1. What the team does (the function)
Concierge is not a switchboard and not a sales desk. Its work is relationship and case continuity, and it is genuinely two-way: the team places more outbound calls (1,932) than it answers inbound (1,233), proactively reaching clients as well as fielding them. Calls are long (8.5 minutes average) because they are substantive case conversations, not quick lookups.
Three core jobs (what clients call about)
| Call type | Share | What the staff member is doing |
|---|---|---|
| General inbound client | 78% | Status, questions, reassurance, capturing life and medical updates |
| Client callback / status check | 8% | Returning a client who was asked to call, or checking where the claim stands |
| SSA / agency | 5% | Social Security or an agency calling about a client |
| New client intake | 4% | A prospect or referral that has to be handed to Intake |
| Medical provider / records | 2% | Provider or records inquiry |
How the calls end (disposition mix)
| Outcome | Share | Meaning |
|---|---|---|
| Resolved with info on the call | 60% | Client got a real answer |
| Callback scheduled | 14% | Staff must call SSA or follow up and call back |
| Message taken | 11% | Passed to another person or team |
| Partial / transferred | 6% | Handed off mid-call |
| Not resolved | 6% | Client left without an answer |
| No contact (IVR / voicemail) | 3% | Never connected |
Roughly 4 in 10 calls end in an open loop (a scheduled callback, a message to pass, a transfer, or unresolved). Every one is a promise that has to be worked to closure, and today nothing systematically guarantees the loop gets closed. That is the single largest operational risk on this team.
2. How the team performs today
The open is the weakest moment, and it is mostly structural
| Signal (measured from the call structure) | Result |
|---|---|
| Clients who came through a reception transfer before reaching Concierge | 42% |
| Clients routed directly to the Concierge queue (one hold, no reception) | 58% |
| Hold prompts before a human answered (median, up to 9) | 2 |
| Staff who re-asked the name reception had just collected | ~12% |
The 42% who arrive via reception get the worse experience: a second hold, then a staff member who re-asks their name and reason because reception's information does not travel (no screen-pop). To the client, that reads as "they do not know who I am," in the first ten seconds. Add that 43% of calls are repeat callers who often ask for the specific person they spoke to before and are told "we do not have case managers anymore," and the open frequently starts from a confidence deficit.
Method note: self-identification and status quality vary widely staff to staff, but automated transcript keyword counts are unreliable for those because the transcription mis-hears "Newlin" and redacts detail. Those are characterized from direct call review (not keyword counts); the structural numbers above are from the CDR and are measured.
3. The opportunities
1. Win the first 30 seconds
4 in 10 clients get a double hold and a re-ask. The open is where confidence is made or lost, and it is currently the team's weakest point. Fixing it is the highest-visibility improvement available.
2. Restore continuity
43% of calls are repeat callers. The pooled-team model removed the "my case manager" relationship without giving staff a way to rebuild trust on each call. Clients notice, and some threaten to leave over it.
3. Close every loop
Roughly 40% of calls create a follow-up obligation. Clients report promised callbacks that never come. Un-closed loops are the most direct driver of frustration and churn on this team.
4. Make quality consistent
No standard exists, so the same call is excellent or poor depending on who answers. A simple, shared definition of a good call is the fastest lever on overall experience.
4. Solutions
Start now No new technology required
| Move | What it fixes |
|---|---|
| A "Confident Open" standard (15 seconds): name plus firm every time, one smooth verify, lead with a specific status. Never "how can I help you?" into dead air. | Opportunity 1, 4 |
| One reframe script for the pooled-team model: "You are no longer waiting on one person; our whole team can see your case and help you the moment you call." | Opportunity 2 |
| Close-the-loop discipline: every promised callback becomes a tracked task with an owner and a due time; the QA team verifies closure. | Opportunity 3 |
| The seven-point call standard (Section 5) as the shared definition of a good call, taught before it is scored. | Opportunity 4 |
Soon (Twilio) Near, not here: the structural fixes
| Move | What it fixes |
|---|---|
| Screen-pop on answer: the staff member sees who is calling and where the claim stands before saying hello, and opens with "Hi Mr. Diaz, I have your file up, I can see your hearing is scheduled..." | Opportunity 1, 2 |
| Route the 42% straight to Concierge: IVR plus client lookup sends known clients directly to the queue, killing the second hold and the re-ask. | Opportunity 1 |
| Skill-based routing and follow-the-owner: calls and SMS reach the right sub-team and the current case owner without manual transfers. | Opportunity 2 |
The now-moves and the Twilio-moves are complementary. Standing up the standard and the QA program now means that when routing and screen-pop land, the team is already trained to use the better tools well.
5. Starting the QA journey
The firm has light QA today. The goal is real standards with a human-in-the-loop QA team that takes action. The path is crawl, walk, run, and it starts with a shared definition of a good call.
The v0 rubric: the Concierge call arc (seven fundamentals)
| # | Fundamental | The standard | QA check (yes / no) |
|---|---|---|---|
| 1 | Open | Name plus firm plus offer to help | Did they identify themselves and the firm? |
| 2 | Verify before disclosing | Confirm identity before any case detail | Was identity verified before case info was shared? |
| 3 | Confirm the ask | Reflect the reason for the call back in one sentence | Did they confirm what the client needed? |
| 4 | Own it plus set expectations | Plain status; if unresolved, say what, who, and when | Did they give a concrete next step with a timeframe? |
| 5 | Capture plus correct | Log updates; read back; fix bad data on the spot | Were updates captured and confirmed? |
| 6 | Close the loop | Recap next step, confirm channel, warm close | Did they recap what happens next? |
| 7 | Rapport (scored) | Recognition plus warmth: quickly identify the client and get oriented on the case, warmly, easing friction, never robotic or hold-heavy | Scored low / medium / high (not yes / no) |
- Low: robotic, cold, hold-heavy, makes the client repeat themselves, no acknowledgment.
- Medium: polite and competent, but flat.
- High: warm, quickly oriented on the file, acknowledges the client's situation, eases the friction; the client audibly relaxes.
This is the dimension keyword scoring cannot judge, because it is tone and flow, not words. It is scored by the human reviewers and an LLM judge that reads the whole call. It is the one that answers the real question: would the client feel taken care of?
If you install only three: #2 (verify before disclosing), #4 (own it, with a concrete next step and timeframe), and #7 (Rapport) most separate a good call from a bad one.
Crawl, walk, run
- Crawl (orient): teach the seven fundamentals with anonymized "good call" clips. No scoring yet. The team learns what good sounds like before anyone is measured.
- Walk (score): the AI QA engine scores every call against the six yes/no checks plus the Rapport score and flags exceptions. Humans review a sample plus all flagged calls. Scores are for coaching first.
- Run (accountability): published per-person scorecards, a coaching cadence, and the fundamentals folded into performance expectations. Accountability arrives only after the team has been oriented and coached.
The human-in-the-loop QA operating model
AI scores everything; humans act on what matters. The transcript QA engine already runs nightly on this team, so it can score all calls against the seven fundamentals and surface the exceptions. The human QA team then:
- Reviews a daily sample plus every flagged call (failed verify, no next-step, open loop, unhappy client).
- Coaches the staff member with the specific clip and the specific fundamental, not a vague score.
- Verifies closure of the ~40% open-loop calls: was the promised callback actually made?
- Escalates systemic issues (the pooled-team friction, the callback follow-through gap) rather than blaming staff for structural problems.
This is the bridge from the light QA of today to a program that changes behavior: the machine handles coverage (every call, every day), and the people handle judgment and action.
Flag both ways: correct and recognize
The engine does not only catch weak calls. It flags both directions, and both feed the journey:
- Exceptions (weak calls: failed verify, no next step, low rapport, an open loop, an unhappy client) route to coaching and correction.
- Exemplars (standout calls: high rapport, clean execution, a hard situation handled well) route to recognition, and become the anonymized "good call" clips that train the rest of the team.
Catching staff doing it right, not only wrong, is what keeps the accountability journey motivating rather than punitive, which matters most for a team moving from no measurement. It also sets up a light recognition and gamification layer once the basics are established: call of the week, per-fundamental streaks, and team scoreboards built on the same seven-point standard.
6. The immediate next moves
- Finalize the seven-point standard and a one-page "Confident Open" script for orientation.
- Turn on AI scoring against the standard for Concierge and pull a two-week baseline before coaching.
- Stand up the open-loop tracker so every promised callback has an owner and a due time.
- Run the same read on one more team (Case Status) to confirm the fundamentals hold firm-wide before publishing.