A good AI calling list needs 18 fields in four groups: identity and consent, relationship history, the reason for the call, and call logistics. Each field gives the agent one thing to say or one rule to follow.
How to read these 18 fields
Every field below has to change what the agent says or does. If a column never alters a sentence or a rule, leave it out. Each item follows one pattern: what the field is, why it matters, and what the agent does with it on the call. Items 1 to 5 are identity and consent. Items 6 to 10 are relationship history. Items 11 to 15 are the reason for the call. Items 16 to 18 are call logistics.
1. First name and preferred name
Store the name the customer uses, not only the one on the account. A wrong name loses trust in the first five seconds. The agent opens with it, as in "Hi Funmi, this is…" Add a pronunciation note for names in languages the voice may stumble on, such as Yoruba or Igbo.
2. Phone number and number type
Store the number in full international format, with a tag for mobile or landline. Mobile numbers get the call. A landline may need a different time window or a human caller. Local numbers are more likely to be answered, so use a number from the customer's region where your carrier gateway allows it. If a number fails, send an SMS.
3. Preferred language
Store the language the customer wants to be called in, not the one on their paperwork. The agent starts in that language and does not switch unless asked. Buyers keep asking for local accents and languages, and a lender serving informal traders needs non-English calls. Voice campaigns can run in Yoruba and Igbo. The manager picks the language before the voice.
4. Consent status and source
Store whether the customer agreed to be called, plus where and when. In the US, the FCC's 2024 ruling treats AI-generated voices as "artificial" under the TCPA, so AI-voice calls need consent. Customers often ask how you got their number. The agent answers from the source field: "You gave us this number when you opened your account."
5. Do-not-contact flag
Mark every customer who has opted out, and check the flag before each dial. A duplicated campaign leaves out anyone added to the list since the original, so old copies do not call people who have opted out. If a customer says "stop calling" mid-call, the agent confirms and ends the call.
6. Customer segment or tier
Tag each contact with a segment such as retail, premium or high-net-worth. The tag sets the agent's tone and offer. A financial firm told us its older investor clients distrust unknown callers and expect a dedicated relationship manager. For that kind of tier, the agent opens with the firm's name and offers a human handover early.
7. Tenure and account age
Store the date the customer joined, or the account age in months. Tenure sets how deep the script goes. A new customer needs a short welcome and one clear next step. A long-standing one has heard the basics and wants the agent to get to the point. Existing customers are also the best list to call first, because they know your name.
8. Products held
List the products each customer already has, such as savings, a card or a loan. The agent skips any product the customer already owns, because pitching it again shows you have not looked at their account. It leads with the product that fits next, such as a card offer for a savings-only customer. If the list is empty, it asks an open question instead.
9. Last interaction date and channel
Store when the customer last dealt with you and how: call, email, WhatsApp, branch or app. The agent opens with that touch: "Last week you emailed us about your policy." Channels often live in separate tools. Log each one against a single customer ID, as insurers do with policy numbers. Then the agent never calls as if you had never spoken.
10. Last interaction outcome or open issue
Store how the last contact ended: resolved, pending or complaint open. If a complaint is open, the agent skips the pitch and gives a status update. Complaint handling is regulated in many places. In Nigeria, the central bank fined institutions N430M in 2025 for slow handling. Operators also call early about outages and delays.
11. Reason for this call (trigger event)
Store the event that makes this call happen, such as a signup, a KYC drop-off, an outage or a due date. The agent opens with it: "You started verifying your account yesterday." Triggered calls fire on an action in your CRM, and you can set a delay, such as a day after drop-off. Scheduled and recurring campaigns suit surveys and routine check-ins.
12. Journey stage and drop-off point
Store the last step the customer reached, such as signed up but not funded, or KYC started but not finished. The agent asks about that step: "You were one step from finishing." Lenders and investment firms lose customers at these same points, and emails alone do not fix it. A lender with thousands of approved leads who never borrowed learned that many assumed the loan was cash.
13. Amount due, due date and payment history
Store the amount due, the due date and whether past payments were on time. The agent states the exact amount and date, then reminds the customer to fund the account so the debit can go through. After a failed debit, it offers another payment channel. Reminders are not collections, and a happy customer pays. A clean history gets a friendly tone. A missed payment gets help.
14. Offer or approved limit
Store the exact offer for each customer: the approved limit, the rate, or the discount and its end date. The agent quotes that figure, such as "You're approved for up to 500,000", instead of a generic pitch. A specific number shows the call is about this customer. Some customers are approved for more than they asked for, so store the limit, not the request.
15. Stated churn reason or last feedback
Store the reason the customer gave for leaving, or their last survey answer or complaint tag. Win-back calls to lapsed customers lose many people because they do not recognise the caller. The agent names your brand first, then speaks to the reason: "You told us delivery was slow." If the reason is blank, it asks. A clean NPS score can hide real objections.
16. Best time window and time zone
Store each customer's best calling window and their time zone, so the agent never dials at a bad hour. Test your windows. For loan calls, mornings and evenings can answer better than midday, so check your own answer rates by hour. If the customer is busy, the agent offers a callback time.
17. Attempt count and last outcome
Store how many times you have tried each contact and how the last try ended. A carrier's temporary refusal should not use up a retry, and a call where only the AI spoke does not count as reached. Cap attempts at three or four, since many customers answer on the second or third try.
18. Human handover contact or account owner
Store a named relationship manager or a queue for each customer, so the agent knows where to send a handover. Gartner's 2026 survey found 87% of 3,566 customers say a human option is essential. Buyers ask who picks up a missed handover. EU AI Act Article 50 also requires the agent to say it is AI.
Map fields, then test the call
Upload your CSV, map each column to a field, then set the calling window, retry rules and do-not-contact settings. In Insight7 (beta), you can add contacts by CSV or by typing them in. A campaign also has consent marking, calling hours, retry rules, a do-not-call list and a test call to your own phone. You can also talk to the agent in your browser, with no phone number needed. Listen for a wrong name, a skipped consent line or a pitch for a product the customer already owns.
Drop any field the script never uses. Pass the agent only the personal data the call needs. Regulated buyers ask where call data lives, which AI models touch it and who owns it.
Related reading: How to Personalize AI Outbound Calls With CRM and Product Usage Data, AI Phone Surveys vs Online Surveys Compared by Response Rate, Cost and Insight Depth and How to Run Customer Research Interviews at Scale With AI Voice Agents.
Questions people also ask
What data should be in an AI calling list?
Include four groups: identity and consent, relationship history, the reason for the call, and call logistics. Keep a column only if it changes what the agent says or a rule it follows.
How do you personalize AI voice calls?
Open with the preferred name, then give the reason for the call, such as a drop-off, a due date or an approved offer. Use history to skip what the customer already owns.
What is the minimum data needed to start an AI outbound campaign?
You need a name, a phone number, consent status and one reason for calling. Add a calling window and a do-not-contact flag before launch. Start small, then add fields as needed.
Do you need consent to call customers with an AI voice agent?
Often yes. In the US, AI-voice calls already need consent under the TCPA. Other regions have their own rules, so store consent status and source for every contact.
How many times should an AI agent retry a contact?
Three or four attempts is a common cap. Space retries across different times of day. Never retry anyone on the do-not-contact list.
How do you keep customer data safe when feeding it to an AI voice agent?
Send only the fields the call needs. Know where recordings live, which models touch them and who owns them. Residency rules are tightening, such as Nigeria's central bank deadline of 1 January 2027 for payment data.
Should the AI tell customers it is an AI?
Yes. Disclosure is becoming law: the EU requires it and Colorado from 2027. Have the agent answer truthfully when asked, and consider saying so at the start.
Check consent before you dial
Every field earns its place by changing a sentence the agent says. Before you launch, check that every contact has a consent status filled in.

