How matches are scored

The Overview explains the idea in plain language. This page gives you the exact formula, so the Strong / Medium / Weak tag on a suggestion is never a mystery. There is no AI and no hidden model: every score is a sum of fixed points you can add up yourself.

The signals

When you open a request, Second Chance reads its summary, description, request type, and labels, pulls out the meaningful keywords (dropping filler like "the" and "please"), and searches your knowledge base. Each candidate article then earns points from five signals:

Signal	Points	Earned when
Request type	up to 30	The article title contains the request type's name.
Summary keywords	up to 25	The request's summary keywords appear in the article title or body.
Description keywords	up to 15	The request's description keywords appear in the article.
Label overlap	15	A label on the request is also a Confluence label on the article.
Freshness	+3 / −3	+3 if the article was updated in the last ~6 months; −3 if it is older than ~18 months; nothing in between.

The request type is weighted highest on purpose: if an article is written for the kind of request you are looking at, that is the strongest signal there is.

How the keyword points are worked out

Summary and description points are proportional, not all-or-nothing. The points scale with the share of the request's keywords that the article actually contains:

points = (matching keywords ÷ keywords the request had) × the signal's weight

The divisor is the number of keywords the request genuinely carried, capped at 10. So a short, on-topic request that matches every keyword scores the full weight, rather than being penalised for being brief.

Title boost. If any matched keyword appears in the article's title (not just its body), the summary and description points are multiplied by 1.5. A title match is a strong indicator the article is genuinely about the request.

Adding it up

The final score is simply:

request type + summary + description + label + freshness

That total maps to the tag you see:

Score	Tag	Meaning
35 and above	Strong	A confident match.
22 – 34	Medium	A reasonable match worth a look.
15 – 21	Weak	A loose match, shown only if nothing better exists.
Below 15	not shown	Filtered out as noise.

15 is the default minimum match score, and you can raise it per project on the Enable a project screen if you want to hide weaker matches.

A worked example

A request of type "VPN access" whose summary yields four keywords. An article titled "Set up VPN access to the office" contains the request type's name and two of the four summary keywords, in its title, and was updated last month:

• Request type in the title → +30
• Summary: 2 of 4 keywords → (2 ÷ 4) × 25 = 12.5, and because they are in the title, × 1.5 → +18.75
• Updated recently → +3
• Total ≈ 51.75 → Strong

Which articles you see, and in what order

Only articles scoring at or above the minimum are shown, and never more than the top three. When scores tie, Second Chance prefers the article with more title matches, then the more recently updated one, then alphabetical order: a stable, predictable ranking.

The percentage a customer sees is not this score

If you choose the Percentage confidence cue on the portal (see Show suggestions to customers), the "n% match" a customer reads is not the raw score above. The raw score realistically tops out in the 50s, so a genuinely strong match in the mid-30s would read as a misleadingly low "35%".

Instead, the percentage is a calibrated 0–99 scale anchored to the very same bands:

• Weak band → 50–64%
• Medium band → 65–84%
• Strong band → 85–99%

So the number a customer sees always agrees with the qualitative band, and a strong article reads as high confidence. The admin panel states this plainly, so the figures never look contradictory.

Where to next

You now know exactly how a suggestion earns its place. Next, see how the people on both sides of the desk tell you whether it actually helped: Feedback from agents and customers.