Workflow template for customer support and service
Call summary check against the transcript
Before anyone acts on the summary of a service call, your quality lead sees every amount, date, outcome and statement in it that the transcript does not back, each with the line and time to check, and what the summary left out. It changes no record; the reviewer decides what is corrected.
Listed under
Add it to a workflow in the Fully Powered Planner, beside the teams and roles that will run it. Book a fit call
The outcome
The record of a call says what was actually said. A summary is the part of a call everyone else relies on: the next agent, the team approving a credit, the person answering a complaint. When it says a credit was applied and the agent only asked for one, or that the customer agreed to a payment plan when they said they would think about it, the next decision rests on something that never happened.
This workflow template reads the summary one claim at a time. Every amount, every date, every outcome and every statement about who said, asked or promised what gets one verdict from the transcript: supported, not supported, contradicted or cannot tell. Each verdict carries the transcript line, the time on the recording and the speaker, so the reviewer can check it in a minute. Offers recorded as done, conditions dropped, "next Wednesday" turned into the wrong date and amounts with two digits swapped all come to the surface.
Then it reads the transcript the other way, for what a service summary must carry and this one does not: a request for a supervisor, a disputed charge, a complaint, a promise with no date, a mention of an ombudsman. It works the same way on a summary an agent wrote and on one your call platform generated, so it also gives your team a fair way to judge generated summaries before trusting them.
The quality lead or team leader decides what happens next. The summary, the case and the agent's record stay exactly as they were.
Who starts the work, and when
The contact-center quality lead, a team leader or a quality analyst starts it once a call is over and has both a transcript and a summary: when the call is picked for quality review, when someone is about to act on the summary, or when a customer says the record of a call is wrong. Teams trying out generated summaries run it on a set of calls, one call at a time.
The quality lead or the team leader reads the check and decides, for each finding, whether the summary or the case is corrected and by whom.
What they supply
- The transcript, exported with a line number, the time from the start of the call, the speaker and what was said. The template names the column headers it expects; when yours differ, it stops and asks which is which.
- The summary, exactly as it stands on the case.
- The call details: call and case references, the date of the call, the queue, the agent, the outcome recorded on the case, and whether the summary was written by the agent or generated.
- Any later corrections noted on the case.
- Your settings: who reviews, who receives mentions of a regulator or a safety risk, your working days and closed days, and the list of what every summary must carry, if yours differs from the default.
What comes out
One check for one call, in a fixed order. The header gives the call, the limits of the transcript (hidden and unheard passages) and any line to raise straight away with your escalation role. The result line counts the claims by verdict and shows how many need the reviewer, and the outcome code on the case is compared with how the call actually ended.
Then the claims, contradicted first, then not supported, then cannot tell, each with the summary's words beside the transcript's, and the working for every date. What the summary left out follows, each item with its line and time, then the later corrections and what they touch, and the claims the transcript supports. Card numbers and security answers are never repeated.
An invented example
Illustrative example with invented data. The company, the customer, the call, the transcript, the summary and every figure below are invented. Company R is an invented supplier of electricity and gas to homes, with a billing team of twelve agents. Its call platform writes a summary of each call onto the case. A customer called about a high bill, and the billing team leader was about to send the case on for a credit. First come the inputs the billing team leader supplied, in full, then the check the workflow template returned.
The inputs
1. Settings
Reviewer: quality lead
Escalation role: complaints team leader
Run by: billing team leader
Today: Thursday 5 November 2026
Working days: Monday to Friday
Closed days: Wednesday 11 November 2026
Hidden mark: [masked]
Unheard mark: [inaudible]
Must-carry list: default
2. Call details
call_id: C-20418
case_id: CASE-7731
call_date: 2026-11-03
time_zone: Eastern
queue: Billing
agent: Agent 14
recorded_outcome: Resolved - credit applied
summary_source: generated by the call platform
summary_written: 2026-11-03 10:42
3. Transcript
line,time,speaker,text
1,00:00,System,"This call may be recorded for quality and training."
2,00:06,Agent,"Thanks for calling Company R billing. Can I take your account number?"
3,00:11,Customer,"It's [masked]."
4,00:19,Agent,"Thank you. And the first line of the address?"
5,00:23,Customer,"[masked]."
6,00:31,Agent,"That's verified. How can I help today?"
7,00:35,Customer,"My October bill is three hundred and twelve dollars. It's normally around a hundred and forty. I think the meter reading is wrong."
8,00:52,Agent,"Let me look. I can see the October bill was based on an estimated reading, not an actual one. The estimate was high."
9,01:10,Customer,"So it's wrong."
10,01:14,Agent,"It looks that way. I can request a credit of forty-two dollars sixty for the overestimate. It needs my team leader's approval before it goes on the account."
11,01:31,Customer,"Only forty-two? The bill is more than double."
12,01:37,Agent,"The rest is the higher usage in October from the cold spell, as far as I can see. Once we have an actual reading we can check it again."
13,01:50,Customer,"Fine. When can someone read the meter?"
14,01:56,Agent,"I can book a meter reader for next Wednesday, between eight and twelve."
15,02:04,Customer,"Wednesday the... sorry, which date?"
16,02:07,Agent,"Wednesday the [inaudible]."
17,02:10,Customer,"OK."
18,02:15,Agent,"Once the reading is in, the corrected bill will be sent within five working days."
19,02:24,Customer,"And can I pay the old amount in the meantime, the hundred and forty?"
20,02:31,Agent,"You can set up a payment plan for the balance. I could put you on three monthly payments of a hundred and four dollars."
21,02:42,Customer,"I'll think about it. I don't want to agree to anything until the bill's fixed."
22,02:49,Agent,"Understood."
23,02:52,Customer,"And I want a supervisor to call me back about this. This is the second time it's been estimated."
24,03:01,Agent,"I'll pass that on."
25,03:04,Customer,"If it's not sorted by the end of the month I'm going to the energy ombudsman."
26,03:12,Agent,"I understand. I've noted your account. Is there anything else?"
27,03:16,Customer,"No."
28,03:18,Agent,"Thanks for calling."
4. Summary, as it stands on the case
Customer called about a high October bill of $312.00. Agent confirmed the
bill used an estimated reading. A credit of $46.20 was applied to the
account. Meter reading booked for Wednesday 4 November, 8am to 12pm.
Corrected bill to be issued by 10 November. Customer agreed to a payment
plan of three monthly payments of $104. Issue resolved.
5. Later corrections
2026-11-03 10:50, agent, case note: "Correction: $42.60 credit requested,
pending team leader approval, not applied yet."
The output
CALL SUMMARY CHECK: call C-20418, case CASE-7731
For: quality lead. Prepared Thursday 5 November 2026 for billing team leader.
Call: Tuesday 3 November 2026, Eastern, Billing, Agent 14. Transcript: 28 lines, 03:18. Summary: generated by the call platform, written 2026-11-03 10:42.
Settings: working days Monday to Friday; closed days Wednesday 11 November 2026; must-carry list default.
Transcript limits: hidden 2 (lines 3, 5); unheard 1 (line 16); speaker unknown 0 (none).
Raise with complaints team leader: line 25 (03:04), the customer will go to the energy ombudsman if the bill is not sorted by the end of the month.
RESULT
Claims checked: 14 = supported 6 + not supported 2 + contradicted 5 + cannot tell 1.
Left out: 6. Needs the reviewer: 14.
Recorded outcome "Resolved - credit applied": does not match: line 10 (01:14) the credit needs the team leader's approval before it goes on the account; line 18 (02:15) the corrected bill waits on a reading; line 23 (02:52) the customer asked for a supervisor callback; line 25 (03:04) the customer will go to the ombudsman if it is not sorted by the end of the month.
CONTRADICTED
S4 Amount. Summary: "A credit of $46.20". Transcript: line 10 (01:14) Agent: "I can request a credit of forty-two dollars sixty for the overestimate." Note: forty-two dollars sixty is $42.60.
S5 Outcome. Summary: "was applied to the account". Transcript: line 10 (01:14) Agent: "It needs my team leader's approval before it goes on the account." Note: the agent offered to request the credit; no line says it was requested, approved or applied.
S10 Date. Summary: "by 10 November". Transcript: line 18 (02:15) Agent: "Once the reading is in, the corrected bill will be sent within five working days." Working: the date of the meter reading can be read two ways (see S7). From Wednesday 4 November: Thursday 5, Friday 6, Monday 9, Tuesday 10, Thursday 12 (Wednesday 11 closed): by Thursday 12 November. From Wednesday 11 November (a closed day, counted from all the same, starting the next working day): Thursday 12, Friday 13, Monday 16, Tuesday 17, Wednesday 18: by Wednesday 18 November. Tuesday 10 November matches neither.
S11 Statement. Summary: "Customer agreed to a payment plan". Transcript: line 21 (02:42) Customer: "I'll think about it. I don't want to agree to anything until the bill's fixed."
S14 Outcome. Summary: "Issue resolved." Transcript: line 10 (01:14) Agent: "It needs my team leader's approval before it goes on the account."; line 23 (02:52) Customer: "And I want a supervisor to call me back about this."; line 25 (03:04) Customer: "If it's not sorted by the end of the month I'm going to the energy ombudsman." Note: steps and a customer request were still open at the end of the call.
NOT SUPPORTED
S6 Outcome. Summary: "Meter reading booked". Transcript: discussed at line 14 (01:56) Agent: "I can book a meter reader for next Wednesday, between eight and twelve."; line 17 (02:10) Customer: "OK." Note: the customer accepted the offer; no line says the booking was made.
S9 Statement. Summary: "Corrected bill to be issued". Transcript: the condition at line 18 (02:15): "Once the reading is in". Note: the summary drops the condition.
CANNOT TELL
S7 Date. Summary: "Wednesday 4 November". Transcript: line 14 (01:56) Agent: "next Wednesday"; line 16 (02:07) Agent: "Wednesday the [inaudible]." Why: two readings: said on Tuesday 3 November, "next Wednesday" is Wednesday 4 November or Wednesday 11 November, and the date at line 16 is unheard; Wednesday 11 November is a closed day, so under that reading the visit would fall on a closed day. Listen at: 02:07.
LEFT OUT
L1 Sum of money. Transcript: line 7 (00:35) Customer: "It's normally around a hundred and forty." Note: the customer's rough figure for a usual bill.
L2 Complaint, dispute or refusal. Transcript: line 11 (01:31) Customer: "Only forty-two? The bill is more than double." Answer given: line 12 (01:37): "The rest is the higher usage in October from the cold spell, as far as I can see. Once we have an actual reading we can check it again."
L3 Customer request. Transcript: line 19 (02:24) Customer: "And can I pay the old amount in the meantime, the hundred and forty?" Answer given: line 20 (02:31): "You can set up a payment plan for the balance." Note: no line says whether paying $140 in the meantime is accepted.
L4 Customer request. Transcript: line 23 (02:52) Customer: "And I want a supervisor to call me back about this." Answer given: line 24 (03:01): "I'll pass that on." Note: no date and no owner given for the callback.
L5 Complaint, dispute or refusal. Transcript: line 23 (02:52) Customer: "This is the second time it's been estimated." Answer given: none.
L6 Lawyer, regulator, injury, safety or threat. Transcript: line 25 (03:04) Customer: "If it's not sorted by the end of the month I'm going to the energy ombudsman." Date read as: the end of the month is Monday 30 November 2026.
LATER CORRECTIONS
2026-11-03 10:50, agent: "Correction: $42.60 credit requested, pending team leader approval, not applied yet." Touches S4 and S5. The transcript supports part of it: the amount and the pending approval at line 10; it does not show that the credit was requested, only that the agent could request it.
SUPPORTED
S1 Statement. "Customer called about a high October bill": line 7 (00:35).
S2 Amount. "$312.00": line 7 (00:35).
S3 Statement. "Agent confirmed the bill used an estimated reading": line 8 (00:52).
S8 Date. "8am to 12pm": line 14 (01:56). Note: the transcript says "between eight and twelve" without am or pm, and the booking it belongs to is not supported, see S6.
S12 Amount. "three monthly payments": line 20 (02:31). Note: figures as offered ("I could put you on"); the plan was not agreed, see S11.
S13 Amount. "$104": line 20 (02:31). Note: figures as offered; the plan was not agreed, see S11.
DATA NOTES
None.
Draft for quality lead. The summary, the case record and the transcript are unchanged; nothing has been sent, and no agent has been scored.
Where it fits
- Customer support and service, where the contact center's quality lead owns the record of every call.
- E-commerce and retail, where a call summary promises refunds, replacements and delivery dates that the next agent will honor.
- Insurance, where a claim call's summary carries amounts, dates and what the policyholder said, and the claim moves on that record.
- Energy, where billing calls turn on estimated readings, credits, payment plans and visit dates, and a customer may take a complaint to the ombudsman.
- Finance and investment, where a service call about a charge or a payment has to be recorded exactly as it happened.
- Established businesses, where call summaries have piled up for years and nobody has read them against what was said.
- Growth & scaling, when call volume grows faster than the quality team and generated summaries start to carry the load.
What has to change for your company
Export transcripts with the column headers and speaker labels the template expects, or tell it yours. Tell it the marks your platform uses for hidden and unheard words. Review the list of what every summary must carry and add what your service must always record, such as a reference number given to the customer. Set your working days and closed days so counted promises such as "within five working days" come out right. Name the reviewer and the escalation role. Decide whether to check every call with a generated summary for a trial period or a chosen sample. It runs in ChatGPT, Claude, Copilot or Gemini from files and text you supply; it needs no connection to your phone or case system.
What has been checked
Fully Powered AI wrote this workflow template as original work. It was editorially reviewed on 11 November 2026. It was tried once on its invented example on 11 November 2026 in Claude; the record is published with it. It has not yet been tested in a client environment. It has not yet been tested as a packaged version in ChatGPT, Claude, Copilot or Gemini.
Testing in your environment means running it on past calls your quality lead has already reviewed by listening, with hard cases chosen in advance: an amount with two digits swapped, an offer recorded as done, "next Wednesday" with the date unheard, a refusal recorded as agreement, a request for a supervisor the summary leaves out, an export with different column headers. The quality lead checks every quoted line against the transcript and every unclear passage against the recording before anyone relies on a finding. Pilot acceptance-test preparation sets those cases up, and what we test before we call an AI workflow ready describes the record we keep.
Who decides
Your people keep
- Whether the summary or the case is corrected, and by whom
- What each unclear passage says, by listening to the recording
It does not
- Changes no record
- Never corrects the summary or the case
- Never repeats card numbers or security answers
- Needs no connection to your phone or case system
Questions people ask
Who is this for?
The contact-center quality lead, a team leader or a quality analyst, once a call is over and has both a transcript and a summary.
What do I need to supply?
The transcript, the summary exactly as it stands on the case, the call details, any later corrections and your settings. It needs no connection to your phone or case system.
Does it work on summaries our call platform generated?
Yes. It works the same way on a summary an agent wrote and on one your call platform generated.
Does it change the summary or the case?
No. It changes no record. The quality lead or team leader decides whether the summary or the case is corrected and by whom.
The full workflow template
Everything above is the outline. The complete workflow template is the working version: the roles, the inputs, the procedure, the exceptions and stop conditions, the output format, the reviewer checklist, the acceptance tests, the instructions for an AI assistant and what to customize for your company. It comes with a download and the link to its folder in Fully Powered’s public repository.
Free with an account. Sign in by email: no password and no card.
Next step
Put this workflow template to work for your team: on your plan, or built with you in a sprint.
Build this in a sprint
In a sprint, Fully Powered adapts this workflow template to your rules, proves it on your own work and puts it in the hands of the roles that own it.
Two ways to begin: