1. Customer response time: Customer message to first reply measured Replies
| Measure | Last 14 days | Month to date | Right now |
|---|---|---|---|
| Machine alone: median | 33.2 min | 2.6 min | · |
| Machine alone: slowest 10% (p90) | 272 min | 73 min | · |
| Machine alone: replies counted (n) | 9 | 15 | · |
| Machine draft, a person approved: median | no data | no data | · |
| Machine draft, a person approved: slowest 10% (p90) | no data | no data | · |
| Machine draft, a person approved: replies counted (n) | 0 | 0 | · |
| A person typed it: median | 9 min | 12.9 min | · |
| A person typed it: slowest 10% (p90) | 122 min | 145.3 min | · |
| A person typed it: replies counted (n) | 49 | 78 | · |
| Machine alone, message arrived while store open: median | 17.7 min | 2.3 min | · |
| Person, message arrived while store open: median | 13.8 min | 12.9 min | · |
| Everyone: answered within 15 minutes | 53.4% | 54.8% | · |
| Everyone: answered within 60 minutes | 79.3% | 81.7% | · |
| Answered after more than 24 hours (left out of the medians) | 3 | 6 | · |
| Customers whose latest message has no reply yet | · | · | 4 |
How this is counted, and where it comes from
Definition. For each customer, minutes from their latest message to the first reply on the thread (the Reply check's lastInboundAt to answeredAt, the same pairs the Home strip's Median uses). Who answered: machine alone = the auto-responder or System; machine draft approved = the approvals sender (a person pressed approve); person = anyone else, a rep or manager typing (scripts/lib/machine-reply.ts rule: a reply from the shared login that lines up within 2 minutes with a machine send is the machine). Windowed by the day it was answered. Replies later than 24 hours are counted separately, not in the medians (a reply days later is not an answer to that message). Clock minutes, nights and Sundays included; the 'store open' rows keep only messages that arrived inside sales hours (config/store-hours.json).
Read with care.
- One pair per customer: if a customer wrote twice, only the newer message is on file, so this is a sample, not every message. Pairs are saved in the daily file as they are seen, so month to date keeps them after a customer writes again.
- The Reply check reads VinSolutions about every 2 hours; replies typed by a rep inside VinSolutions are in it, texts sent outside VinSolutions are not.
- Medians over fewer than 10 are noise; n is shown. A p90 of a small n is just its slowest few.
Source files.
memory/reply-verify-latest.jsonas of Nov 19, 1:23 PM CTmemory/message-log.jsonas of Nov 19, 9:41 PM CTdata/store/metrics-daily.jsonas of 6:50 AM CT
2. Overdue and unanswered: Unanswered messages and overdue follow-ups measured Turnover
| Measure | Last 14 days | Month to date | Right now |
|---|---|---|---|
| Customers who wrote us | 83 | 117 | · |
| Of those, latest message still has no reply | 4 | 4 | · |
| Average per day: customers waiting with no reply | 0.3 | 0.2 | · |
| Overdue: arrived in store hours, no reply within 60 min (proposed standard) | 15 | 20 | · |
| Held replies that expired with no person answering | 10 | 13 | · |
| Promises and callbacks due, still not done | 43 | 57 | · |
| No reply yet, all dates | · | · | 4 |
| Promises and callbacks overdue now | · | · | 69 |
By day (last 14 days)
| Date | Customers who wrote | Latest message, no reply yet | Overdue (store hours, 60+ min) | Held replies expired | Promises due, not done |
|---|---|---|---|---|---|
| 2026-11-07 | 4 | 0 | 0 | 0 | 0 |
| 2026-11-08 | 1 | 0 | 0 | 1 | 0 |
| 2026-11-09 | 18 | 0 | 2 | 0 | 2 |
| 2026-11-10 | 9 | 0 | 1 | 0 | 4 |
| 2026-11-11 | 5 | 0 | 0 | 1 | 4 |
| 2026-11-12 | 11 | 0 | 3 | 2 | 4 |
| 2026-11-13 | 6 | 0 | 1 | 1 | 1 |
| 2026-11-14 | 15 | 0 | 1 | 2 | 2 |
| 2026-11-15 | 0 | 0 | 0 | 0 | 2 |
| 2026-11-16 | 13 | 0 | 1 | 0 | 5 |
| 2026-11-17 | 10 | 0 | 2 | 1 | 6 |
| 2026-11-18 | 9 | 2 | 2 | 0 | 6 |
| 2026-11-19 | 10 | 2 | 2 | 2 | 5 |
| 2026-11-20 | 2 | 0 | 0 | 0 | 2 |
How this is counted, and where it comes from
Definition. No reply yet = a customer whose latest message the Reply check shows nobody (person or machine) answered, and we have not texted since (the Home strip's 'No reply'). Wrote us = distinct customers with a message in the window (reply ledger + Reply check). Expired holds = customer replies the machine held for a person that ran out of time with no person answering (/turnover 'Expired, still unanswered'), counted on the day they expired. Overdue (PROPOSED standard, 60 minutes) = a message that arrived while the store was open and was answered later than that, or is still unanswered. Promises due and not done = commitments, video promises and missed-call callbacks whose due time (/promises rule: 4 hours, or noon next business day) fell in the window and are not marked fulfilled.
Read with care.
- No reply yet is a snapshot of who is waiting now, grouped by the day they wrote; a customer who wrote twice is counted once, on the newer day.
- /turnover only looks back 14 days, so Expired holds for a long month come from the saved daily file (starts when this job first ran).
- Missed-call callbacks are most of the promises count and are never closed by the customer, so that number only falls when a person marks the call.
Source files.
memory/reply-verify-latest.jsonas of Nov 19, 1:23 PM CTmemory/reply-ledger.jsonlas of 6:33 AM CTmemory/message-log.jsonas of Nov 19, 9:41 PM CTmemory/pending-approvals.jsonas of 12:59 AM CTmemory/draft-class-ledger.jsonlas of Nov 19, 1:43 PM CTmemory/commitments.jsonas of Nov 19, 5:15 PM CTmemory/video-promises.jsonas of Nov 19, 7:00 PM CTmemory/missed-call-callbacks.jsonas of Nov 19, 7:25 PM CT
3. Re-engaged opportunities: Customers who wrote back after 14 or more quiet days measured Replies
| Measure | Last 14 days | Month to date | |
|---|---|---|---|
| Customers re-engaged | 46 | 64 | · |
| after our outreach (we sent in the 72 hours before) | 27 | 38 | · |
| on their own | 19 | 26 | · |
| first reply ever on an old lead | 20 | 28 | · |
| still open (not closed lost, sold or dead) | 35 | 47 | · |
| Left out: a new lead (wrote within 2 days of the lead being created) | 119 | 181 | · |
| Left out: nothing to measure the quiet from | 0 | 1 | · |
How this is counted, and where it comes from
Definition. A customer counts when they write us and the gap since their previous message from them (any text or email we know of: the reply ledger since 8/19, the Reply check, and the last 20 notes on the lead) is 14 days or more. With no earlier message at all, the gap is measured from the day the lead was created ('first reply ever'). A message within 2 days of the lead being created is the new lead itself and is left out. Counted once per customer per window. 'After our outreach' = we sent them something in the 72 hours before they wrote; the rest wrote on their own. 'Still open' leaves out customers the Reply check has since closed as lost, sold or dead.
Read with care.
- Quiet means the customer was silent; we may have been texting them the whole time.
- A customer with no message on file and no lead row cannot be measured and is skipped (count shown).
- Out-of-window history is only as deep as the lead's last 20 notes plus the ledger from 8/19, so a quiet stretch can be overstated for a customer whose older messages fell off.
Source files.
memory/reply-ledger.jsonlas of 6:33 AM CTmemory/reply-verify-latest.jsonas of Nov 19, 1:23 PM CTdata/store/leads.json (Notes Summary, createdTime)as of 6:48 AM CTmemory/message-log.jsonas of Nov 19, 9:41 PM CT
4. Appointments: Appointments set and completed measured Appointments
| Measure | Last 14 days | Month to date | |
|---|---|---|---|
| Appointments set (dated in the window) | 74 | 96 | · |
| Completed (customer came) | 46 | 62 | · |
| Missed | 20 | 25 | · |
| Cancelled | 6 | 7 | · |
| Still set or confirmed | 2 | 2 | · |
| Other status (rescheduled) | 0 | 0 | · |
| Show rate: completed / (completed + missed) | 69.7% | 71.3% | · |
| Appointments the machine booked | no data | no data | · |
How this is counted, and where it comes from
Definition. Appointments dated in the window, from the VinSolutions appointment report (scripts/lib/appointments-truth.ts, the same reader Home, /appointments and /reps use). Set = every appointment dated in the window, any status. Completed = VinSolutions status COMPLETED (the customer came). Missed = MISSED. Show rate = completed / (completed + missed): appointments still open or cancelled are left out of it.
Read with care.
- The feed keeps one appointment per customer (their latest) over a trailing 30 days, so a customer with two appointments in the window counts once.
- RESCHEDULED appointments are not a status the readers count (shown as 'other').
- The feed does not say who booked the appointment, so how many the machine booked is not measurable here (see the last row).
Source files.
data/store/vs-appointments.jsonas of 6:05 AM CT
5. Active deals and stalled negotiations: Deals being worked, and the ones gone quiet definition proposed Opportunities
| Measure | Last 14 days | Month to date | Right now |
|---|---|---|---|
| Active deals (What needs me, state Deal) | · | · | 34 |
| Stalled: a deal with no touch in 48 hours | · | · | 17 |
| Touched in the last 48 hours | · | · | 17 |
| Stalled share of active deals | · | · | 50% |
| Compare: queue 'live negotiation' bucket | · | · | 1 |
| Compare: queue 'stalled deal' bucket (no outbound in 7 days) | · | · | 32 |
How this is counted, and where it comes from
Definition. A deal is a customer with a price, payment or finance line in the notes this week, a live negotiation in the opportunity queue, a stalled deal that has written in the last 14 days, a showroom desking note in the last 2 days, or a credit app a person has since touched. Counted once per customer; someone more urgent (in the showroom, a reply to send, a hand-off, numbers owed, an appointment) shows in that state instead. Stalled means no touch, a message either way or a note, in 48 hours. Live, read when the page opens.
Read with care.
- The saved daily file still holds its own older count (55), from a different rule. This page shows the engine's number so /needs-me, the Showroom and /metrics agree.
- Notes are synced into the lead store about hourly, so a touch in the last hour can be missing.
- A machine cadence text counts as a touch. If a person's touch is the one that matters, the stalled number goes up; that is the open question.
- The queue's own 'stalled deal' bucket is a different, looser rule (no outbound in 7 days, any open lead) and is shown for comparison only.
Source files.
data/store/leads.json (Notes Summary, status fields)as of 6:48 AM CTmemory/message-log.jsonas of Nov 19, 9:41 PM CTdata/store/vs-appointments.jsonas of 6:05 AM CTdata/store/vs-sold.jsonas of 6:05 AM CTmemory/master-queue.jsonas of Nov 19, 8:42 PM CTlive: the What needs me engine (scripts/lib/needs-me.ts), the same object /needs-me and the Showroom read
6. Lead to sale conversion: Lead to sale by source, last 90 days measured Sources
| Measure | Last 14 days | Month to date | Right now |
|---|---|---|---|
| Leads (2026-08-23 to 2026-11-20) | · | · | 1,627 |
| Appointments set | · | · | 378 |
| Appointments shown | · | · | 253 |
| Sold | · | · | 279 |
| Lead to sale, all sources | · | · | 17.1% |
| Units sold (Axcessa retail + lease, date sold) | 30 | 38 | · |
By source (90 days)
| Source | Leads | Appointments | Shown | Sold | Lead to sale % |
|---|---|---|---|---|---|
| Referral / repeat | 327 | 66 | 52 | 73 | 22.3 |
| Phone-up | 228 | 100 | 71 | 57 | 25 |
| Walk-in / drive-by | 190 | 43 | 23 | 55 | 28.9 |
| Credit-app tools | 171 | 40 | 30 | 37 | 21.6 |
| TrueCar | 165 | 10 | 4 | 4 | 2.4 |
| Dealer site / Ford Direct / OEM | 161 | 50 | 33 | 22 | 13.7 |
| CarGurus | 111 | 15 | 8 | 9 | 8.1 |
| AutoTrader | 82 | 21 | 13 | 11 | 13.4 |
| VinCue trade tools | 69 | 20 | 13 | 6 | 8.7 |
| Carfax | 55 | 1 | 1 | 0 | 0 |
| Facebook / chat | 43 | 9 | 5 | 4 | 9.3 |
| Other | 25 | 3 | 0 | 1 | 4 |
How this is counted, and where it comes from
Definition. The /sources report's own 90-day window (scripts/lead-source-report.ts buildReport): leads by source group, and how many were sold. A lead counts as sold if VinSolutions says Sold, or the buyer's name is on an Axcessa retail/lease delivery dated on or after the lead (120-day cap), or the VinSolutions sold feed has the same customer. Lead to sale = sold / leads. Credit-app tool leads are folded into where the customer really came from when the lead type says so. Ford ICI list imports and cross-store transfers are left out. Units sold (window columns) = Axcessa retail + lease rows by date sold.
Read with care.
- Lead data pulled 2026-11-19T13:54:15.766Z.
- The lead pull (VinSolutions reporting API) refreshes weekly, Mondays; the report says when. Leads from the last few weeks have not had time to close, so the newest sources look worse than they are.
- Matching a delivery to a lead is by buyer name, so a buyer who used a different name on the lead is missed.
Source files.
data/store/lead-source-rows.jsonas of Nov 19, 7:54 AM CTdata/store/axcessa-cardeals.jsonas of Nov 19, 12:22 PM CTdata/store/vs-sold.jsonas of 6:05 AM CT
7. Manager time saved: Replies the machine handled without a person assumption Agents
| Measure | Last 14 days | Month to date | |
|---|---|---|---|
| Replies the machine sent on its own | 21 | 40 | · |
| Estimated manager hours saved (x 4 min, assumed) | 1.4 h | 2.7 h | · |
| Machine drafts a person approved and sent | 9 | 9 | · |
| Sent by a person through desk-send | 120 | 134 | · |
| All logged sends (any script) | 662 | 951 | · |
How this is counted, and where it comes from
Definition. Machine-handled replies = texts and emails the auto-responder sent on its own to a customer's message (message log, script vin-auto-responder; the Home strip's 'Auto-reply'). Time saved = those replies x 4 minutes. THE 4 MINUTES IS AN ASSUMPTION (a manager reading the message, looking up the vehicle and typing an answer), not a measurement; change MINUTES_PER_REPLY in scripts/metrics-baseline.ts.
Read with care.
- Only replies the machine sent alone count. Drafts a person approved are shown but not counted as saved.
- Texts a rep types inside VinSolutions are not in the message log, so the machine's share of all replies is overstated.
- Counts come from the saved daily file once the message log has rolled past a date.
Source files.
memory/message-log.jsonas of Nov 19, 9:41 PM CTdata/store/metrics-daily.jsonas of 6:50 AM CT
8. Reliability and automation: Scheduler, feeds and the dashboard measured Jobs
| Measure | Last 14 days | Month to date | Right now |
|---|---|---|---|
| Job runs | 35,524 | 50,508 | · |
| Failed runs | 692 | 738 | · |
| Scheduler failure rate | 1.9% | 1.5% | · |
| Timeouts (killed for running too long) | 39 | 41 | · |
| Silent-job alerts sent to Will | 13 | 16 | · |
| Stale-feed findings (health check, per feed per day) | 23 | 39 | · |
| Dashboard restarts by the watchdog | 6 | 15 | · |
| Feeds stale or unreadable now (of 22 with a cadence) | · | · | 3 |
| Jobs in a failure state now (some may be paused on purpose) | · | · | 35 |
| Health check: checks failing now | · | · | 3 |
| Auto-responder: answered on its own | 21 | 40 | · |
| Auto-responder: drafted for a person (held) | 226 | 303 | · |
| Auto-responder: stopped by a safety gate | 48 | 60 | · |
Jobs that failed most, last 14 days
| Job | Failed runs |
|---|---|
| vin-first-touch | 259 |
| vin-auto-responder | 201 |
| credit-app-clock | 37 |
| vs-daily-feeds | 31 |
| covideo-rep-activity | 25 |
| grid-delta-run | 15 |
By day (last 14 days)
| Date | Job runs | Failed | Timeouts | Silent-job alerts | Stale-feed findings | Dashboard restarts |
|---|---|---|---|---|---|---|
| 2026-11-07 | 2,976 | 5 | 0 | 0 | 1 | 0 |
| 2026-11-08 | 1,862 | 4 | 2 | 0 | 3 | 0 |
| 2026-11-09 | 2,831 | 3 | 0 | 0 | 2 | 0 |
| 2026-11-10 | 2,809 | 4 | 2 | 0 | 2 | 2 |
| 2026-11-11 | 2,673 | 2 | 0 | 0 | 2 | 1 |
| 2026-11-12 | 2,694 | 3 | 0 | 0 | 0 | 1 |
| 2026-11-13 | 2,777 | 14 | 0 | 0 | 3 | 1 |
| 2026-11-14 | 2,880 | 7 | 0 | 0 | 3 | 1 |
| 2026-11-15 | 1,912 | 0 | 0 | 0 | 3 | 0 |
| 2026-11-16 | 2,894 | 61 | 5 | 1 | 3 | 0 |
| 2026-11-17 | 2,939 | 7 | 0 | 0 | 0 | 0 |
| 2026-11-18 | 2,926 | 17 | 0 | 0 | 1 | 0 |
| 2026-11-19 | 2,951 | 563 | 30 | 12 | 0 | 0 |
| 2026-11-20 | 400 | 2 | 0 | 0 | 0 | 0 |
How this is counted, and where it comes from
Definition. Job runs = 'Finished:' lines in the scheduler log. Failure = a non-zero exit, except exits the failstate line calls a known expected outcome. Failure % = failures / runs. Timeouts = jobs the scheduler killed for running too long. Silent-job alerts = 'JOB SILENT' alerts the heartbeat watchdog sent Will (a job missed its own cadence; the watchdog sends at most one per job per cooldown). Stale-feed checks = distinct feeds the health check listed as stale or out of date, counted once per check per day. Dashboard restarts = times logs/watchdog.log found the dashboard down and restarted it (includes restarts while code is being deployed). Feeds stale now = Home's own data stamps (dashboard/views/stamp.ts) older than their cadence or unreadable.
Read with care.
- A job that exits 0 without doing its work is not a failure here. Health-check findings cover the known cases.
- The scheduler log is archived about 15 days back; older days come from the saved daily file once this job has run for them.
- Silent-job alerts have a cooldown, so a job silent for a day is one alert, not many.
Source files.
logs/scheduler.logas of 6:50 AM CTlogs/archive/scheduler-*.log.gzlogs/heartbeat-watchdog.logas of 6:28 AM CTlogs/watchdog.logas of Nov 14, 3:31 PM CTmemory/scheduler-failstate.jsonas of 6:45 AM CTmemory/health-status.jsonas of Nov 19, 6:50 PM CTmemory/responder-decisions.jsonlas of 6:49 AM CTdata/store/metrics-daily.jsonas of 6:50 AM CT
9. Additional opportunities identified: Opportunities the system found, and what was done with them measured Calls
| Measure | Last 14 days | Month to date | Right now |
|---|---|---|---|
| Missed-call callers recorded | 108 | 178 | · |
| acted on (called, reached, or phone-up created) | 4 | 7 | · |
| still open | 104 | 171 | · |
| Promises recorded (machine commitments + video promises) | 46 | 64 | · |
| acted on (marked fulfilled) | 13 | 20 | · |
| overdue and open | 24 | 24 | · |
| Queue: stalled deals to re-engage | · | · | 32 |
| Queue: missed appointments to reschedule | · | · | 34 |
| Queue: new leads with no first touch yet | · | · | 7 |
| Queue: live negotiations needing a person | · | · | 1 |
| Queue candidates acted on | no data | no data | · |
How this is counted, and where it comes from
Definition. Missed-call callers = numbers that called the sales line and got no one (CallRevu), first missed in the window, junk excluded; acted on = a person marked it called, reached, or a phone-up was created. Promises = commitments the machine made and video promises found in the thread, made in the window; acted on = marked fulfilled. Queue candidates = the opportunity queue's buckets right now (stalled deals to re-engage, missed appointments to reschedule, new leads with no first touch); the queue keeps no history of what was acted on.
Read with care.
- Opportunities are counted when the system first records them, not when they would have closed.
- 'Acted on' for the queue is not recorded anywhere, so it is no data.
- A missed call from a number already in the CRM may be a customer we were already working.
Source files.
memory/missed-call-callbacks.jsonas of Nov 19, 7:25 PM CTmemory/commitments.jsonas of Nov 19, 5:15 PM CTmemory/video-promises.jsonas of Nov 19, 7:00 PM CTmemory/master-queue.jsonas of Nov 19, 8:42 PM CT
10. After the visit: Customers who came: sold within 7 days, and followed up after a visit with no sale definition proposed
| Measure | Last 14 days | Month to date | |
|---|---|---|---|
| Customers who came (appointment completed), at least 3 days ago | 39 | 55 | · |
| sold within 7 days | 25 | 32 | · |
| on the same day as the appointment | 25 | 32 | · |
| on a later day | 0 | 0 | · |
| no sale on record | 14 | 23 | · |
| Sold within 7 days, share of those who came | 64.1% | 58.2% | · |
| Bought on a later day, share of those who did not buy that day | 0% | 0% | · |
| Showroom visits ended 24+ hours ago, no sale (judged) | 30 | 47 | · |
| person-typed call, text or email within 24 hours of the visit ending | 14 | 27 | · |
| Follow-up within 24 hours, share of those visits | 46.7% | 57.4% | · |
Same measures over the last 30 days (the whole appointment feed)
| Measure | Last 30 days |
|---|---|
| Customers who came (appointment completed), at least 3 days ago | 100 |
| sold within 7 days | 53 |
| on the same day as the appointment | 53 |
| on a later day | 0 |
| no sale on record | 47 |
| Sold within 7 days, share of those who came (%) | 53 |
| Bought on a later day, share of those who did not buy that day (%) | 0 |
| Showroom visits ended 24+ hours ago, no sale (judged) | 66 |
| person-typed call, text or email within 24 hours of the visit ending | 33 |
| Follow-up within 24 hours, share of those visits (%) | 50 |
How this is counted, and where it comes from
Definition. Sold from shown: appointments VinSolutions marks COMPLETED on a day in the window and at least 3 days old (time to close), matched to a sale from that day through 7 days later: an Axcessa retail, lease or fleet delivery matched by buyer name, or the VinSolutions sold feed matched by customer id or name. 'Same day' = a sale record dated the day of the appointment in either file (the other file may date the same deal a few days later, a marking or delivery lag, which is not a second sale); 'later day' = the only sale records are 1 to 7 days after. Follow-up: each showroom visit the reps logged (data/store/showroom-visits) that ended 24 or more hours ago with no sale that day, the customer in the lead store, not a buyer or opt-out, and the lead re-read after those 24 hours; followed = a call, text or email typed by a person within 24 hours of the visit ending (machine sends left out; the same reader as /needs-me visits, scripts/lib/follow-through.ts buildVisits). Both are proposed definitions (baseline.md measures 4 and 5, 10/21/26); the follow-up denominator is the /needs-me visits rule, which leaves out buyers and opt-outs, so it is smaller than baseline.md's first reading (every visit not flagged sold that day) and the two should not be compared.
Read with care.
- Same-day sales include customers who came to pick up a car already sold (their appointment is the delivery), so the sold share overstates how many new shoppers buy. The useful figure is the 'later day' share of those who did not buy that day.
- A buyer whose Axcessa name differs from the appointment name is missed (understates). Sales after the newest Axcessa file are not in yet, which is why the last 3 days of appointments are left out.
- A touch is read from VinSolutions note lines (the newest 20 per lead); a call made on a personal phone is invisible, and a visit nobody logged is not counted. Small windows swing: read n beside every percentage.
- The appointment feed holds one appointment per customer (their latest) over a trailing 30 days.
Source files.
data/store/vs-appointments.jsonas of 6:05 AM CTdata/store/axcessa-cardeals.jsonas of Nov 19, 12:22 PM CTdata/store/vs-sold.jsonas of 6:05 AM CTdata/store/showroom-visits/*.jsondata/store/leads.jsonas of 6:48 AM CT