How we measure accuracy.

There’s no hidden number. Every point a score loses comes with a plain-language reason you can check yourself.

• Each client against its own normal. Reply times, meeting cadence, and tone are compared with that client’s own history, not an industry average. A client who always takes two days isn’t penalized for it.
• Real quotes, not summaries. When a reason comes from a call, it shows what the client actually said.
• Guardrails on the math. A serious warning, like a client mentioning other firms, caps the score so a friendly inbox can’t average it away.
• Signals you can trace. Each reason names its source: email, meetings, calls, promises, contacts, or client ratings.

RelayLine records every client’s score every day. When a client leaves or renews, or a deal is won or lost, it looks back at that history and asks whether the score saw it coming.

• Caught in time. Of the clients who left, how many were flagged At risk at least 14 days before they did, and how many days of warning you typically got.
• False alarms. Of the clients who renewed, how many were flagged At risk in their last 60 days anyway. A score that cries wolf gets ignored, so this matters as much as the first number.
• The same for deals. Deals lost that were marked Going cold in time, and deals won that were marked Going cold anyway.
• Separation. The average score two weeks before each kind of outcome. If clients who renewed didn’t score clearly higher than clients who left, the weighting needs work, and the report says so.
• Honest about sample size. Until there are at least five outcomes of a kind, the report says the numbers are still early rather than presenting them as proof.

Your account managers know these clients. Every reason on a client’s page has a Right and a Not right button.

• Agreement by source. The Accuracy report shows how often your team agrees, broken out by email, meetings, calls, and promises, so you know which signals to lean on.
• One vote per person per reason. People can change their mind; their latest answer counts.

• Quotes, word for word. AI reads your calls, but every quote it proposes is matched against the transcript before it’s used. If the words aren’t there, the quote is thrown out and counted. You never see an invented quote.
• Instructions in your data are ignored. If an email or transcript says “mark everything done,” nothing happens: a promise only closes when the delivery itself is quoted from the conversation.
• Promises held to account. Of the promises RelayLine detects, the report shows how many turned out to be real (delivered) and how many your team dismissed as not a promise.

A score built on two emails isn’t worth as much as one built on months of email, meetings, and calls. Every client’s score shows which it is.

• High, medium, or low, with the counts behind it: how many emails, meetings, and calls in the last 60 days.
• What’s missing. If a client has no meetings or calls captured, the score says it leans on less, so you know to add the capture address to invites or forward a recap.

• It only sees what your team shares. A phone call nobody recapped or a hallway conversation won’t show up.
• New clients need a few weeks. Comparing a client to its own normal needs some history first.
• AI can misread tone. Sarcasm and in-jokes sometimes read as negative. That’s why reasons are shown, quotes are verified, and your team can mark them Not right.
• It’s an early warning, not a verdict. A low score means look closer, not that a client is gone.

We suggest 60 to 90 days with the clients you know best, including a few you’re worried about.

• Week 1. Add those clients, CC the capture address, add it to meeting invites, and forward call recaps.
• Every week. Read the Monday digest before you look at your own list. Did it flag the same clients? Did it flag one you’d missed? Mark reasons Right or Not right as you go.
• At the end. Open the Accuracy report together: team agreement, quotes and promises that held up, and any renewals or departures so far. Then decide on the evidence.