The Database Is Not a Call List
Co-Founder & CEO of CLRT
You have years of enquiries sitting in a CRM. Portal leads who asked for a floor plan and went quiet, viewing requests that never became viewings, investors who said maybe in a year that has since passed. You now have a voice agent that can hold many conversations at once and never tires. The arithmetic writes itself: point the agent at the database, ring every name, and harvest whatever interest is left. It is the first idea in almost every conversation about outbound AI in this city, and it is a reasonable instinct. It is also, under rules in force in the UAE since 2024, a fast way to commit several separate violations at once. The question is not whether the agent can make the calls. It is whether your database was ever a list you were allowed to call.
Start with what the database actually is. It is a record of people who once wanted something from you, captured for that purpose and rarely for any other. Some of them asked to be contacted about one launch and never about the next. Some said no on a call two years ago, and nobody wrote it down. Some have since added their number to the national Do Not Call Register. Some numbers now belong to someone else entirely. A human sales team works through this mess slowly, guided by memory and a sense of who is worth a second call. A voice agent pointed at the same file has no memory of any of it unless someone built that memory in. It will ring the refusal, the registered number and the stranger with the same warm opening, at the same pace, all afternoon.
The UAE regulated this exact activity in August 2024. Cabinet Resolution 56 of 2024, explained by the Ministry of Economy on 29 August that year, applies to every licensed company that markets by phone, free zones included. Its obligations read like a specification for a campaign rather than a call: prior approval from the competent authority, local numbers registered to the company, no contact with numbers on the Do Not Call Register, and marketing only to consumers who have said they want to hear it. On the call itself, calls run between 9am and 6pm, the caller states its identity and purpose at the start, every call is recorded and the recording is announced at the start, and the consumer is asked whether to continue. Afterwards, a refusal ends contact, repeat attempts are capped at once a day and twice a week, and every call is kept on record. Notably, the resolution expressly contemplates automated systems within its provisions. It did not ban the machine. It governed the campaign.
The companion resolution, Cabinet Resolution 57 of 2024, sets the price of getting it wrong. Resolution 57 lists 18 types of violation, with penalties from AED 10,000 to AED 150,000 depending on the type and nature of the breach. Individuals who market from numbers in their own name face AED 5,000 and the suspension of every number registered to them until the fine is paid, then AED 20,000 and three months without those numbers for a repeat within 30 days, then AED 50,000 and twelve months without telecom service for a further repeat within the next 30 days. Those numbers matter less than the multiplier. A tired salesperson who breaks the twice a week rule breaks it a handful of times. An agent pointed at an ungoverned list breaks it on every record it touches, in the same hour, and produces a perfect log of having done so.
The UAE is not an outlier, and the direction elsewhere is the same. For licensed financial institutions, the Central Bank's telemarketing regulation, described by Pinsent Masons in April 2026, requires prior express consent that names the language, channels and products a customer accepts, gives customers the choice between human agents, robocalls and AI agents, and requires logs kept for at least five years. In the United States, the FCC ruled in February 2024 that AI-generated voices are artificial voices under the robocall law, so calls using them need prior express consent, must identify who is calling and must offer an opt-out. In the European Union, the AI Act's duty to tell people they are dealing with an AI system became applicable on 2 August 2026. Four regimes, four different gates, one shared premise: the call is a permissioned act.
This is where the instinct inverts. The database is not a call list. It is the raw material for one, and turning it into one is the real work: deciding who has asked to hear from you and about what, which numbers are off limits today, which line the call leaves from, what hours it may run, and what a refusal does to every future contact. Done well, an outbound agent can be more disciplined than the status quo it replaces. Where salespeople still ring prospects from their own mobiles, which is precisely the case the individual penalties were written for, a governed agent calls only inside the hours, only from registered numbers, and records a no the moment it is spoken. Done badly, it is the same fine schedule with a better voice.
The law did not ban the machine. It governed the campaign, and the campaign is where the fines are.
A deeper dive
The hardest record in an outbound programme is the refusal, because it arrives as speech rather than data. Buyers rarely say the regulated word. They say not now, we bought already, take me off your list, call my husband, or simply end the call. Each of those means something different for recontact, and the rule that a refusal on the first call ends contact only works if the system decides, consistently, which of them was a refusal. That decision then has to travel. It must reach the human brokers who share the same CRM, the next campaign that someone loads in three months, and every other channel the firm uses, or the agent will be compliant on its own calls while the business breaks the rule around it. The frequency cap has the same shape: once a day and twice a week limits how often the company reaches a consumer, so it has to be counted across every caller, human or machine, not per campaign. And the register is not a one time scrub. A consumer can join it at any point, so the list that was clean when the campaign was planned is not necessarily clean on the day it runs. None of this is a model problem. It is a records problem that only looks like a voice problem.
The second-order trap sits in the opening seconds. The rules prescribe what happens before any selling begins: who is calling, why, that the call may be recorded, and whether the person wants to go on. That is the moment buyers are most likely to hang up, and a weak agent loses them there. The temptation that follows is predictable, which is to trim the opening until the numbers recover, and a firm that does so is trimming the law rather than the script. The only durable answer is the reverse: keep every required sentence and make the conversation after it good enough to be worth staying for. That is a quality problem, not a compliance one, and it is where most outbound programmes quietly fail. The firms that do this well will look, from the outside, as if they simply have better calls. What they actually have is a governed list, a remembered no and an opening that earns its next thirty seconds.
Work with CLRT
CLRT builds Closer, the sales agent that talks with buyers and shows them what they asked about on their screen, and its outbound calling is designed to run within the UAE's calling rules. The agent is the easy part to switch on. The work CLRT does first is the programme around it: who may be called and about what, when, and what a refusal does to every future contact, so the first campaign is one you would be comfortable defending. Talk to Tom, the agent itself, at clrtstudio.com/closer, then talk to CLRT before you load a single list.

Vishal Sachar is the Co-Founder and CEO of CLRT in Dubai. He helps businesses put agentic AI to work, leads practical AI workshops for leadership teams and operators, and writes on agentic systems, AI governance and the economics of automation. Reach him at vishal@clrtstudio.com or on LinkedIn.


