AI and automation can remove delays from onboarding, payments, compliance, risk management and support. But when clients cannot understand a decision, challenge it or reach someone accountable for it, efficiency becomes a source of distrust.
A client rarely cares whether a withdrawal, document check or support request was processed by a person or a machine when the result is fast and correct. The distinction becomes important when a payment is delayed, a document is rejected or an account is restricted without a useful explanation. Financial firms are increasing their use of AI, with an ESMA survey across EU securities markets finding that 70% expected to raise AI investment between 2025 and 2027, but the technology is entering client journeys in which a faster answer is not always a better one.
For FX and CFD brokers, automation now affects many of the moments that shape a client’s confidence in a firm. It verifies identities, classifies clients, monitors transactions, approves withdrawals, detects risk, routes enquiries and provides first-line support. Behind the client interface, it also handles reconciliation, reporting, liquidity management and the operational information employees need when investigating a problem.
FinanceFeeds approached Daniel Aristidou, Team Leader in Quantitative Research at Exness; Christoforos Soutzis, Europe CEO at Capital.com; Remonda Kirketerp-Møller, CEO of Muinmos; Tom Higgins, CEO of Gold-i; and Lars Holst, founder and CEO of GCEX, for their views on where automation improves service and where firms risk going too far.
The contributors cover both sides of the market. Exness and Capital.com provide the retail broker perspective, while Gold-i supplies trading, liquidity and risk technology to brokers. GCEX provides FX, CFD and digital-asset services and technology to institutional and professional clients, while Muinmos automates regulatory onboarding and client lifecycle processes. GCEX uses Muinmos within its compliance infrastructure. Holst chairs the RegTech provider’s board, while Kirketerp-Møller sits on the GCEX board.
Where Automation Removes Real Friction
Daniel Aristidou, Team Leader in Quantitative Research at ExnessAutomation is often discussed as one technology, although the term covers several different functions. Workflow automation moves information and routes cases. Rule-based systems apply predefined payment, compliance or risk policies. Machine-learning systems identify patterns, while large language models interpret questions, summarise information and generate responses.
The strongest use cases tend to be procedural and repeatable. Clients receive little value from waiting for an employee to copy information between systems, compare standard documents manually or provide an answer already available in the broker’s records.
Daniel Aristidou, Team Leader in Quantitative Research at Exness, said automation should remove that type of friction without eliminating access to support.
“Automation should make a client’s life easier, not harder. Clients shouldn’t have to wait on a support agent for information they could get instantly, so the point isn’t to replace human contact, it’s to cut out the friction and let the support team spend their time where it actually helps. Done right, it’s better for the client and the team.”
Exness applies that model to payments through automated withdrawal approvals available around the clock, alongside 24/7 in-app live chat. A routine withdrawal is a strong candidate for automation because the client normally wants the process completed rather than discussed. The service test comes when a transaction falls outside the standard journey and requires investigation.
Christoforos Soutzis, Europe CEO at Capital.comCapital.com has applied automation to another early point in the relationship. The broker expanded Trulioo’s Person Match identity-verification technology across 17 countries. Capital.com and Trulioo reported that the firm was verifying and onboarding 80% more new customers in Latin America and 28% more in Asia following the integration. Those figures describe changes in volume, rather than approval quality, processing time or the portion of growth attributable solely to the technology. FinanceFeeds previously examined the deployment in its coverage of Capital.com’s expansion of automated identity verification.
Christoforos Soutzis, Europe CEO at Capital.com, said automation delivers its clearest benefit when the client is waiting for a factual outcome rather than a decision requiring discretion.
“Automation delivers the clearest benefit in the parts of the client journey that are procedural and repeatable: identity verification, document checks, routine account administration, and the monitoring systems behind risk management. These are areas where consistency and speed genuinely improve the experience, because the client is waiting for a factual outcome, not a judgement call.”
Muinmos extends automation beyond identity verification into client classification, suitability, appropriateness and risk assessment. These processes determine what a firm may offer a client, through which entity and under which regulatory regime. FinanceFeeds has previously examined how KYC workflows, surveillance and reporting are becoming connected across CFD brokers as compliance technology moves deeper into their operating infrastructure.
Remonda Kirketerp-Møller, CEO of MuinmosRemonda Kirketerp-Møller, CEO of Muinmos, said firms should not confuse manual administration with meaningful personal service.
“If I’m opening an account and the onboarding team emails me three times for information they could have sourced digitally, that isn’t meaningful human interaction. It’s an administrative burden with a person attached.”
“Most of KYC and KYB onboarding can now run end to end without any form of interaction; data collected and verified in the background, checks running as the client moves through the journey, the questions adapting automatically to the client type and to the risks that actually surface. Thus making the experience for the client pleasant and almost frictionless and that’s how our Muinmos platform is designed.”
Removing a person from those stages does not necessarily make the service less personal. It can release employees from routine files and give them more time for the smaller number of cases where the standard process cannot provide an adequate answer.
Where Clients Still Need Human Judgement
The dividing line changes when a decision involves distress, disagreement, vulnerability or material financial consequences. A system may gather the facts and identify the rule involved, but a client disputing a trade or explaining financial difficulty needs a response that accounts for the circumstances rather than repeating the standard policy.
Soutzis said complaints, disputed transactions and support for vulnerable clients require a person who can adapt and remain accountable.
“Where clients still expect a person is anywhere a decision involves discretion, distress, or nuance: complaints, financial difficulty, disputed trades, or situations where something has gone wrong. Vulnerable clients in particular need someone who can adapt to their specific circumstances rather than follow a script. Our approach is to let automation handle the structured, factual layer of a request and hand anything involving judgement to a person who stays accountable for the outcome.”
For brokers serving UK retail clients, that approach is consistent with the FCA’s expectations under the Consumer Duty. In its review of support for vulnerable customers, the regulator said firms should prevent unreasonable barriers, design support around customer needs and monitor the quality of the support delivered. It also recognised that some friction can be appropriate when it helps customers understand and assess their options.
A fast journey is therefore not automatically a good journey. An appropriateness assessment completed in seconds may be efficient, but the result is of limited value if the client does not understand it or cannot obtain an explanation of why access to a product was restricted.
Kirketerp-Møller said automation relocates the human element towards exceptions and higher-risk cases rather than removing it.
“None of this removes the human element, it relocates it. Every routine file an analyst isn’t manually pushing through is time returned to the cases that genuinely need judgment: complex ownership structures, cross border complexity, exceptions, higher-risk clients, the file that doesn’t fit the pattern. Those cases have always existed. What changes is whether the team has the capacity to look at them properly, or whether they get the same fifteen minutes as everything else in the queue.”
“And where clients do want a person, it is usually for one thing: an explanation. Why am I being asked for this? Why have I been classified this way? Why was I declined? Which is in effect what the regulators are also asking. Automation can remove almost every other touchpoint in onboarding. It never removes the obligation to have someone who can answer that.”
The need for human involvement also appears in B2B relationships, although the issues differ from retail support. GCEX serves hedge funds, brokers, family offices and professional traders across FX, CFDs and digital assets. Its clients may be less concerned with navigating a standard form, but onboarding complex ownership structures, arranging custody, negotiating liquidity or resolving settlement and execution issues requires people who understand the institution’s mandate.
Lars Holst, founder and CEO of GCEXLars Holst, founder and CEO of GCEX, said firms should decide task by task rather than treating automation as one scale running from manual to autonomous.
“Where clients still expect a human is anywhere a decision carries judgement. Hedge funds, brokers, family offices and overall professional traders are not looking for a faster form necessarily. They want someone who understands their mandate and their risk appetite, and who can have a real conversation when execution gets complicated or conditions shift. Onboarding a large mandate, structuring custody arrangements, negotiating liquidity terms; all these still need someone on the phone.”
“The mistake firms make is assuming automation is a single spectrum from manual to fully automated that applies evenly across the business. It does not. It is task by task and getting the split wrong in either direction costs you client trust.”
The Trust Problem Begins When Nobody Can Explain the Decision
Clients can accept that a regulated broker must perform checks, restrict some activities or request additional information. Trust deteriorates when the broker cannot explain which information produced a decision, which rule was applied or who has authority to review it.
For brokers using external compliance, risk and support systems, the underlying processing may be outsourced, but responsibility for the outcome remains with the regulated firm. ESMA has identified customer support, fraud detection, risk management and compliance as AI use cases covered by existing MiFID requirements. The use of a vendor or model does not create a separate category of service outside the firm’s regulatory responsibilities.
Holst said automation can perform the work without taking ownership of the result.
“Nobody is running onboarding and KYC manually anymore, and nor should they be. For document verification, sanctions screening, transaction monitoring, machines are better at these than people are. They are faster, more consistent and they do not make the fatigue-driven errors that turn into real regulatory exposure. The caveat is that a person still has to own the outcome. So, automation does the work, but accountability does not transfer with it.”
That accountability becomes particularly important when automation operates across several regulatory regimes. GCEX has a UK entity authorised by the FCA for FX and CFDs, a Danish investment firm for FX and CFDs, a separate Danish crypto-asset service provider authorised under MiCA, and a Dubai entity licensed by VARA for broker-dealer services.
Holst said this regulatory coverage makes automation necessary at institutional scale, but it does not change who must answer for a decision.
“At GCEX this is where automation earns its keep first. We operate under FCA, MiCA, MiFID and VARA rules simultaneously across our different entities, and running that manually across regimes is not realist at institutional scale. Our partner Muinmos handles that layer.”
“Firms overreach when they treat automation and AI as a substitute for accountability rather than a support for it. The moment a client asks why something happened and the honest answer is “the system decided,” you have lost something you will not easily get back. Automation should make a decision easier to explain, not harder. Automate risk management without keeping a human who can explain and own the outcome, and you are building a liability, not an advantage. At the end of the day, a regulator asking why a transaction was flagged, or why it was not, needs an answer from a person.”
Kirketerp-Møller said explainability depends on whether the financial firm understands and controls the rules its system applies.
“Where firms go too far isn’t really a question of how much they automate. There is no volume of automation that is inherently too much. The line is whether the firm can still explain what its automation did, and why.”
“That comes down to how the rules are configured. If a firm can see the rule set, set it itself, and trace exactly which rule fired for a given client and on what basis, a fully automated journey is entirely defensible, the firm has automated the processing, not the thinking. If the system simply returns an answer and nobody can trace where that answer came from, the firm is no longer applying its own risk appetite. It’s applying the vendor’s, without realising it.”
The regulatory evidence supports that concern. The European Banking Authority reported that more than half of serious AML and counter-terrorist-financing compliance failures reported to its EuReCA database involved improper use of RegTech. The authority identified inadequate internal expertise, poor governance and insufficient oversight as obstacles to responsible implementation.
Kirketerp-Møller said the finding demonstrates why firms cannot outsource their understanding of a compliance decision.
“The EBA reported last year that over half of serious compliance failures reported to its EuReCA database involved the improper use of RegTech tools. Improper use, not defective tools. Firms deploying systems they didn’t fully understand.”
“So the firms that get this right won’t be the ones that automate the least. They’ll be the ones that chose their rules deliberately, can explain any decision the system reached, and can defend it in a room with a regulator. You can outsource the processing. You can’t outsource the decision, or the understanding of how it was made, and this is a focus area for all regulators.”
B2B Automation Shapes the Service Clients Receive
Client service is often associated with chat, email and telephone support, but some of its most important determinants sit behind those channels. A broker cannot explain an execution problem quickly if trade and liquidity data are split across systems. It cannot provide predictable settlement if reconciliation remains delayed, and it cannot resolve an account discrepancy efficiently if different departments are working from conflicting records.
This is where the B2B perspectives from Gold-i and GCEX become central. Their automation is focused less on replacing a client conversation and more on improving the operating infrastructure behind it.
Gold-i’s MatrixNET platform provides multi-asset liquidity aggregation and distribution, while Visual Edge consolidates trading and risk information across platforms. Visual Edge also produces automated alerts and scheduled reports for dealing, risk, compliance and other operational teams.
For the client, the benefit is indirect but material. Earlier identification of an exposure can prevent a larger execution or balance issue. Consolidated data can shorten an investigation, while automated reporting can give the employee handling an escalation a clearer account of what occurred.
Tom Higgins, CEO of Gold-i,Tom Higgins, CEO of Gold-i, said AI is most useful when it equips an employee with accurate information rather than attempting to remove the employee from the interaction.
“I think clients will always want and should be given a real human to speak to. Where AI automation helps hugely is giving that human accurate and detailed information to help the client. The AI system can tell the human exactly why a process has failed and most often suggest the correct course of action to fix it.”
Gold-i’s technology shows how automation can improve service without being visible to the end client. Visual Edge uses automated alerts and consolidated reporting to help brokers identify risk across clients, books, instruments and trading platforms. The platform has also added Value at Risk, stress testing and Negative Balance Protection analysis, which FinanceFeeds examined in its coverage of Gold-i’s expansion of Visual Edge.
GCEX faces a similar operational challenge from the institutional-client side. Its clients expect liquidity, custody, collateral, settlement and trading technology to operate as a connected service. Automation can make settlement timing more predictable, improve visibility into counterparty exposure and reduce the manual work required to operate across asset classes and regulatory entities.
Holst said firms often begin with visible client-facing automation while leaving more difficult operational processes partly manual.
“The advantage will go to firms that automate the boring, high-volume, high-error work like reconciliation, reporting and monitoring. Then they use the time that frees up to put better people in front of clients, on the decisions that actually need judgement. Much of the industry is doing the reverse. It automates the client-facing layer first because it is visible and cheap and leaves the operational backbone semi-manual because fixing it is hard.”
That reversal can produce the appearance of efficiency without improving the underlying service. A chatbot may answer immediately, but the employee receiving an escalated case will still struggle if trade records, payment information and compliance data remain fragmented.
The stronger model automates both the routine client journey and the operational work required to support it. FinanceFeeds previously examined how broker technology is moving in this direction as onboarding and risk intelligence become competitive differentiators alongside execution and trading features.
Measuring Outcomes Rather Than Cost Reductions
The contributors also challenged the metrics commonly used to justify automation. Cost per contact, average handling time and the number of cases processed can show whether a system is cheaper or faster. They do not establish that the client’s problem was resolved.
A broker can reduce handling time by closing cases prematurely. It can increase automated containment by making it difficult to reach an employee, and it can report fewer contacts even if clients have abandoned a process in frustration.
Soutzis said firms should examine resolution, escalation and understanding rather than focusing only on operating expenses.
“Cost savings measure whether automation is cheaper to run, not whether it’s serving the client well. More useful measures look at whether an issue was actually resolved, not just closed faster; whether complaint and escalation rates moved after automation was introduced; and whether clients understood what happened and why. If automation is working, front-line staff should be spending more time on harder, judgement-based cases, not just covering more volume with fewer people. A broker that only tracks handling time and headcount will optimise for how many cases it gets through and miss whether trust in the service is holding up.”
Useful support measures include first-contact resolution, repeat-contact rates, successful transfers from automation to an employee, complaints arising from each journey and the number of cases reopened after closure. Those figures should be read alongside customer satisfaction rather than replaced by it.
Onboarding requires a different but related scorecard. Kirketerp-Møller said a high pass rate can be a poor measure because a compliance system is supposed to identify the right clients, not approve the largest possible number.
“Cost per file tells you what onboarding costs you. It tells you nothing about what it costs the client.”
“Start by discarding the metric the industry leans on hardest. A lot of KYC vendors quote high pass rates as proof of a good customer experience. But the purpose of a compliance solution is not to pass as many clients as possible, it is to make sure the right clients pass. A high pass rate is at least as likely to be a warning sign as a selling point. We deliberately don’t publish one.”
Muinmos says more than 97% of KYC and KYB journeys initiated on its platform are completed across every region. The figure is self-reported and measures whether the journey reaches an outcome, not whether the applicant passes the checks.
Kirketerp-Møller said completion must be assessed alongside contact, rework and decision times.
“Four measures that do say something:
Journey completion rate. Of the clients who begin onboarding, how many reach the end? Drop-off is where a bad experience actually shows up. Across all journeys initiated on our platform KYC and KYB, every region over 97% are completed. That means the journey is smooth, efficient, not burdensome – and all that without compromising compliance.
Touch count. How many times did you go back to the client for something you could have obtained yourself?
Rework rate. How many approved files were later reopened, escalated or corrected? Speed that produces rework isn’t speed. This is also the number that keeps completion rate honest: you can always finish more journeys by asking fewer questions, and rework is where that shows up.
Time to decision, measured from the client’s first click rather than from the moment the file reached an analyst’s desk.
And measure all four by client segment and risk band. Averages hide exactly the clients whose experience is worst.”
For institutional firms, exception rates and predictability may say more than the number of transactions processed. A single incorrect settlement instruction or misclassified risk event can outweigh savings produced across many routine cases.
Holst said falling enquiry volumes should not automatically be interpreted as better service.
“Cost savings is the easiest metric to report and the least useful one for judging client experience, because you can cut cost by making a process worse just as easily as better. What matters in an institutional context is whether automation is reducing ambiguity for the client, not just cost for the firm.”
“The questions worth asking are practical ones. Are settlement times more predictable? Is counterparty risk genuinely clearer? And when client queries drop, is that because the process answered them, or because nobody can reach a human anymore?”
“Exception rates matter more than volume processed. Institutional clients do not forgive automated mistakes the way retail clients might. A wrong settlement instruction or a miscategorised risk flag does more reputational damage than the efficiency gain was eve worth. If exceptions are not falling at the same rate volume is rising, you have not improved anything. You have just moved the same risk faster.”
AI Should Make Employees Better, Not Merely Fewer
The assumption that AI automatically lowers costs is also open to challenge. LLMs require integration, inference, data controls, monitoring and quality assurance. Regulated firms must additionally prevent sensitive information from entering inappropriate systems and verify answers used in consequential client processes.
ESMA’s survey provides broad securities-market context rather than FX and CFD broker-specific evidence, but its findings show why the cost case should not be assumed. While efficiency was the main driver of investment, 23% of surveyed firms expected no cost savings from generative AI. Investment firms also reported lower expected savings than several other financial sectors.
Aristidou said client outcomes should determine whether a deployment is worthwhile.
“The mistake many make is judging automation only on cost savings. Tools like LLMs are expensive to run at scale, and sometimes they cost more than they save, so the reason to use them should be better outcomes for clients, not the technology itself. The real question is: are clients getting better support? The answer is in satisfaction scores and in what clients tell you directly, not in how much you cut from operating costs.”
Higgins similarly said he measures automation through productivity, service quality and the time required to address an issue.
“I never measure automation by cost savings, but instead, look at the increase in productivity, the increase in service quality, the reduction in time to market and the time to fully address a client issue. With the support of AI, people can be 5-10 times more productive.”
His productivity figure is an estimate, but the underlying distinction is important. An employee who receives the relevant account, execution, liquidity and risk information in one place can resolve more complex cases without reducing the quality of the interaction.
FinanceFeeds previously explored the difference between useful machine learning and generative-AI enthusiasm with Higgins and other executives in Beyond the Buzz: Industry Experts Unpack AI’s Rise in Retail Trading. Newer systems are moving AI further into operational configuration, including tools that can translate AML policies into workflows, as seen in Sumsub’s use of AI agents for compliance environments. As capability grows, firms need more control over configuration, review and accountability.
Higgins warned that using automation primarily to remove employees from client-facing roles can produce longer-term damage.
“Some brokers and tech providers in our space have already started replacing humans with automated machines, and it has not been appreciated! Firms that go for cost-saving over quality improvement, in my view, will suffer long-term harm. Adding AI machines to a strong brokerage or tech firm greatly improves the whole operation and user experience.”
The firms that benefit most from automation may therefore be those that make it less visible. Instant processing should handle routine work, while connected operational systems should give employees the information required to investigate exceptions. When judgement, vulnerability or disagreement enters the case, the client should be able to reach someone with the authority to explain and review the result.
Automation earns trust when it removes work from the client without removing the broker’s responsibility. Once a client needs an explanation, the measure of efficiency is no longer how successfully the firm avoided human contact. It is how quickly automated processing gives way to informed accountability.