Summary
Investment banks struggle to operationalize unstructured data trapped in PDFs, emails, legal agreements, loan tapes, and fragmented systems. The result is slower client response, manual reconciliation, and delayed decision-making. Closing the gap requires domain-aware data infrastructure that improves workflow automation, data access, governance, and enterprise AI readiness.
The processing of unstructured data in investment banking is time-consuming and resource-intensive. If banking data infrastructure can't adequately process unstructured data in a centralized, standardized data repository, individual business units are unable to grow their business.
In organizations as immense as global banks, a natural tension simmers between top-down technology decisions and the frontline execution needs of individual business units. Banks can close the unstructured data gap by taking practical approaches to enterprise data strategies, which help drive measurable returns for business unit stakeholders.
Top-down decisions drove most enterprise data infrastructure investment. Yet, this infrastructure must put the right answers in the hands of counterparties, sales traders, and relationship managers. In a recent article, we’ve proposed that every function within a bank should have confidence that the information they are working with is reliable and accurate.
About 90% of data users in banks reported that the data they need is often unavailable or takes too long to retrieve.i If banks have built data ecosystems oriented for front-office oversight and compliance while neglecting middle- and back-office operations, these teams pay the price.
Unstructured data is information that lacks a predefined data model or isn’t organized in a preset manner, making it difficult for traditional legacy systems to ingest and analyze. The proliferation of unstructured data has made manual management completely untenable. Data pipelines and models that can't make sense of unstructured data leave operational teams dependent on manual spreadsheet extraction. Eventually, many analysts and managers work with “learned helplessness” when it comes to deriving insights from their data.ii
Some global banks are spending $50 million or more annually for an enterprise data lake for prime brokerage that doesn’t work for them, while a specialized solution costing $15–20 million could save them over $30 million a year. The gap between enterprise infrastructure and middle- and back-office operations manifests most clearly in data latency, which prevents business units from being competitive.
You have a hot trade coming in from a buy-side client, and you have to answer immediately. Or a buy-side portfolio manager calls with a question, and your whole desk has to chase data across systems to collate an answer, maybe even doing a manual reconciliation if it’s a past trade. The “let me get back to you” refrain erodes trust and drains cycle time. These are also typically the most complex, highest-margin, most relationship-intensive client relationships. The follow-up window needs to be as brief as possible.
If a large asset manager, CFTC regulator, or board member asks for up-to-the-minute information, the answer needs to be immediate.
Answering the regulator:
Answering the client:
Answering the boss or the board:
If a bank’s buy-side client’s own data infrastructure is more advanced, they might already have independently generated the answer before their sell-side counterpart can confirm it. It’s even worse if the numbers don’t gel, and it takes the bank an eternity to generate it. If a buy-side counterparty has to build its own reconciliation and position-tracking layer, it’s like they’re subsidizing your lack of data capability.
A rates desk running vanilla swaps has a relatively clean data pipeline. But sell-side operations people encounter challenges with wrapper creation, trade capture, and reconciling trades, positions, and corporate actions at scale. A bank’s data infrastructure musttransform colossal volumes of unstructured data from a morass of documents into a single authoritative source for downstream reporting and analytics.
Common sources of unstructured data in investment banks include:
Different vendor sources tag information regarding geography, sector weightings, and Greeks inconsistently, essentially acting as unstructured data. In the private credit business, detailed attributes such as FICO scores, delinquency ratios, prepayment history, and borrower income must be extracted from messy loan tapes or agreements. Operations must manually map fields and reconcile naming inconsistencies.
This complexity slows down the onboarding of deals in terms of executing financing approvals and collateral eligibility checks. Risk teams may be working with stale risk calculations. Compliance teams may gain unwanted attention from regulators from poor lineage tracking or uncontrolled spreadsheets in calculating Basel capital levels.
Instead of building data infrastructure solely around the front-office data challenges using rigid enterprise-wide systems, banks should deploy flexible data infrastructures focusing on practical day-to-day operational needs. Everybody benefits from smooth data access, data workflow automation, and transparency across the organization. The financial services sector presents enormous challenges for data systems. Our industry is so complex and nuanced with regard to the density of nomenclature that only data platforms with domain expertise can sufficiently solve the unstructured data problem.
“While unstructured data is abundant in raw form, high-quality labeled data for supervised learning in finance is often scarce. Creating labeled datasets (e.g., news articles labeled with true market impact, or audio calls labeled with “positive/negative” tone) requires domain expertise and can lag behind the pace of model development. Unlike domains such as computer vision, financial data labels can be ambiguous or hard to define.” - Duane, Morgan, and Carter, Indiana State Universityiii
Recently, deep learning has become exceptionally adept at processing high-dimensional data like text, audio, and images. Available techniques include natural language processing (NLP) for parsing financial text, speech and audio processing for spoken communications, computer vision for financial documents, and multimodal approaches that combine heterogeneous sources.iv
Domain-aware data platforms allow all business units to benefit from a consolidated view of portfolios and positions, real-time accuracy, and single-query access across structured and unstructured sources. In this way, institutions can give high-revenue business units the specialized tools they need to grow.
Large banks have invested heavily in data ecosystems, but most have not solved the unstructured data problem. The business units paying the price know they can generate more revenue with improved access to unstructured financial data. The cream of the banking crop will move to seal the gap between enterprise data infrastructure and front-line execution capability. Enterprise data infrastructure isn’t purpose-built for the high-speed, high-volume requirements of modern sell-side business, like financing equities on swaps, managing structured products, or trading millions of contracts daily. Flaws in foundational enterprise IT systems disrupt business-as-usual operations on a weekly basis for 58% of financial services executives.v
On the buy side, large hedge funds, multi-strategy platforms, or firms with a lot of vehicles shouldn’t have to factor the shortcomings of their sell-side counterparties into their own planning. It’s a great way to make clients start courting other sell-side counterparties that can reduce that burden. The edge goes to banks that can deliver clean, consolidated, real-time data directly into the buy side’s workflow.
Ted O’Connor
Ted is a Senior Vice President focused on Business Development at Arcesium. In this role, Ted works with leading financial institutions in the capital markets to optimize data, technology, and operational needs.
Sources:
i Deloitte, October 30, 2025. https://www.deloitte.com/us/en/insights/industry/financial-services/financial-services-industry-outlooks/banking-industry-outlook.html
ii HBR, 2012. https://hbr.org/2012/06/learned-helplessness-in-organi
iii Duane, J., Morgan, A., & Carter, E. (2025). A review of financial data analysis techniques for unstructured data in the deep learning era: Methods, challenges, and applications. https://osf.io/download/gdvbj
iv Duane, J., Morgan, A., & Carter, E. (2025). A review of financial data analysis techniques for unstructured data in the deep learning era: Methods, challenges, and applications. https://osf.io/download/gdvbj
v KPMG, 2024. https://kpmg.com/xx/en/our-insights/transformation/kpmg-global-tech-report-2024/financial-services.html
No spam. Just the latest releases and tips, interesting articles, and exclusive interviews in your inbox every week.
