After a campaign ends, you decide who gets a second call, who gets another channel and who is left alone. That choice depends on what each call produced, not the answer rate. Here is how to read AI call campaign results.

Count only calls that truly reached someone

Start by cutting the list down to calls where a person actually spoke. An answered call is not a reached call. Voicemail, a line that connects to silence and a call where only the AI spoke all show as "answered" in a raw log.

Carrier refusals need the same care. A temporary refusal is a network problem, which makes it a poor reason to retire a contact. Make sure it does not use up one of the contact's retry attempts.

Then rebuild the funnel in four steps:

  1. Dialled: every contact the campaign tried.
  2. Answered: someone or something picked up.
  3. Real conversation: the customer spoke and the call went past hello.
  4. Goal met: the customer did what the campaign asked.

Read transcripts to fill in steps 3 and 4. Yes/no answers and call outcomes should come from the transcript, so you are not guessing.

Here is one lender's test, rounded. Across three rounds it called about 150 customers, and about 130 calls were answered. Around 65 became real conversations, and about 30 customers asked for a loan. Half the answered calls turned into conversations, and about a quarter ended in the goal. Those two rates tell you who to call next.

Sort every contact into outcome segments

Give every contact one outcome label, and write the rule for each label before you assign any. Each label can then be matched to the campaign goal.

Use these seven segments:

  • Goal met: the customer did what the campaign asked, such as requesting a loan.
  • Interested, not done: the customer said yes or maybe but did not finish the step.
  • Asked for callback: the customer named a time or asked for a person.
  • Not reached: voicemail, no answer or a carrier refusal.
  • Answered, no speech: the line connected but the customer said nothing. Only the AI spoke, so this does not count as reached.
  • Declined or do-not-contact: the customer said no or asked you to stop.
  • Wrong number or failed: a dead line, the wrong person or a technical error.

Campaign type changes what "goal met" means. A finished survey is a win for a customer survey. For a win-back, only a customer who agrees to return counts. Write the goal in one sentence per type.

Then check the labels by hand. Read a small sample of transcripts from each segment, and grow the sample for any segment that looks messy. Look for calls in the wrong bucket, such as a "declined" call where the customer only asked to be called later. If you find several, fix the rule and re-run.

Store consent and do-not-contact flags on the contact. A "goal met" customer who later opts out must stay out of every future list.

Read patterns by time, attempt and script

Split your reached contacts by when, how often and how far the call went. Then look for the cuts where the rates differ.

Time of day. Group answer rates by the hour of the dial. Midday often answers worse than mornings and evenings, but test it on your own list before moving your start time.

Attempt number. Count how many answers came on the first, second and third dial. Many customers only pick up on a later try, so three to four attempts often beat one. If most answers land on attempt one, extra retries mostly add cost.

Script. Read transcripts to see where customers went quiet or hung up. A drop-off at the same sentence across many calls points to that sentence. Asking customers to use an app mid-call is a common weak spot, because mobile data can drop.

Check each group's size before you change a rule. A cut with a few dozen calls can flip on a handful of answers.

Calling hours and consent are also legal questions, and they differ by region. In the US, an FCC ruling from February 2024 brought AI-generated voices under the TCPA. In the EU, the AI Act requires AI that talks to people to say it is AI. Other regions have their own rules, so check yours before you move your window.

Pick the follow-up audience and set its rules

Map each segment to one action, one campaign type and one channel. Write the map down before you build.

Segment Follow-up action Campaign type Channel
Goal met Thank, then leave alone None Suppress
Interested, not done Nudge to finish the step Lead follow-up Retry call or SMS
Asked for callback Call at the time they named Lead follow-up Human call
Not reached Retry in a better window Same as original Retry call
Answered, no speech Retry once, then switch Same as original SMS
Declined or do-not-contact Never contact None Suppress
Wrong number or failed Fix the data first None Suppress until fixed

Send interested or sensitive contacts to a person. A disputed charge or an upset customer should never meet a second AI call. Gartner's August 2026 survey of 3,566 customers found that 87% say a company using GenAI for service must offer a way to reach a human. Plan the handover before launch.

Then reset the rules for the new goal:

  • Calling hours: use the window your time-of-day cuts supported.
  • Retry rules: set the attempt count from your attempt data, and space retries over different days.
  • Wave cap: decide how many follow-up waves one contact may get before they are retired. Adjust the number to your own data.
  • Consent marking: confirm every contact is marked.
  • Test call: talk to the agent before launch.

Insight7's voice campaigns are in beta, and the setup below follows its spec, so check it in the app. A campaign has a type, consent marking, calling hours, retry rules, a do-not-call list and a scorecard when it ends. Calls where only the AI spoke are not counted as reached, and a carrier's temporary refusal does not use up a retry. A reviewer can read each transcript when the call ends. To save time, duplicate the old campaign with its contacts. People added to the do-not-contact list since the original are left out of the copy. You can test the agent by talking to it in the browser, with no phone number needed.

Prove the follow-up worked with a clean test

Hold out a control group before you launch. Pick a random slice of your chosen segment and do not contact it. The gap between its conversion and the called group's is the lift. Without a control, you cannot tell whether customers would have acted anyway.

If you run calls and SMS side by side, give each channel its own audience. One option is a sublist that only voice can reach, so any conversion from it belongs to the call. A second is a parallel voice arm on the same customer set as an SMS campaign, so you can compare conversion rates.

Define the success metric before launch. Use conversion to the campaign goal, such as loan requests per 100 reached customers. Calls made and answer rate show how busy the system was.

Measurement is also a weak spot in AI service. A Gartner analysis of 432 use cases found only a quarter showed a return, and 42% could not be measured.

Set a stop rule too. Divide the campaign cost by the extra conversions over the control. Keep going only while that cost stays below what a customer is worth to you.

Related reading: AI Outbound Campaign Report Template for Sharing Results With Leadership, How to Define Success Criteria for an AI Voice Agent Pilot and Outbound Call Campaign Benchmarks for Answer Rate, Completion Rate and Conversion.

Questions people also ask

What is a good answer rate for an AI call campaign?

No single number fits every list. Compare each round with your own earlier rounds. Track reached customers and real conversations per answered call, because "answered" includes voicemail and silence. Local numbers are also more likely to be answered.

How many times should you redial a customer in an outbound call campaign?

Start with three to four attempts and test from there, because many customers only pick up on a later try. Spread them across different hours and different days.

What is the best time of day to run an outbound call campaign?

Test it on your own list. Mornings and evenings often answer better than midday. Split results by dial hour and check that each hour has enough calls to trust.

How do you measure the success of an outbound call campaign?

Measure conversions to the campaign goal per 100 reached customers. Compare that with a held-out control group, and divide the campaign cost by the extra conversions. Calls made only show activity.

Who should you include in a follow-up call campaign?

Include contacts who said yes or maybe, asked for a callback, or were never reached. Cap the waves so no one is chased forever. Leave out customers who met the goal, declined or gave a wrong number.

How do you compare AI calls and SMS in the same campaign?

Give each channel its own audience and judge both on the same goal. A parallel voice arm on the same customer set as an SMS campaign lets you compare conversion rates.

When should an AI call hand the customer to a human?

Hand over when the customer asks for a person, raises a sensitive issue or disputes an amount. Design that handover before launch, including who chases a handover nobody picked up.

Close

Results are only as good as the segments you build from them. Label every contact by what its call produced, hold out a control group, and keep the do-not-contact list clean. Rule of thumb: count conversations, then call back only people who said yes or maybe.

Book a demo of Insight7