippet as the fastest fix for a single recurring question, then promote it to a full article once the question recurs at meaningful volume.

Multi-question email threads

Email messages often contain several questions at once. Fin should answer what it can from content and escalate the remainder rather than ignoring unanswered parts.

Email threading

Keeping Fin's reply inside the existing conversation thread so full history and context are preserved. Breaking the thread loses prior messages and confuses the customer.

AI disclosure on email

Stating in the email auto-reply that Fin is an AI assistant, along with how to reach a human. This sets expectations before the customer reads the answer.

Email sender configuration

Domain authentication and a valid reply-to address must be configured before enabling Fin on email, or replies may fail deliverability checks and land in spam.

Fin Procedure anatomy

A procedure defines its trigger and scope, required inputs, ordered steps, decision branches, and a clear completion or handoff outcome. Each step should reference specific content.

Procedure branching fallback

Every decision point inside a procedure needs an explicit path to a human when its condition is not satisfied. Without a fallback, Fin improvises at the branch.

Procedure scope rule

Build one narrow procedure per task, such as a refund eligibility check, rather than a single large procedure covering all billing topics. Narrow scope makes testing and review feasible.

Sales qualification fields

The predefined attributes Fin collects conversationally before handing a lead to a rep, typically intent, company size, and use case. They are defined before the conversation flow is built.

Sales handoff triggers

Topics Fin must never decide, including discounts, contract terms, and delivery dates. These route immediately to a sales rep rather than being answered from content.