How to Monitor an AI Calling Campaign in Real Time

A campaign that looks healthy on a blended dashboard can be burning one country's list at a 6% answer rate while the overall figure reads 28%. AI call campaign monitoring is the shift discipline that catches that inside half an hour instead of at the weekly review. It's the difference between a retention head who knows why Tuesday went badly and one who has a feeling it did.

This guide walks through what to set before the first dial, the numbers to read every 30 minutes, the calls to listen to, how to keep spend on pace, and the exact conditions under which you pause. It's written for the ops lead or retention manager on shift while an AI agent dials dormant traders, unfunded registrations or KYC leads.

Key Takeaways

• Track dials, answered calls and real conversations as three separate numbers; a blended answer rate hides a single country or carrier that has stopped connecting.
• Under the US Telemarketing Sales Rule, abandonment must stay at or under 3% of calls answered by a person, measured per campaign over each 30-day period.
• FCA rule SYSC 10A.1.14R keeps relevant telephone recordings for five years, so a missing recording is a compliance gap before it's a QA gap.
• Topcalls bills $0.35 per minute all-inclusive, so spend on shift is talk minutes times 0.35, and pace is checkable on a calculator, not a finance report.
• Six conditions pause a campaign on the spot: a dialable opt-out, improvised investment advice, drops over ceiling twice, failed transfers, spend past cap, and a dead CRM sync.
• A warm-up batch of 20 to 50 numbers with three full listens catches most launch-day failures before the rest of the list is released.

1. What does AI call campaign monitoring actually watch?

AI call campaign monitoring is the live check an ops person runs while an AI agent is dialing: connect and answer rate, drops and silence, transfers reached, sentiment and opt-out flags, spend against cap, and a short list of stop triggers. It runs on a timer, every 30 minutes for a brokerage list, and ends in a shift log the retention head and the compliance officer can act on.

It's not the same job as quality assurance. QA samples recordings after the fact and scores the conversation; monitoring reads the numbers while they're still moving and decides whether the campaign keeps dialing. A campaign can pass a QA review on Friday and fall apart on Monday because a carrier flagged the caller ID over the weekend.

The broker angle matters because the list isn't a cold prospect file. It's dormant traders with money still in an account, registrations that never funded, and KYC leads who stalled at document upload. Every bad call costs a relationship the compliance desk can be asked about later, so monitoring is half a compliance function and half a revenue one. Topcalls shows the live figures in its real-time analytics, but the habit of reading them on a timer is yours.

2. What should you set before the first dial?

Before the first dial, name the person on shift and their escalation contact, record the list with its countries and the date of its last do-not-call screen, write down the floor for answer rate and the ceilings for drops, failed transfers and opt-outs, confirm the dialing window per trader time zone, and set the spend cap in the platform. Then dial a warm-up batch of 20 to 50 numbers and listen to three full calls.

Thresholds come from your pilot or your last human campaign on a similar list, not from a vendor deck. A first campaign with no history uses its own first hour as the baseline and tightens from there.

The time zone check bites brokers more than most. A list with UK, Cyprus and Dubai numbers has three dialing windows, and in the US the Telemarketing Sales Rule limits outbound calls to a residence to between 8:00 a.m. and 9:00 p.m. local time at the called person's location. The platform should hold a number until its own window opens. Smart campaigns in Topcalls do that per number, but read the setting rather than assume it.

Then check the CRM is receiving outcomes from the warm-up batch. Open three trader records and confirm the disposition, recording link and next action landed. A campaign that dials cleanly but writes nothing back has to be reconciled by hand later, and that's the kind of afternoon nobody logs.

3. Which numbers do you read every 30 minutes?

Every 30 minutes, record dials attempted, calls answered and calls that reached a real conversation, then drops, silence and failed dials. Compare each to its floor or ceiling and flag anything under or over for two consecutive checks. One bad half hour is lunch. Two in a row is usually a carrier problem, a flagged caller ID or the wrong time zone on part of the list.

Keep the three connection numbers apart. Answered includes voicemail and instant hang-ups; a conversation means the trader spoke back after the greeting. Teams that track only "answered" report a healthy campaign that's mostly talking to voicemail.

Split by country and by segment. A blended 28% can hide one country at 6%. The segments you built the list from (months dormant, last deposit size, language) are the same ones to read the answer rate by. And watch the share of answered calls that end within five seconds: a rising instant hang-up rate points at the caller ID, the opening line or a voice that doesn't fit the market. Call one of your own numbers on each carrier and read what the phone shows. A "Likely spam" label explains a sudden drop faster than any dashboard, and warming up a number is the fix, not dialing harder.

Drops need a listen as well as a count. Two dropped calls tell you whether the network cut the audio, the agent went silent or the trader hung up on a pause, and the three have different owners: carrier, platform, script. Topcalls answers in under 500 milliseconds; if a platform reports a rising response delay, traders start talking over the agent and calls degrade well before they drop.

Concurrency is the other quiet one. A campaign set to 40 concurrent calls that's running at 12 finishes the list on a different day and at a different cost per hour. Read the live figure, not the setting, and if it keeps slipping, our post on concurrent AI calls for brokers covers why.

The monitoring checklist puts all of this on one sheet: the thresholds to set before launch, the half-hourly connect, drop, transfer, sentiment and spend checks, seven stop triggers, and the end-of-shift log.

4. How do you watch transfers and human handoff?

Record three transfer numbers per check: offered, accepted by the trader, and reached a live agent. The gap between accepted and reached is your desk's problem, not the AI's. If traders keep saying yes and nobody picks up, stop offering transfers until the desk is staffed, or switch the agent to booking a callback with a time, an owner and a phone number.

Set a ceiling on hold time and make the desk hold it. A trader who agreed to talk about funding and then waited two minutes has been talked out of it. If the wait climbs, lower concurrency instead of losing transfers you've already earned.

Sit with one agent and watch a transfer arrive. They should see the trader's name, account state, what the AI already said and why the transfer happened, before the trader does. Then check the desk's schedule against the dialing window for every country on the list. Transfers offered at 18:30 local to a desk that closed at 18:00 all fail, and a failed transfer with a trader still saying yes on the line is the most expensive call of the day. The mechanics of a clean human handoff on AI calls deserve their own post, and they have one.

5. What should you listen for as well as count?

Listen to five random completed calls per hour for the first two hours, then three per hour, picked by call ID or time stamp rather than the ones the dashboard surfaces. You're checking four things: every opt-out lands on the suppression list within minutes, the agent answers plainly when asked if it's a machine, each frustration flag has one cause you can name, and nobody was promised a return.

Opt-outs are the one that doesn't wait. Pull one opted-out number and search for it in the calling platform and in the CRM. If it's still dialable in either, pause and fix the sync before you continue.

The financial promotion check is the broker-specific one. Listen for improvised answers to "is it a good time to trade gold". The script should redirect; if the agent states a return, a guarantee or an opinion on a specific instrument, that's a stop trigger and the recording goes to the compliance officer, not the shift log. Same for any trader who mentions a complaint, a regulator, a lawyer or a withdrawal problem. Those aren't reactivation leads. They're service cases with a recording attached.

Confirm recordings and transcripts are being written for every completed call while you're there. Under the FCA's SYSC 10A.1.6R a firm must take all reasonable steps to record telephone conversations that relate to activities in financial instruments, and SYSC 10A.1.14R keeps those records for five years, up to seven if the FCA asks. Open the last five completed calls and check each has both. Topcalls includes recording and transcription in the per-minute price, so a gap here is a settings problem, not a budget one.

6. How do you keep spend on pace?

Spend on shift is talk minutes times the per-minute rate. With Topcalls at $0.35 per minute all-inclusive, billed by the second, 400 conversations averaging two and a half minutes is 1,000 minutes, or $350. Every check, divide spend so far by hours elapsed, multiply by hours left, and compare that projection to the shift cap. If it passes the cap, lower concurrency now rather than hitting a hard stop mid-conversation.

Cost per dial always looks cheap. Cost per conversation and cost per transfer reached are the numbers your retention head asks about, and they move whenever answer rate or handoff success moves. Compare both to the pilot at every check.

Watch average conversation length too. Shorter than the pilot by more than a minute usually means traders are hanging up early, which sends you back to the connect-rate section. Longer usually means the agent is looping on an objection. Note voicemail, no-answer and failed-dial minutes separately from conversation minutes as well; on a big dormant list, voicemail drops are short but they add up.

Caps get raised during a good hour and never lowered again. Read the setting in the platform, don't trust memory. To size the cap in the first place, the dormant trader revenue calculator turns list size, expected answer rate and average deposit into the minutes a reactivation campaign can afford.

7. When should you pause an AI calling campaign?

Pause on the spot for any of six conditions: a single opted-out number found dialable, the agent heard giving investment advice or skipping the AI disclosure, drop rate over ceiling for two consecutive checks, transfers accepted but fewer than half reaching a human for a full period, projected spend past the campaign cap or the CRM no longer receiving outcomes, and answer rate on any country under half its floor for two checks.

Each has its own restart condition, and they're not interchangeable. A dialable opt-out restarts only after the suppression sync is verified on that number and on a sample of ten others. A compliance trigger restarts when the compliance officer signs it, not when the script is patched. A drop-rate pause restarts when a 20-call warm-up batch comes back clean. A country under half its floor gets its caller ID label and local time checked first, because dialing harder into a flagged number makes the flag permanent.

US brokers have one more hard line. Under 47 CFR 64.1200(a)(7), a telemarketer may not abandon more than three percent of calls answered live by a person, measured over a 30-day period for a single calling campaign, and a call counts as abandoned if it isn't connected to a live sales representative within two seconds of the person's completed greeting. The FTC's Telemarketing Sales Rule carries the same 3% figure and requires records establishing compliance. How that rule reads when an AI agent answers instantly is a question for your counsel; what monitoring can do is log every transfer attempt and every hold time, so the record exists. When a pause turns into a full outage, the campaign failure recovery post covers the restart order.

8. What goes in the end-of-shift log?

The shift log records the same nine totals in the same order every shift: dials, answered, conversations, transfers reached, callbacks booked, opt-outs, drops, minutes and spend. Then every trigger that fired with the time and who restarted, every setting changed, the call IDs of five calls worth hearing, and what the next shift should watch first. It goes to the retention head, the compliance officer and the vendor contact the same evening.

Nine numbers in a fixed order means the week's logs read as a table without anyone building one. Five call IDs picked with a reason each, two good, two bad, one odd, get listened to; fifty don't. Threshold changes are proposals in the log, decided by the escalation contact, never made in the log itself.

A shift log is what you compare a dashboard against, not the other way round. If you're choosing what the screen on the wall should show, the AI calling campaign dashboard post lists the panels that earn their place, and our AI cold calling benchmarks give industry ranges for connect and conversion rates to set the first floors against.

9. When this doesn't fit

Half-hourly monitoring doesn't fit a campaign of a couple of hundred numbers, a single-country list on a caller ID that's been clean for months, or a broker with no desk to transfer to. The first finishes before the second check, the second rarely moves between checks, and the third has no transfer section to watch. Run the before-launch section and the shift log, and skip the timer.

It also doesn't replace QA. If your question is "was that a good conversation", that's a recording review with a scorecard, and it happens weekly, not on shift. And if compliance hasn't signed the script and the disclosure wording, monitoring is the wrong tool; nothing on this sheet catches a script that was never approved.

If you'd rather set your thresholds with someone who has watched a few of these run, book a 30-minute call. We'll look at your list, your desk hours and your countries and send a proposal within 48 hours. Reactivating dormant traders is the campaign most brokers run first, and the customer reactivation page shows what that looks like end to end, from list to shift log.