Designing AI Voice Flows for Common Trader Objections

The first thing a dormant trader says on a reactivation call is almost never yes. AI voice objection handling for brokers is the work of deciding, before the first dial, what the agent says to "I lost money", "is this a robot?" and "just send me an email", and where each answer may go. Most brokerage teams write the pitch and improvise the rest, and that's how a $0.35-a-minute call becomes a complaint on the compliance desk.

This guide covers the ten objections dormant traders raise most, the three-move branch that handles each one, the lines a regulated agent must never cross, and how to test the flow before it runs at volume. Topcalls handles 63,000+ AI calls a day, and what separates a campaign that reactivates accounts from one that generates opt-outs is the objection branches, not the opening line.

Key Takeaways

• An objection branch has three moves: acknowledge in the trader's words, answer with one checkable fact, offer exactly one next step. Two rounds per objection, then close.
• The FCA's PS19/18 caps retail CFD margin at 30:1 to 2:1 and bans cash inducements to trade, so "will I make it back?" gets approved risk wording, never a figure.
• ESMA's 2018 CFD measures require a standardised risk warning stating the percentage of retail accounts that lose money. An AI agent reads the approved sentence; it doesn't paraphrase it.
• The FCC ruled on 2 February 2024 that AI-generated voices are "artificial" under the TCPA, so US calls need prior express consent and an identity disclosure at the start.
• Each objection gets its own disposition code, OBJ_FEES rather than "not interested", reviewed weekly so you know which branch to rewrite next.

1. Which objections do dormant traders raise most on AI calls?

Ten objections cover nearly every dormant-account call a broker makes: I lost money, no time to trade, moved to another broker, who are you, is this a robot, fees too high, a withdrawal problem, just email me, not interested, and I'll think about it. Pull the last 50 call notes from your retention desk, tally them, and rank. If your top objection isn't on that list, add it before writing anything else.

The ranking matters because the branches aren't equal. "I lost money" is the emotional one. "Fees too high" is the only one you can answer with a number you publish. And "I had a problem with a withdrawal" isn't an objection at all, it's a complaint wearing one, and it goes straight to a person.

Objections also sound different by market. A trader in Madrid says he moved brokers with a shrug; a trader in Dubai wants to know who gave you the number first. Write a row per language where the wording differs, not a translation of the English line. Topcalls runs calls in 32 languages, so the limit is your approved wording, not the platform.

Still on the opening line? Start with AI caller scripts for dormant traders, then come back here.

2. How does AI voice objection handling for brokers work?

Every branch runs the same three moves. Acknowledge the objection in the trader's own words, without softening it. Answer with one fact the trader can check: the account is still open, the balance, the published spread on the pair they traded. Then offer exactly one next step, a dated callback, one email, a transfer or a clean close. Cap each objection at two rounds. A third attempt at the same objection is where complaints start.

"I understand your concern" fails because every call centre says it. "So the last gold trade went against you and you closed the app" works because the trader hears that the agent read the account. Two options in one sentence get a no to both.

Write the branch for the trader who says nothing, too. Silence after the answer is common on dormant-account calls, and the agent needs a bridging line ("Does that match what happened on your side?"). Traders also talk over the answer; how an AI call handles interruptions covers the mechanics, and the flow rule is that the branch still resolves when the objection lands mid-opening.

The Trader Objection Handling Flow Template lays out all ten objections with their acknowledge, answer and next-step cells, the banned money lines, the end and handoff rules, and a disposition code per objection, in a form your sales lead and compliance officer can sign off together.

3. What can an AI agent say when a trader says "I lost money"?

Acknowledge the loss plainly, state only what's true about the account today, and offer an account-manager callback or a clean close. The agent gives no return figure, no view on where gold or EUR/USD is heading, and no "now is a good time". Under the FCA's PS19/18 and ESMA's CFD measures, a retail trader hears a standardised risk warning, not a reason to trade again.

The rules are specific. The FCA's PS19/18, in force for CFDs since 1 August 2019, limits retail leverage to between 30:1 and 2:1 by volatility of the underlying, closes positions at 50% of required margin, requires a standardised risk warning stating the percentage of the firm's retail accounts that lose money, and stops firms offering "cash or other inducements to encourage retail consumers to trade". ESMA's measures, agreed on 23 March 2018, set 30:1 for major currency pairs down to 2:1 for cryptocurrencies, with the same risk warning and incentive restriction.

For the flow, that turns into four hard rules the agent can't talk around:

• "Will I make it back?" The agent reads the approved risk wording, then offers education material or a human. No estimate, no reassurance about outcomes.
• "What should I trade?" No instrument, no direction, no position size. The branch redirects to education material or to a licensed person. "I'd start small" is advice, and it's banned too.
• "Is there a bonus?" One approved offer sentence, written by compliance, or "there's no offer". The agent never improvises an incentive.
• "I can't afford to lose more." That's a vulnerability signal. The sales part of the call ends, the agent gives support contact details, and the call is flagged for compliance review.

Each answer cell carries a dated sign-off. That log is what you hand the regulator if a trader complains about what the agent said, and it's a large part of why brokers run customer reactivation on an AI agent rather than a script deck every rep reads slightly differently.

Put your own dormant-account count and average deposit through the dormant trader revenue calculator. Whatever the number, the losses branch is the one carrying most of it, because past losses are the main reason those accounts went quiet in the first place.

4. How does an AI voice agent answer "Is this a robot?"

With a yes, first time and every time. "Yes, this is an automated call from [Brokerage] about the account you opened in March. A person is available if you'd prefer." The agent then says what the call is for and continues only if the trader is fine with that. No dodge, no pause, and the same answer whether the question comes at the start, mid-answer or after the close line.

In the United States this is settled law. On 2 February 2024 the FCC adopted Declaratory Ruling FCC 24-17, confirming that the TCPA's restrictions on "artificial or prerecorded voice" "encompass current AI technologies that generate human voices". An AI-voice call needs the called party's prior express consent (written consent for telemarketing), an identity statement at the start of the message, and an opt-out method. In the FCC's release, Chairwoman Jessica Rosenworcel said: "Bad actors are using AI-generated voices in unsolicited robocalls to extort vulnerable family members, imitate celebrities, and misinform voters. We're putting the fraudsters behind these robocalls on notice."

For a broker that's a design constraint, not a threat. The disclosure line lives in the opening, next to the brokerage name and the reason for the call. "Take me off your list" is a command, not an objection: confirm, log the suppression, end. Disclosure rules differ by country, so the exact line is written per market with your compliance officer; the wider picture is in AI disclosure laws by jurisdiction.

Two more identity checks belong here. The data-source line has to be true: "your number is from your registration with us" only works if that's where it came from, and a bought list needs a GDPR and TCPA review before the first dial. And the trader needs a way to verify the call, a callback number on your website or an email from the official domain. Topcalls AI voice agents carry the disclosure, recording notice and opt-out per campaign within a TCPA, TSR, DNC and GDPR compliance posture, so the compliance desk reviews one flow instead of a hundred recordings.

5. When should the AI call end or hand off to a person?

Hand off any withdrawal, deposit or account-access complaint, any suitability question, any request for a person, and any trader above the balance threshold you set. End on an explicit opt-out, a second refusal, hostility, a wrong number, a vulnerability signal or the time cap. Every trigger has a written close line, a disposition code and a rule on whether a retry is allowed.

Handoffs aren't failures of the flow; they're the flow working. A withdrawal complaint gets a ticket reference and a callback window, never an answer branch. A suitability question gets a callback with a licensed person. "Can I speak to a person?" gets "of course, connecting you now".

What travels with the transfer matters as much as the trigger. The objection heard, the branch taken, the trader's own words and the CRM record link all go with the call, so nobody asks the trader to repeat himself. When nobody's available, the fallback is a callback at a set time, confirmed on the call and written to the CRM. Warm transfer versus callback is covered in human handoff on AI calls for forex brokers.

The end triggers are shorter and stricter:

• Explicit opt-out: "Don't call me", "remove my number" and "stop" all hit the same branch. Confirm, suppress, thank them, end. Disposition DNC, no retry.
• Second refusal: one acknowledge-and-answer round is fair. The second "not interested" gets the close line and OBJ_NOT_INTERESTED, with no further attempts in this campaign.
• Hostility: profanity aimed at the agent, threats or repeated shouting. The agent closes politely and the call is flagged for human review, not queued for a retry.

6. How do you test an objection flow before it goes live?

Run one test call per objection row, ten calls minimum, and check that each reaches a next step or the close without the agent inventing a line. Interrupt the agent mid-answer on three of them, stack two objections in one breath, and ask "is this a robot?" at three different points. Your compliance officer listens to every recording and signs off the answer cells, the risk wording and the disclosure lines with a date.

Then check the plumbing. Ten calls, ten disposition codes, ten next actions in the CRM. If a code is missing or mapped wrong, fix it before the first real dial, because without OBJ_FEES you'll never learn that fees are the objection killing the campaign. One code per objection, not one per outcome, each with an owner. AI call disposition automation covers the CRM side.

Once live, review the codes weekly. The branch with the most volume and the lowest next-step rate is the one to rewrite first, and a change log per answer cell (date, old wording, new wording, who approved) tells you which edit moved the number. On Topcalls the dispositions land in real-time analytics per campaign as calls complete, and the transcripts feed the call quality assurance checks your QA lead runs weekly.

First setup takes about 15 minutes and a campaign is typically live within two weeks, most of that being sign-off. To run the ten test calls with us on the line, book a 30-minute call and bring your current script; you'll have a proposal within 48 hours.

7. When doesn't a scripted objection flow fit?

A written objection flow doesn't fit high-balance accounts that already have a named account manager, complaint-heavy lists where most calls are really support tickets, or markets where you don't yet have approved risk and disclosure wording. In those cases the AI call is a router, not a closer, and the honest design is a short identification line plus a handoff.

• Top-tier accounts. If a trader above your threshold knows his account manager by name, an automated objection branch reads as a downgrade. Route those calls to the person.
• No approved wording. If compliance hasn't signed the risk statement and the disclosure line for a market, don't dial that market. Improvisation is the thing the flow exists to prevent.
• Numbers without consent. A bought list with no consent record isn't an objection problem. No branch fixes it, and in the US the FCC ruling makes the call itself the violation.

Everywhere else, the flow is the product. Write the ten branches, get them signed, run ten test calls. After that, objections stop being surprises and become a weekly report with a code next to each one.