Human trafficking is often described as a hidden crime, yet it leaves a trail that is anything but invisible: a trail of money.

Behind the coercion and the exploitation sits a business model, and like any business it has to collect revenue, pay costs and move profits. Those profits are vast. The International Labour Organization estimates that forced labour in the private economy now generates $236B in illegal profits every year, a figure that has risen by more than a third in a decade [1]. 

Almost every pound and dollar of it passes through the financial system at some point. That makes banks an unlikely but powerful frontline in the fight against modern slavery, and it makes financial intelligence one of the most promising tools we have to detect it. 

This paper looks at the scale of the problem, at how financial crime enables trafficking, and at how a combination of artificial intelligence and skilled human judgement can help institutions find it.

The scale of a hidden economy

The headline numbers are difficult to absorb. On any given day in 2021, almost 50 million people were living in modern slavery, including 27.6 million in forced labour, according to the Global Estimates of Modern Slavery produced by the ILO, Walk Free and the International Organization for Migration [2]. 

The crime is also growing more profitable. The ILO's 2024 report, Profits and Poverty, put the annual illegal profits from forced labour at $236B, an increase of $64B, or 37%, since 2014, equivalent to roughly $10,000 wrung from each victim [1]. 

Sexual exploitation accounts for nearly three-quarters of those profits despite involving a smaller share of victims, and the largest regional total sits in Europe and Central Asia at $84B [1].

Detection is rising too, which reflects both a worsening problem and better awareness. The United Nations Office on Drugs and Crime (UNODC) found that the number of trafficking victims detected globally in 2022 was 25% higher than in pre-pandemic 2019. Victims trafficked for forced labour rose by 47%, children now make up 38% of those detected, and trafficking for forced criminality, including the online scam compounds that have spread across South-East Asia, climbed from 1% of detected victims in 2016 to 8% in 2022 [3]. 

The financial scale of that trend is striking: the United States Department of State estimates that forced-labour scam operations generated between $25B and $64B globally in 2023, and defrauded US citizens alone of around $10B in 2024, prompting the first use of Global Magnitsky sanctions against officials complicit in forced labour in online scams [4]. These are only the cases that came to light; the true totals are certainly higher.

A profit-driven crime, and why money is its weak point

Trafficking persists because it pays. Unlike contraband that is sold once, a trafficked person can be exploited repeatedly, generating revenue over months or years. That revenue has to be collected, whether in cash from a brothel or car wash, in wages skimmed from an exploited worker, or in digital payments through an online platform. It then has to be moved, pooled and laundered before it can be spent or reinvested. Each of those steps requires the financial system, and that is precisely where the crime becomes visible.

This is why following the money has become so important. Victims are frequently too frightened, traumatised or controlled to come forward, and prosecutions that rely solely on their testimony are hard to present. Financial investigation offers a different route. The accounts, transfers and businesses that surround a trafficking operation can reveal the network, identify the controllers, and corroborate a case even when a victim cannot or will not testify. The same financial plumbing that enables the crime is the channel through which it can be disrupted.

The laundering itself follows familiar stages. Proceeds are placed into the system, often as structured cash kept beneath reporting thresholds, before being layered through transfers between accounts, money service businesses, shell companies, and, increasingly, virtual assets. Finally, they are integrated back into the legitimate economy through property, vehicles, and cash-intensive businesses. 

Each channel, from remittance corridors to online payment platforms to crypto exchanges, carries its own indicators, which is why an effective detection programme has to look across products rather than at any one in isolation [5] [6].

A trafficked person can be exploited repeatedly. That makes trafficking a business, and every business leaves a financial trail.

What trafficking looks like in the data

Financial regulators and bodies such as the FATF and FinCEN have published detailed red-flag indicators that give trafficking a recognisable financial signature (FATF, 2018; FinCEN guidance). No single indicator proves anything, but in combination they form a pattern. Common signs include:

•  Funnel activity: cash deposited into an account in many different locations and quickly withdrawn or transferred elsewhere, often inconsistent with the customer's stated profile.

• Third-party control: one individual appearing to operate or benefit from numerous accounts, or wages paid to workers that are immediately moved back to a controlling party.

•  Clusters of accounts that share an address, phone number, email or device, or that are opened together in a short window.

• Cash-intensive front businesses, such as nail bars, massage parlours, car washes, agriculture or hospitality, whose revenue is inconsistent with their size or location.

• Payments to online classified or adult-services platforms, recruitment-fee patterns, and frequent small transfers to higher-risk source countries.

• Account holders whose spending shows little or no normal cost of living, suggesting their finances are controlled by someone else.

The footprint also varies by type of exploitation. Sexual exploitation tends to surface in card payments to adult-services and classified sites, late-night cash deposits near known venues, and many small transfers. Labour trafficking shows up differently, in wage accounts opened in bulk by an employer or gangmaster, identical payroll payments routed straight back to a single controlling party, and recruitment debts deducted from pay. Recognising these distinct profiles helps tune detection to the actual risk.

Crucially, no single bank sees the whole picture. A controller will often spread accounts across several institutions precisely so that each sees only a fragment of the activity. That fragmentation defeats siloed monitoring, and it is a central reason why both advanced analytics and information-sharing partnerships matter so much.

Individually these behaviours are unremarkable. The challenge, and the opportunity, is that the signal of trafficking lives in the relationships between transactions, accounts, and people, not in any single payment.

Consider how that looks in practice. A modest nail bar deposits irregular cash across branches in three different cities. The same mobile number is attached to nine personal accounts opened within a fortnight. Each receives an identical monthly payroll credit that is withdrawn in full the next day at a single cluster of cash machines, and none of the account holders spends anything on rent, food, or transport. Alone, not one of these facts is damning. Taken together they describe a controlled workforce, and they are precisely the kind of pattern that network analytics can assemble and a human investigator can act on.

Why rules alone fall short, and where AI helps

Traditional transaction monitoring relies on fixed rules and thresholds. Against trafficking it struggles on two fronts at once: it misses cases because the indicators are weak, contextual, and spread across many accounts and even many institutions, and it overwhelms analysts with false positives on the crude rules that do fire. The result is alert fatigue on one side and undetected exploitation on the other. This is the gap that artificial intelligence is well suited to close.

Applied responsibly, AI strengthens both screening and monitoring in several ways:

• Entity resolution and network analytics link accounts, people, and businesses that share hidden identifiers, exposing the clusters and controllers that rules-based systems treat as unrelated.

• Behavioural and anomaly detection learns what normal looks like for a given customer and flags the funnel patterns, structuring, and sudden changes that signal control.

• Typology models trained on confirmed trafficking cases recognise combinations of weak signals that a human could not feasibly tie together at scale.

• Natural-language processing screens adverse media, court records, and unstructured data to enrich name and sanctions screening with real-world context.

• Risk-based prioritisation ranks alerts by genuine likelihood, so analysts spend their time on the cases most likely to involve real harm.

Screening deserves particular attention. Traffickers and their facilitators show up on sanctions and watch lists, in adverse media and in law-enforcement alerts, but legacy name-matching drowns analysts in false matches, especially across different scripts and transliterations. 

AI-driven screening that understands names, context, and relationships raises genuine detection while cutting false positives, and when it is paired with perpetual KYC it keeps the customer picture current, rather than frozen at onboarding. Regulators increasingly expect institutions to build human-trafficking indicators into their AML frameworks, so this is fast becoming a supervisory expectation as much as a moral one.

The combined effect is to detect more genuine cases, surface whole networks rather than isolated alerts, and do so while reducing the noise that buries investigators. AI does not replace the bank's controls; it makes them see in colour rather than black and white.

The indispensable human

If AI provides scale and pattern recognition, people provide judgement and care, and in this domain both are essential. AI surfaces possibilities; humans decide what they mean and what to do about them. 

The human role is non-negotiable for many reasons:

• Context and proportionality. A trafficking indicator may point to a perpetrator, but it may equally point to a victim whose account is being controlled. A purely mechanical response, such as exiting the customer or freezing an account, can cut a victim off from their only resources and push them deeper into harm. Trained analysts apply a victim-centred lens that automated systems cannot.

• Quality of intelligence. The output that matters is not an alert but a suspicious activity or transaction report that is timely, accurate and genuinely useful to law enforcement. Skilled humans turn a model's signal into the narrative that helps investigators identify a network and, ultimately, reach victims.

• Collaboration and learning. Some of the most effective work happens through public-private partnerships, such as Project Protect in Canada, the Joint Money Laundering Intelligence Taskforce in the UK and the Finance Against Slavery and Trafficking initiative internationally, where banks, regulators, law enforcement, and survivors share typologies and feedback. Survivor insight in particular sharpens the indicators that models are trained on. 

The right operating model is human-in-the-loop by design: AI to find the signal at scale, expert analysts to interpret it with judgement, and explainable systems so that every decision affecting a real person can be understood and defended.

Real-world impact

The $236B that flows from forced labour each year is proof that traffickers treat human beings as a business. 

The encouraging corollary is that businesses can be disrupted by following their money. Banks sit at a chokepoint that victim-led investigation alone can rarely reach, and the tools to act on that position now exist. AI brings the scale, the network view, and the pattern recognition that fixed rules never could. Human analysts bring the context, the care and the high-quality intelligence that turn detection into protection. Partnership between institutions, regulators, law enforcement and survivors connects the two. 

United, they can turn routine compliance monitoring into something far more meaningful: a genuine instrument for finding the hidden, and helping to free them.