Every campaign call your team makes, or a call center makes in your name, leaves a recording. That recording is the only full account of what was said, yet a manager can hear only a handful, and the rest get judged by an outcome code.
How to review call recordings at scale: define a good call, decide whether a sample answers your question, score against the script but read the words before deciding, keep the customer's own words beside the score, check conduct on every kind of call, show what was not checked, and have a person record each decision.
This guide is for managers who run sales, survey, charity or client campaigns, in-house or through an outside call center. The last section turns the steps into a checklist for choosing call review software and tests one option, Dagmor, against it.
1. Write down what a good call is
You cannot run a call quality review against a standard that lives in a manager's head. Put it on one page:
- The required steps, in order: greeting, reason for the call, the questions or offer, the close, the next step. Mark which steps must happen in sequence.
- The required wording: lines that must be said exactly, such as a disclosure or a consent question, written word for word.
- The lines nobody crosses: false claims, pressure after a clear no, rudeness or abuse, on every call whatever the outcome.
Treat the page like any other standard operating procedure: one owner, a version date, an update whenever the script changes. If an outside call center makes the calls, it is also the brief you hand over.
2. Call sampling or every call?
Sampling is sound if you know which question it answers. The NIST/SEMATECH e-Handbook of Statistical Methods says the main purpose of acceptance sampling "is to decide whether or not the lot is likely to be acceptable, not to estimate the quality of the lot." Applied to calls: a sample can tell you whether a campaign looks broadly on script. It cannot hand you the particular calls that need action.
A worked example (arithmetic, not measured results): a campaign has 1,000 completed calls, and some skipped a required disclosure. Each cell is the chance that a random sample of that size contains none of them.
| Calls that skipped it | Listen to 20 | Listen to 50 | Listen to 100 | Listen to 200 |
|---|---|---|---|---|
| 10 (1 in 100) | 82% | 60% | 35% | 11% |
| 20 (2 in 100) | 66% | 35% | 12% | 1% |
A manager who listens to 20 calls has an 82% chance of hearing none of the ten; at 200 calls the sample still misses all ten about one time in nine.
- A sample is enough when the question is about the whole campaign, such as whether a new opener works.
- Check every call when one call matters on its own: a missing disclosure, a false claim, a sale that may need reversing, or a client who wants the evidence.
3. Check calls against a script, then read the call
A score lets you compare agents and weeks. Keep three measures apart, because each points to a different fix: steps covered, order, and required wording said as written. An agent who covers every step out of order needs coaching; one who skips the disclosure needs a different conversation.
Treat a low score as a pointer, not a verdict. Before marking a call down, read the transcript or listen to the flagged part. The customer may have hung up before the close, or the agent may have covered the step in other words. Speech-to-text also mishears names, numbers and accents.
4. Keep the customer's own words beside the score
Outcome codes and ratings are closed answers: sale, no sale, callback. Easy to count, they lose the reason. Pew Research Center's guide to writing survey questions shows what a fixed list can hide: in one poll it describes, 43% of people asked the open-ended version gave an answer not listed in the closed-ended version.
On calls (an illustrative example): a charity call is coded "declined", but the person said, "Not this month, try me after payday." That is a callback the code drops.
So group the reasons behind each outcome (objections, refusals, requests to call back), report each group's share with a few real quotes underneath, and flag any call where the score and the words disagree.
5. Check conduct on every kind of call, not only the sales
Review effort follows the money: sales and completed surveys get scored because they are paid for. Conduct problems do not stay in that pile: pressure after a clear no shows up on refusals, and a required line goes missing on a short call.
Run a conduct check across every call type for false or misleading claims, swearing or abusive treatment, unacceptable pressure, and missing required wording. Each concern should point to the words that raised it.
6. Know what was checked and what was not
Every review has gaps: files that never arrived, recordings that were silent or cut off, calls the transcription could not read, flags nobody has opened. A report that shows only the checked calls looks cleaner than the campaign was. Put a coverage line on every report: recordings received, checked, unreadable (and why), and results still waiting for a person.
7. Record the decision, made by a person
Software can find and sort calls; a person decides what they mean. For every flag that leads to action, record the call (date, agent, recording file name), what was found quoted from the call, the decision (cleared as a mistaken flag, coaching, customer called back, client informed), and who decided.
That record lets an agent see why a call was marked down and a client see that problems were dealt with. With an outside call center, it is the review point our guide to delegating without losing control of quality asks for.
How to choose a call review tool or service
First, the honest case for not buying one. With a handful of agents and a few dozen calls a week, a manager with the written standard and a review sheet can listen to all of them. A tool earns its place when volume outruns listening, or when you must show a client that every call was checked. If nobody on your team has set up call review before, find vetted operations consulting firms to write the standard and routine once.
| Look for | Why | Ask |
|---|---|---|
| Works from completed recordings you already have | Test without changing how you call | Can we start with one finished campaign? |
| Uses your script and required wording | The score should measure your standard | How do we load or correct the script? |
| Shows the evidence behind each flag | A flag without the words is an accusation | Can I open the transcript and the original recording? |
| Shows coverage and gaps | You need to know what was not checked | Where do unreadable recordings show up? |
| Keeps the customer's words | Codes lose the reason | Can I see the words behind each answer group? |
| A person decides | Agents and clients can question it | How is a mistaken flag cleared and recorded? |
| Price and delivery agreed up front | Review work is easy to under-scope | What will this campaign cost, and when do we get it? |
| A small first test | Judge it on your calls, not a demo | What does a first campaign commit us to? |
Weight the rows by risk: where one bad call can cost a client or harm a customer, evidence and the human decision come first.
A worked option: Dagmor against the checklist
Dagmor describes itself as call recording review and campaign reports, in two parts: Dagmor Calls checks the work (script, Hero Score, conduct and coverage), and Dagmor Analysis keeps people's own words beside the numbers. Its summary line: "We check completed calls. People make the final decisions."
- Completed recordings. You send a completed campaign of call recordings and keep your originals; Dagmor works from copies. "No change to your phone system is needed for a first test." It checks finished recordings, not calls in progress: "Dagmor does not listen live or automatically prove someone lied."
- Your script. You can give Dagmor your exact script and required words, or let it draft the repeated questions and steps from the calls for you to correct. Its own caution: "A draft cannot tell you about a required step that every example left out."
- Evidence. The Call Log finds a completed call by date, recording filename, agent or result and opens its highlighted transcript; the filename leads back to your original recording. Hero Score, by agent on scored sales and completed calls, "averages the agreed performance checks", shown with missed steps, script covered and script order.
- Conduct beyond the sales calls. The Dagmor conduct check covers honesty, customer treatment and the client's required wording, across the other calls too. "A concern points to the spoken words; a person decides what happened."
- Coverage. Reports show which recordings were received and checked, which could not be read and which results still need a person. "Gaps stay visible."
- The customer's words. What People Said, in Dagmor Analysis, groups answers with their percentages and people's exact words underneath (reasons, objections, callbacks) and flags scores the words disagree with.
- A person decides. Human Review brings concerns into one place to read the words, clear a mistaken flag or record what needs attention. "Nothing is confirmed against an agent by the machine alone."
- Price and delivery. In its words, "We agree the work, price and delivery before starting," and "Processing is not instant." Dagmor's price is 4 cents per recorded minute, Calls and Analysis included, rounded to the nearest euro per campaign, with "No bands, bundles or minimum charge."
Dagmor was "built on a call centre floor in Carndonagh, Co Donegal, Ireland", and it has run there every day since 2025. Dagmor's founder on hand review there: "A manager can listen to two or three calls in every hundred." It suits teams that check campaign calls against a script, including for a client, and you need not learn the system: "Or simply send the recordings and take the reports."
Before you start, agree the delivery date for your volume, which Dagmor settles with you along with the work and price, and pick a first campaign whose calls are typical of your customers. On the first campaign, have one of your reviewers read some flagged and unflagged calls and compare. If you need someone listening while calls happen, this is the wrong kind of tool.
FAQ
Does call review software replace my QA people?
No. Software can transcribe, score and flag; a person still reads the words, clears mistaken flags and decides what happens to an agent, a customer or a client. The work moves from finding calls to deciding about them.
What if an outside call center makes the calls for us?
Write into the agreement that you receive the recordings of completed calls, which script and required wording apply, and how flags are handled. Then share the decision record, so disputes are settled from the words, not memory.
Does checking every call mean nothing slips through?
No. Recordings can be missing or unreadable, transcripts can mishear, and a check only finds what the standard asks about. That is why the coverage line and the human decision matter as much as the score.
How to review call recordings from your next campaign
Start with the one-page standard from step 1, run one finished campaign through the seven steps, and compare the result with what your usual sample told you.
The overview of Dagmor's call recording review software shows its example reports, built on fictional calls. If you want Dagmor to check a campaign, its terms are: "Try one campaign with no payment card. Stop after it and owe nothing; continue and pay the agreed price for it." On Dagmor's site, a button and the footer link read "Arrange a first campaign"; the brand's short form is "Try one. Stop, owe nothing."
Disclosure: this guide was produced in collaboration with Dagmor, which appears here as the worked example.