Recruiter Call Coaching

The useful output of recruiter call coaching is a coaching note with two kinds of cites: the transcript turn, and the rubric line it maps to. That is the quality outcome.

A note that says "needs more rapport" is not usable. A note that says "Rubric: Confirm next step and owner before close. At 12:40 the recruiter said I'll have someone follow up. The family never named a date, a portal task, or a person" is usable. The manager can replay that stretch and practice a closer that matches the local standard.

Do not invent a rubric score to make the record look complete. If your shop wants a one-to-four on next-step clarity, a human still assigns it after reading the cites. Model-generated totals invite arguments about the number instead of the call. Empty stays empty when the audio cannot support a cite. A blank note is an honest result. A guessed score is not.

The manager still owns coaching. The note is a briefing, not a verdict, not a personnel write-up, and not a substitute for sitting with the recruiter.

Start with audio you would be willing to quote in a 1:1. Capture the recording under your existing consent and retention rules. Transcribe it with speaker labels and timestamps when you can tell recruiter from family, counselor, or student. If two speakers collapse into one blob, treat that as a transcription defect, not a personality finding.

Then apply a hard gate: is this transcript good enough to cite? Unusable audio includes dropped mobile calls, gym crosstalk, a headset that only captured the recruiter, a language the model cannot reliably transcribe, or a file that cuts out before the close. In those cases, stop. Do not score.

Scoring from a bad transcript is the first failure mode. You will coach the recruiter for interrupting when the model merged two voices, or for skipping financial aid when the aid question was never captured. The recruiter learns to distrust the process. The manager coaches a ghost.

If the gate passes, score only against the local rubric: the same checklist your enrollment shop already uses in ride-alongs. Score the lines you already train, such as restating the student's concern, naming the next deadline and owner, and refusing aid promises the file does not support. The model checks whether those lines show up in the talk. It does not add a line because another campus trains recruiters to use a first name three times.

Inventing a best-practice the rubric does not contain is the second failure mode. If your rubric is silent on storytelling, the note must not ding a recruiter for a weak narrative arc. If your rubric does not score talk-to-listen ratio, do not output one. Extra criteria quietly rewrite the job.

Where the rubric is binary, say observed or not observed, with a cite. Where it is qualitative, quote the attempt and name which part of the line it met or missed. Do not average those into a single call-quality index. Averages hide the one miss the manager meant to practice this week.

Keep CRM objects out of the scoring loop. Salesforce, Slate, and Technolutions systems can store that a call happened, who owned it, and whether a task was created afterward. They are not the rubric. A missing CRM task is a process finding for the manager to check after the conversation, not a model-invented coaching item, unless that check is already on the printed rubric.

After mapping, each observation should point at a turn: a timestamp plus a short quote, or a speaker-labeled line if that is how you store transcripts. The cite has to be short enough to read in a huddle. Dumping the whole transcript under evidence is not a cite.

Here is one illustrative pass, not a measured result. A regional recruiter calls a first-generation family about a spring overnight. The local rubric includes four lines: restate the student's academic interest in one sentence; explain overnight logistics without assuming the family has a car; name the financial-aid next step and who owns it; close with a date the family can repeat.

The transcript shows the recruiter restating "nursing, maybe" as "our nursing pathway," which maps to the first line as observed. At 08:12 the parent asks how they get to campus from the bus station. The recruiter answers with parking lots. Line two is not observed; cite 08:12 to 08:35. At 14:02 the recruiter says aid is generous and the family will be fine, with no FAFSA, CSS Profile, or counselor name. Line three is not observed; cite 14:02. The close is "I'll shoot you an email." Line four is not observed.

The coaching note lists those four mappings and stops. It does not assign two out of four. In the 1:1 the manager replays 08:12 and 14:02, then they practice a bus-to-campus answer and a single aid next step the family can write down. That is the loop: transcribe, score against the local rubric, cite the turn, and the manager coaches.

If a later stretch of the same call is unintelligible, leave those rubric lines blank rather than inferring them from the clear opening. Partial notes are allowed. Filled-in guesses are not.

Using the note as a PIP is the third failure mode. A coaching note with cites is material for development. It is not documentation of misconduct and not a substitute for your HR process. If a call contains a policy problem, such as a promise the institution cannot keep, the manager escalates through the real channel. Do not let a weekly batch of model notes become an unofficial dossier.

The recruiter should see the same cites the manager sees, or the exercise becomes surveillance with extra steps. The 1:1 is where judgment lives. Maybe the family was in a parking lot and the email close was the only humane option. The note cannot know that. Empty notes stay empty. Contested transcripts get a human relisten, not a harsher invented score.

Write the 1:1 prompt as a question. "At 14:02 you told them they would be fine on aid. What did you want them to do on Monday?" is a coaching move. The model can draft that question from the cite. The manager still decides whether to ask it.

Call coaching is about the conversation that already happened. It is not a forecast of whether the student will deposit, and it should not be blended into predictive yield scoring. A warm call that misses the aid next step is still a coaching item.

The follow-up message after the call is a different job. If you generate that message with a model, keep it on personalized outreach generation, using facts the recruiter actually said, not the coaching critique. Do not auto-send an apology the manager has not approved.

File evaluation is also a different job. Recruiter talk tracks should not leak into reviewer decision assistance as shadow evidence about the applicant. Later counseling conversations belong with a placement counselor co-pilot, not with this recruiting rubric.

Put the coaching note where coaches already look: a shared doc, a private 1:1 agenda, or a quality queue. If a CRM record needs a breadcrumb, store that a note was prepared, or a link your policy allows, not the model's prose in a field counselors or the student might one day see. Salesforce, Slate, and Technolutions records are durable. Coaching language is not.