What it does and does not do

Scope matters. This page is deliberately honest about the boundaries, so you know what you are installing before you install it.

What it does

• Recommends articles on the issue. Up to three knowledge base articles per request, scored Strong / Medium / Weak, each showing the keywords behind the match.
• Shows matching articles to customers. Optionally, behind an admin toggle, the same ranked matches appear on the customer's request in the portal, read-only, so a customer can help themselves while they wait.
• Measures coverage by request type. A point-in-time read of how many recent requests already had a strong matching article, weakest request types first, with the sample size always shown.
• Respects who is allowed to see what. For agents it searches as the viewing user, so they only ever see content they already have permission to read. For customers it searches as the app, but only ever returns articles published to the portal for that service desk, so a customer never sees anything restricted or internal.
• Installs on Jira alone. No Confluence install, no second product, no space picker, no setup beyond turning it on.
• Stays inside your tenant. Built on Atlassian Forge. No external server, no external AI model, nothing sent anywhere.

What it does not do

• It does not write to your Jira or your knowledge base. It is read-only. The only exception is a small per-project flag it sets so each panel knows it has been switched on.
• It does not resolve, close, or reply on anyone's behalf. On the portal it surfaces read-only suggestions to the customer; it never transitions or closes a ticket, and it never auto-replies. It surfaces and ranks, and a human always decides what to do next.
• It does not search internal-only knowledge. It reads the customer-facing knowledge base. Internal runbooks that were never published to the portal are out of scope (and there is a good technical reason for that, covered in How it works).
• It does not use AI. The scoring is rules-based and auditable. That is a deliberate choice, not a limitation we are apologising for.

The promise it refuses to make

Most tools in this space will quote you a deflection rate. Second Chance will not, because deflection cannot be honestly measured. The person who found your article and solved their own problem never raised a request, so they are not in your data. Nobody can count them.

What can be measured is the opposite and honest version: of the requests that did reach you, how many already had an article that would have answered them. That is coverage, not deflection. A request that arrives with no strong match is one your knowledge base was not able to cover.

Showing matches to customers on the portal does not change this. It may well help someone solve their own problem, but the app still will not put a number on how often that happens, because that number cannot be honestly counted.

We always show you the coverage number with its sample size next to it, so a reading from a handful of recent tickets can never be mistaken for a full audit. That honesty is the whole credibility of the app, and we guard it.

Where to next

How it works: the scoring, in plain language.