Score your Front team's replies against your own QA rubric

Build me a conversation quality review desk so our weekly support QA ritual stops living in a spreadsheet. The people who open it are support team leads working through their weekly review round, plus the head of support who wants evidence before a one to one. It reads from Front and logs to Google Sheets. One hard rule that applies everywhere in the app: reviewers only ever see closed conversations, so nobody is graded on work still in flight. Front calls closed conversations archived, so treat archived as closed and never surface anything that is open, spam, or deleted.

The first screen is the review queue. Pull the teammate roster once with List Teammates and use it both for the per teammate filter and for attributing each conversation to the person who handled it. Build the sample for a review period, defaulting to the last seven days, by taking a fixed number of closed conversations per teammate rather than whatever closed most recently. The sampling has to be fair: if the number is five, every teammate contributes five, spread across the period and drawn at random from what is eligible, not the five newest. Get each teammate's candidates with List Assigned Conversations, and use Search Conversations with Front's query filter syntax when the reviewer narrows by date range, inbox, or keyword. Also let a reviewer build a queue from a tag using List Tagged Conversations, for the weeks we want to review everything marked as a complaint or a refund. There is no way to list tags through the Front API, so the tag ids for that live in app settings where an admin pastes them in.

Each queue row shows the teammate, the customer, when the conversation closed, the inbox, the subject or a short preview, and its review state, which is one of not reviewed, drafted, or done. Above the queue put a coverage strip listing every teammate with how many finished reviews they have in the current period, so a lead can see at a glance who has not been reviewed yet, and teammates sitting at zero should stand out. Let a reviewer re roll the sample for a single teammate when a picked conversation turns out to be a duplicate, a test, or otherwise not worth reviewing.

Clicking a row opens the review pane. The left side is the whole thread inline and in order, pulled with List Conversation Messages, showing who wrote each message and when, with the customer's messages and the teammate's replies clearly distinguished so a reviewer can read the exchange without opening Front. The right side is the scoring panel holding our rubric, which starts with tone, accuracy, ownership, and whether the customer's actual question got answered. Every rubric line has a score and a comment box, and underneath sits an overall written feedback field. The rubric lives in app settings so a lead can add, rename, reweight, or remove lines and change the scale without breaking reviews that already exist. Scores, comments, the reviewer's identity and the timestamp all persist in the app, keyed to the conversation and to the teammate who handled it.

Add a Score this thread button that runs a background agent over the open conversation. The agent reads the full message history, drafts a score for every rubric line, and for each line writes the exact quote from the thread that justifies that score, so the reviewer can check the evidence without rereading everything. It also flags anything that looks like a missed escalation, meaning the customer asked for a manager, threatened to leave, raised a legal or billing dispute, or repeated the same unanswered question, and it names the message that made it think so. Its output lands back in the scoring panel as suggestions that pre fill the form, so the reviewer is editing rather than starting from a blank one. Show progress while it runs, and let it be re run on any thread whose review is not finished yet.

Suggested and accepted have to be unmistakably different, because if people cannot tell them apart they will not trust the scorecard. Drafted scores are visually marked as the assistant's suggestion in both the queue and the scoring panel, and they never count toward a teammate's average or toward the coverage strip. A review only counts once a reviewer finishes it. On finishing, record for every rubric line whether the reviewer accepted the suggestion as is or overrode it, and keep both the suggested number and the final number, so we can look back later and see how often the assistant agreed with the humans. A reviewer should also be able to clear all suggestions and score the thread from scratch.

Finishing a review appends a row to a running scorecard in Google Sheets with Append Values: the date, the review period, the teammate, the reviewer, the conversation id and a link to it, one column per rubric line carrying the final score, whether that line was overridden, the missed escalation flag, and the written feedback. The spreadsheet id and tab name live in app settings. Every row carries the review id, so when a reviewer edits a review that was already finished, restate that row in place with Update Values instead of appending a second one and double counting the teammate.

A trends view shows each teammate's average by rubric line over the last few months, built only from finished reviews stored in the app so it stays correct even if somebody edits the spreadsheet by hand. I want a teammate's line by line averages by month next to the team average for the same month, the direction of travel, and the number of reviews behind each figure so one review never looks like a trend. Show the reviews carrying a missed escalation flag as their own list. This view is what a lead brings into a coaching conversation, so keep the evidence one click away: clicking any number opens the reviews behind it, and from there the thread and the quotes.

When a review is finished, let the reviewer push the coaching note back onto the conversation as a private note in Front using Add Comment, so the feedback sits with the thread instead of only in our app. Pre fill the note with the final scores and the written feedback, let the reviewer edit it before it posts, keep it optional per review, and record on the review whether it was posted. Make clear in the interface that this is an internal note visible to the team and never to the customer.

Keep the whole thing usable by a non technical team lead, and keep the Front calls modest. Front rate limits per company, from fifty to two hundred requests a minute depending on plan, and search costs more quota than other calls, so sample a handful of conversations per teammate, fetch message history only for the conversation actually being opened or scored, page with the cursor Front returns rather than guessing, and back off when Front asks you to.