The untested belief
Ask any service business owner whether their phone is answered well and the answer is yes. Ask when they last listened to one of their own calls and the answer is usually a pause.
That gap is the most reliably expensive thing in service-business operations, because the phone is where almost all revenue is decided and it is the least observed part of the company. Every invoice gets checked. Every technician's work gets seen by a customer. The calls happen, end, and vanish.
As of September 2026, the businesses that improve their conversion rate without spending more on marketing are almost always the ones that started scoring their own calls. Not because their staff were bad, but because nobody had ever specified what a good call does, and you cannot consistently deliver an unstated standard.
This guide builds that standard as a short scorecard, sets up a review habit that survives past week three, and explains how transcripts and sentiment analysis turn the whole thing from a listening chore into a filtered queue. The system referenced throughout is Run with Jarvis — the short explanation of what it is sits in the Jarvis AI overview.
The seven things a good service call does
A scorecard has to be short enough to use and specific enough to argue about. Seven items, in the order they occur.
1. Answered fast. Inside a few rings, by something competent. Everything downstream is irrelevant if this fails, and it fails more often than owners believe — usually because the person answering was doing something else.
2. Name and callback number captured early. Early, not at the end. A call that drops at minute four with no number is a total loss; the same call with a number captured at minute one is recoverable. This single item is the cheapest insurance on the card.
3. Job type correctly identified. Not "they want a locksmith" but which service, because the price, the duration, the parts and the technician all depend on it. Misidentification here produces a confidently wrong quote, which is worse than no quote.
4. Specifics captured. Address or location, vehicle or equipment details, the symptom in the customer's own words, access constraints. Whatever your trade needs to dispatch accurately. This is the item that determines whether the technician arrives able to do the work.
5. A price or a clear next step given. Either a number, a range with stated conditions, or an explicit "this needs a look and here is what that appointment costs and what it determines". What fails is the vague middle — "somebody will get back to you with pricing" — which is where most enquiries quietly die.
6. The booking asked for. Explicitly. "Shall I get you on the calendar for Thursday morning?" is the item, and it is the single most commonly missing one across every trade. A call can score well on the first five and still end with nobody having asked for the business.
7. The caller left knowing what happens next. Who is coming, roughly when, what it will cost, what they need to do to prepare. A customer who hangs up unsure of any of those calls back for reassurance, or does not.
Seven items, each observable from a recording. That observability is the whole point: a score derived from evidence can be discussed, while a score derived from impressions becomes an argument about tone.
Scoring without turning it into a tribunal
The mechanics should be almost boringly simple. Each item is a yes, a no, or not-applicable. No weightings, no ten-point scales, no composite index. A call is a string of seven marks, and the useful output is not the individual call's total — it is the pattern across the sample.
The pattern is what tells you something actionable. If item six fails on seven of ten calls, you do not have a staffing problem, you have a missing habit and possibly a missing script line. If item four fails on the calls that later produced a second trip, you have found the cause of your callback rate. If item five fails specifically on one service type, your price sheet has a hole in it.
Two disciplines keep this useful rather than corrosive.
Score the call, not the person. The unit of analysis is the interaction. Naming people in the review turns the exercise into performance management, and the moment it becomes performance management the incentive shifts from handling calls well to producing scoreable calls, while honest reporting of upstream problems stops.
Separate the two causes of a bad score. Some failures are coachable — nobody asked for the booking, the number was captured late, the caller had to repeat themselves. Others are structural — there is no price for that service, the booking rules are ambiguous, the calendar will not accept the slot the customer wants, the discount policy is undefined so the person freezes. Coaching a structural failure is unfair and ineffective, and it is the fastest way to lose the team's cooperation with the whole exercise.
In practice the first honest review round usually finds that a third of the failures are structural. Fixing those is faster and more valuable than any coaching.
A weekly habit that actually survives
Ambition kills this. A business that resolves to review every call reviews none within a month.
Ten calls, once a week, thirty minutes. That is the whole commitment, and it is enough because a pattern common enough to matter will appear in ten calls.
The sample should be deliberately mixed, not random: a few that ended with no booking, a few that booked, one unusually long call, one very short one, and anything the team flagged as strange. Only reviewing lost calls shows you failure without calibrating what good looks like. Only reviewing wins tells you nothing.
The output should be one page and two decisions: one coaching point to raise with the team this week, and one structural fix to put in the queue. Not seven improvements — one of each. Reviews that generate long lists produce no change, because nothing on a long list is anybody's job.
Then the loop closes by watching the scorecard in following weeks to see whether the item you targeted moved. If it did not, the intervention was wrong — usually because the problem was structural and got treated as coachable, or because the fix was a prohibition with no alternative attached. "Stop doing X" without "do Y instead" almost never changes behaviour; the person hits the same situation, has no other move, and does X again.
Where transcripts change the economics
Everything above works with a stack of recordings and a person willing to listen. The reason most businesses stop is that listening does not scale — sampling ten calls blind means listening to a lot of perfectly fine calls to find the two that went wrong.
Transcription inverts that. A searchable text record of every call lets you select the sample instead of drawing it. You can go straight to the calls where a price was quoted and no booking followed, the calls where the customer repeated the same information twice, the ones that ran unusually long, the ones with a specific service mentioned.
Sentiment and intent analysis push it further, by surfacing the calls where tone deteriorated or where the caller's intent did not match what got recorded — which are precisely the calls a random sample misses and the ones with the most to teach. Lead scoring adds the commercial filter: the high-value enquiry that did not convert is worth more review attention than the price-shopper who was never going to book.
On Run with Jarvis, AI transcription, lead scoring, sentiment and intent analysis, and call recording with playback sit in the Pro tier and above; the tier contents are on the pricing page. The practical effect is that a two-hour listening session becomes a twenty-minute review of pre-selected calls, which is the difference between a habit that lasts and one that does not.
There is a further use for the same data once it exists: asking questions of it in plain language rather than building reports. That is what the natural-language operations layer is for, described in the Jarvis AI Brain overview — "which calls quoted over four hundred dollars last week and did not book" is a more useful question than any fixed dashboard answers.
| Scorecard item | Observable from | Typical failure | Coachable or structural |
|---|---|---|---|
| Answered fast | Ring duration, abandon records | Answered by someone mid-task | Structural — capacity, not effort |
| Name and number early | First minute of the recording | Left to the end, call drops first | Coachable |
| Job type identified | Mid-call, versus the invoice later | Generic category, wrong price | Both — often a script gap |
| Specifics captured | Intake fields on the record | Address only, no access or symptom | Coachable with a checklist |
| Price or next step given | Explicit statement on the call | "Someone will get back to you" | Structural — missing price sheet |
| Booking asked for | One sentence, present or absent | Never asked at all | Coachable, highest-yield item |
| Caller knows what is next | Closing thirty seconds | Ends vaguely, callback follows | Coachable |
What not to score
Three things belong off the card, because scoring them does more harm than good.
Tone in isolation. A warm call that captures nothing is a failed call; a brisk, efficient call that books the job is a good one. Scoring friendliness independent of outcome rewards the wrong thing, and it is the item most likely to feel personal.
Script adherence word-for-word. The goal is the seven outcomes, not a recitation. People who are made to read scripts sound like people reading scripts, which customers detect instantly and dislike.
Call length. Short is not efficient and long is not thorough. A three-minute call that books a job cleanly beats a nine-minute one that ends in "we will call you back", and the reverse is also true. Length is a symptom worth investigating, not a target worth setting.
One more caution. Do not score the phone against an imaginary ideal customer. A substantial share of inbound volume is price-shopping, wrong numbers, and enquiries outside your service area, and handling those quickly and honestly is good work, not failure. What good looks like on a price-shopping call specifically is worked through in the handling price shoppers guide — and the answer is usually more specific, not more defensive.
Recording, consent and access
None of this exists without recordings, and recordings carry obligations that vary by jurisdiction and by who is on the call. That makes recording practice a question for your attorney, not a software toggle — and the sequence matters: decide the practice, then build the habit on it, not the reverse.
The operational decisions to make deliberately are what notice callers receive, who inside the business can access recordings, how long they are retained, and what happens to them when someone leaves. The call recording consent and compliance guide covers the shape of those questions. Nothing in this guide should be read as legal advice on any of them.
One internal-governance point is worth stating plainly: recordings are customer data. Access should be limited to people who need it for coaching or dispute resolution, not open to everyone with a login.
Where the scorecard pays off beyond the phone
Two downstream effects tend to surprise people who start scoring calls.
Callback and rework rates fall. Item four — capturing the specifics — is the single largest driver of second trips. A technician dispatched without the access constraint, the equipment model or the real symptom either cannot do the work or does the wrong work. Scoring intake completeness is therefore a direct intervention on your rework cost, which is the same number tracked in the warranty callbacks and rework guide.
Marketing spend gets readable. Once call quality is roughly consistent, differences in conversion between channels start meaning something about the channels rather than about who happened to answer. Before that, a campaign judged on booked jobs is really being judged on whoever picked up the phone that week — which is why attribution and lead scoring only become trustworthy after intake quality stabilises, as the lead scoring guide sets out.
There is also an uncomfortable finding almost every business hits in the first month: the most common single failure is item six, nobody asking for the booking. It is not a skills gap and not a motivation gap. It is that asking was never specified as part of the job, so it happens when the person remembers and not otherwise. Making it an explicit, scored step is usually the highest-return change on the entire card.
What to automate, what to keep human
Automate the capture and the filtering: recording, transcription, the intake fields on the record, sentiment and intent flags, the selection of which calls are worth a human's attention, and the weekly count of how each item is trending.
Keep human the judgement: what a good call sounds like in your trade, whether a failure was the person or the system, what the coaching point is, and whether a price rule needs changing. A model can flag that a call went badly; deciding what to do about it is the owner's job, and it is where the value actually is.
Where to start
Pull ten calls from last week — a mix of booked and not — and score them against the seven items by hand. It will take half an hour and you will learn more about your revenue than from a month of dashboards.
Then fix whatever is structural before coaching anything, because coaching a missing price sheet is unfair. Then put the ten-call review in the calendar as a recurring half hour with a named owner, because a habit without an owner is an intention.
Then, once the habit is real, let transcription and sentiment pick the sample so the review stays a half hour rather than growing into an afternoon. Tier contents and minutes are on the pricing page — three flat monthly plans, no setup fee, month-to-month. To talk it through against your own call volume, get in touch.
