The digital transformation of investment banking client services

Based on observations made by emagine in 2025, Adverse Media Screening (AMS) has emerged as a key use case that is increasingly leveraging AI-based approaches to compliance software and systems design.

AMS is an essential process for identifying potential financial crime, regulatory and reputational risk associated with clients. When undertaking these checks, CIB client services teams must systematically search and monitor global news, social media and public records to determine if a prospective or existing client is or was ever linked to illicit activities such as money laundering, fraud or terrorism financing.

However, increasingly rapid news cycles and a diversified range of media channels for content distribution online means those teams are now challenged to process a potentially vast amount of unstructured data while performing a single AMS check on a single business or individual.

This article – the first in a series of three examining the digital transformation of legacy CIB client services operations – explores why and how institutions are set in 2026 to increasingly turn to AI tools as powerful allies in the effort to stay ahead of AMS-related regulatory compliance mandates.

The Inherent Risks of Manual Adverse Media Screening Processes

CIB Know Your Business (KYB) / Know Your Client (KYC) compliance teams already typically rely on various AI-adjacent tool sets to manage manual AMS processes. The features of these tools typically include:

Search Algorithms – Which help analysts efficiently search through vast amounts of data, finding relevant articles and mentions of clients. They may utilise techniques like those used in search engines, applying filters and keywords;
Natural Language Processing (NLP) – Allows machines to understand and interpret human language. It is used to analyse text from various sources, helping analysts identify potential risks by extracting relevant information from unstructured data;
Machine Learning (ML) – Recognises patterns and improve the accuracy of risk assessments over time. These algorithms learn from historical data to enhance compliance decision-making processes;
Entity Recognition – Focuses on identifying specific entities –such as individuals or companies – within text, assisting in the verification process and categorising information accordingly; and
Data Visualisation Tools – Which interpret and present findings from adverse media checks, enabling easier identification of trends or anomalies.

A single inaccurate MS assessment can raise questions regarding an institution’s ability to accurately assess risk

However, several drivers are increasingly challenging the current paradigm that the manual processes associated with the use of these tools are still valid.

The first driver is the exponential increase in the volume of media content, globally, that could be related to a single business or person. This surge is particularly evident when content present on social media platforms and other new forms of digital media are considered.

The second, related, driver is the growing enshrinement of KYB / KYC compliance checks within regulatory mandates that are extra-territorial or global-in-scope. Where once it was sufficient to only scrutinise source material relating to a business or person within one region, now it is industry best practice to ensure that all content, everywhere, is considered regardless of content source or language.

Both drivers are influenced by the recognition that financial crime knows no borders. Illicit activities can span multiple jurisdictions, making comprehensive due diligence essential. As regulations evolve, regulators now expect firms to use automation technology to maintain a holistic view of potential risks that encompass global news, market trends and social media discourse.

However, even when aided by AI-like solutions, the time-intensive nature of AMS initiated in accordance with layers of manual process presents several inherent risks for both clients and institutions. These risks include, but are not limited to:

Data Leakage – Manual searches on open-Web platforms can inadvertently expose sensitive investigative targets or internal search parameters;
Human Bias & Inconsistency – Manual processes are often non-standardised and rely on differing levels of human ability or experience. Requiring critical thinking analysis based on the interpretation of subjective data can threaten the credibility of any investigative effort if the results are inconsistent across the team;
Point-in-Time Snapshots – A significant risk is the lack of continuous monitoring. Manual AMS screening often means an institution is producing only a single point-in-time snapshot of a client’s risk profile. Once the snapshot becomes outdated by the publication of new content, there is often no trigger built into the process to adjust the risk assessment; and
Commercial & Reputational Risk – Any failure to comply, no matter how minor, creates significant reputational risk. A single inaccurate MS assessment can raise questions regarding an institution’s ability to accurately assess risk, potentially leading to a loss of trust or even the loss of an operating license.

How AI Mitigates These Risks: From Searching to Structured Analysis

Any approach to the design of AMS data analysis software capable of creating AI-like outputs can broadly be categorised across four categories in terms of its characteristics:

ML – Which gained prominence through its use of statistical algorithms to enable systems to learn from data and improve performance over time without explicit manual programming;
NLP – Which built on advances pioneered by ML to enable software to understand, interpret and generate human language;
Large Language Models (LLMs) – Which are trained on vast amounts of data to summarise and translate human language across multiple dialects and formats; and
Generative AI – Which typically uses LLMs to ‘learn’ from patterns in data to produce new outputs, such as summarised reports or risk justifications.

There are, in 2026, 25 third-party vendor solutions that incorporate some or all those approaches to AI software design that are already used by CIB client services teams to manage various elements of the overall KYB / KYC compliance workload.

Third-party vendor solutions that combine the use of ML, NLP, LLMs and Generative AI can significantly enhance the way client services teams conduct AMS. ML enables solutions to continuously learn from new data, allowing improvements in accuracy over time without needing extensive manual input. This adaptability helps analysts focus on more complex tasks, such as higher risk screenings which the model may not be mature enough for, rather than spending their time on repetitive, manual searches.

Likewise, NLP and LLMs provide features and benefits essential for processing and understanding vast amounts of unstructured data in multiple languages, enabling the software to analyse news articles, social media posts and other relevant information quickly and effectively. The use of Generative AI then further streamlines operations by producing summarised reports and justifications of potential risks based on the analysed data.

For compliance teams, the transition afforded by the combination of those approaches to software design not only enables a transition from manual searching to structured analysis, but it also addresses the inherent risks associated with the time-intensive nature of process elements that mandate human intervention. Specifically:

Data Leakage Protection – AI tools can use secure, encrypted searches and limit access to sensitive information, minimising the risk of exposure during the search process. Automated systems can also anonymise search parameters and target data, ensuring that investigative activities remain confidential;
Human Bias & Inconsistency – AI algorithms can standardise the screening process by applying consistent criteria for evaluating data, reducing the influence of individual biases. ML models can be trained on historical data to produce consistent outcomes and provide objective analyses, thereby enhancing accuracy and credibility;
Continuous / Real-time Monitoring – Unlike manual processes, AI can facilitate continuous monitoring by constantly scouring multiple data sources for updated information related to clients. Automated alerts can be set up to notify compliance teams when new adverse media is detected, ensuring real-time updates to a client's risk profile; and
Commercial & Reputational Risk – By improving the accuracy and efficiency of adverse media checks, AI can help institutions make more informed assessments, which reduce the risk of compliance failures. Enhanced data analysis techniques can uncover relationships and trends in data, providing deeper insights that help prevent reputational damage.

AI should be viewed as a longer-term ROI and saving strategy rather than an overnight cost-cutting measure.

Implementation Considerations

There are two considerations compliance teams must undertake before choosing to digitally transform the AMS components of long-standing KYB / KYC processes using solutions that combine several different approaches to AI-led analysis.

At the top of the list and, perhaps, the most-easily addressable question is the classic buy vs. build debate. While often the trickiest element associated with the introduction of new technology into any part of a CIB, the settlement of the argument of whether to use vendor-provided solutions to speed the uptake of the use of AI in AMS tends to hinge on the fact that commercial products often come pre-trained out-of-the-box on historical datasets well before they are ever deployed within a particular organisation. This pre-training allows the tools to mature rapidly, enabling analysis of complex data patterns from day one of implementation.

Next on the list is an institution’s assessment, pre-implementation, of the probability that a selected approach to the integration of an AI-centric solution – in-house built or vendor-built – into manual AMS processes will deliver return on investment. When proceeding with a commercial product selection process, a key consideration is arguably the ability of the application to realise time-save benefits across the most complex screening cases. For example, when dealing with intricate ownership structures or connected individual records, the efficiency of AI increases. In high-performance scenarios, an AI AMS tool can reduce a 10 to 20-minute manual review down to just two minutes. This reduction in manual overhead is also theoretically realizable for the screening of large corporate entities.

However, vendor selection processes must also consider the initial investment associated with commercial products, which is often much higher than in-house building due to licensing, integration and business user training. Therefore, AI should be viewed as a longer-term ROI and saving strategy rather than an overnight cost-cutting measure.

The Future of Compliance: Humans & Controls

The final and, arguably, most crucial consideration for CIB compliance teams seeking greater automation of their AMS processes is the human question. The answer relies not on whether analysts can or should maintain their roles over time; despite the impressive capabilities of AI, institutions tend to still see human expertise in KYB / KYC assessment as indispensable.

Rather, the conversation for or against AI implementation tends to hinge on whether an institution is ready to digitally evolve further elements of its compliance function once it succeeds in fully or partially automating AMS. Herein, the best practices regarding the use of AI in any field are relevant, namely:

Management of False Positives – Humans are required to review machine-generated alerts and request further information when an alert cannot be fully discounted;
High-touch Intervention – For high-risk exercises – for example, screening Politically Exposed Persons – the narrative switches toward a partnership where human analysts and automated toolkits work together for optimised results;
Spot / Point-in-Time Checks – Human operators should perform randomised reviews of AI-augmented systems to ensure an unbiased operation of the underlying code and its outputs; and
Model Refinement – Humans are responsible for refining AI models, ensuring compliance with regulatory and ethical standards, and providing the contextual judgment that machines cannot replicate.

Simply put, CIBs must maintain robust control through human oversight of the usage of AI in any part of the business, but particularly across the realms of regulatory compliance operations. The transition to AI-driven AMS would represent a fundamental shift in how an institution could manage the prevention of financial crime, but the speed and scale at which ‘mostly automated’ assessments of client data or unique scenarios can only be mediated by the contextual judgment and oversight of experience human professionals.

In this partnership, an AI tool functions much like the autopilot function on an airplane; it handles the bulk of the data processing and monitoring, but it escalates critical issues to its human KYC team member for further investigation whenever a nuanced or complex decision is required. This dual approach not only reduces operational costs but also provides a more robust defence against the ever-growing complexities of global financial crime.

Would you like to know more? Talk to emagine to learn more about how AI AMS solutions can optimise adverse media screening across your businesses and reduce risk overall.

The Inherent Risks of Manual Adverse Media Screening Processes

Search Algorithms – Which help analysts efficiently search through vast amounts of data, finding relevant articles and mentions of clients. They may utilise techniques like those used in search engines, applying filters and keywords;
Natural Language Processing (NLP) – Allows machines to understand and interpret human language. It is used to analyse text from various sources, helping analysts identify potential risks by extracting relevant information from unstructured data;
Machine Learning (ML) – Recognises patterns and improve the accuracy of risk assessments over time. These algorithms learn from historical data to enhance compliance decision-making processes;
Entity Recognition – Focuses on identifying specific entities –such as individuals or companies – within text, assisting in the verification process and categorising information accordingly; and
Data Visualisation Tools – Which interpret and present findings from adverse media checks, enabling easier identification of trends or anomalies.

A single inaccurate MS assessment can raise questions regarding an institution’s ability to accurately assess risk

However, several drivers are increasingly challenging the current paradigm that the manual processes associated with the use of these tools are still valid.

Data Leakage – Manual searches on open-Web platforms can inadvertently expose sensitive investigative targets or internal search parameters;
Human Bias & Inconsistency – Manual processes are often non-standardised and rely on differing levels of human ability or experience. Requiring critical thinking analysis based on the interpretation of subjective data can threaten the credibility of any investigative effort if the results are inconsistent across the team;
Point-in-Time Snapshots – A significant risk is the lack of continuous monitoring. Manual AMS screening often means an institution is producing only a single point-in-time snapshot of a client’s risk profile. Once the snapshot becomes outdated by the publication of new content, there is often no trigger built into the process to adjust the risk assessment; and
Commercial & Reputational Risk – Any failure to comply, no matter how minor, creates significant reputational risk. A single inaccurate MS assessment can raise questions regarding an institution’s ability to accurately assess risk, potentially leading to a loss of trust or even the loss of an operating license.

How AI Mitigates These Risks: From Searching to Structured Analysis

Any approach to the design of AMS data analysis software capable of creating AI-like outputs can broadly be categorised across four categories in terms of its characteristics:

ML – Which gained prominence through its use of statistical algorithms to enable systems to learn from data and improve performance over time without explicit manual programming;
NLP – Which built on advances pioneered by ML to enable software to understand, interpret and generate human language;
Large Language Models (LLMs) – Which are trained on vast amounts of data to summarise and translate human language across multiple dialects and formats; and
Generative AI – Which typically uses LLMs to ‘learn’ from patterns in data to produce new outputs, such as summarised reports or risk justifications.

Data Leakage Protection – AI tools can use secure, encrypted searches and limit access to sensitive information, minimising the risk of exposure during the search process. Automated systems can also anonymise search parameters and target data, ensuring that investigative activities remain confidential;
Human Bias & Inconsistency – AI algorithms can standardise the screening process by applying consistent criteria for evaluating data, reducing the influence of individual biases. ML models can be trained on historical data to produce consistent outcomes and provide objective analyses, thereby enhancing accuracy and credibility;
Continuous / Real-time Monitoring – Unlike manual processes, AI can facilitate continuous monitoring by constantly scouring multiple data sources for updated information related to clients. Automated alerts can be set up to notify compliance teams when new adverse media is detected, ensuring real-time updates to a client's risk profile; and
Commercial & Reputational Risk – By improving the accuracy and efficiency of adverse media checks, AI can help institutions make more informed assessments, which reduce the risk of compliance failures. Enhanced data analysis techniques can uncover relationships and trends in data, providing deeper insights that help prevent reputational damage.

AI should be viewed as a longer-term ROI and saving strategy rather than an overnight cost-cutting measure.

Implementation Considerations

The Future of Compliance: Humans & Controls

Management of False Positives – Humans are required to review machine-generated alerts and request further information when an alert cannot be fully discounted;
High-touch Intervention – For high-risk exercises – for example, screening Politically Exposed Persons – the narrative switches toward a partnership where human analysts and automated toolkits work together for optimised results;
Spot / Point-in-Time Checks – Human operators should perform randomised reviews of AI-augmented systems to ensure an unbiased operation of the underlying code and its outputs; and
Model Refinement – Humans are responsible for refining AI models, ensuring compliance with regulatory and ethical standards, and providing the contextual judgment that machines cannot replicate.

Would you like to know more? Talk to emagine to learn more about how AI AMS solutions can optimise adverse media screening across your businesses and reduce risk overall.