The first objection was where the data would live. In 2014, Christian Nentwich pitched representatives of 14 major banks on a cloud service for reconciliation: the painstaking business of checking that records agree. According to Duco’s later account, the room thought financial institutions would never send that information to the cloud. By 2020, eight members of the original group had become customers. It is a pleasing reversal for a company whose daily work consists of finding things that do not add up.
The London software business occupies an unglamorous but consequential corner of finance. A trade, a payment or an investment position can appear in several systems, wearing slightly different disguises. Someone must establish whether those records describe the same thing. When they disagree, someone must investigate. Duco sells tools to do that work with less manual handling, while keeping a record of how the answer was reached.
A spreadsheet can become a very expensive waiting room
Currencycloud offers a useful view of the problem. Before adopting Duco, the payments company tracked transfers across multiple spreadsheets. Investigating discrepancies on just one bank account could consume more than eight hours. Operations staff were doing detective work before they could get back to the business of moving money.
In Duco’s published customer account, Currencycloud’s team built its first reconciliation in two hours, with little training, then added another ten accounts within a week. Staff gained a consolidated view of exceptions and could inspect the underlying transactions. The company reported no financial impact from reconciliation breaks in the first year.
“Without Duco we would not have been able to scale as quickly.”Jon Burrell · Currencycloud
Those figures describe different activities: eight hours investigating an account is not a benchmark against two hours building a control. Their significance is organisational. A recurring investigation problem became a process the team could manage. Burrell estimated that, without Duco, Currencycloud would have needed five additional full-time employees.
Currencycloud’s reported implementation. A customer result, not a promised deployment time.
The product is a sequence, not a magic button
Duco’s proposition begins before matching. Incoming records need to be collected and made comparable. Its data preparation tools combine and transform data; reconciliation checks the resulting records; exception management gives unresolved differences somewhere to go. Users can assign, label and investigate them instead of circulating another attachment.
The platform accepts structured files and, through its document-processing capabilities, information extracted from PDFs, emails and images. Matching includes configurable tolerances and fuzzy matching, which can help distinguish a meaningful discrepancy from a difference in presentation. The business user defines what counts as agreement. That is a powerful responsibility, even when the interface requires no programming.
For an investment manager, the raw material might be cash, positions, allocations or transactions arriving from separate counterparties and systems. Duco’s asset-management offering addresses that fragmentation. A practical benefit is the ability to introduce or adjust a control when the business changes, rather than letting a new relationship accumulate a spreadsheet workaround.
This places Duco between the systems that generate financial activity and the people who must explain it. It is infrastructure for operations teams: a place to establish agreement, expose disagreement and pass usable information onwards. Its expertise combines data engineering with the details of financial workflows, where a plausible answer and a defensible answer are different things.
Getting past “we would never”
Duco’s account of its early rejection points to a broader change in buying habits. It argued that browser access, shared infrastructure, easier expansion and fewer installation projects made cloud software more attractive. The pandemic supplied another practical reason to value systems that colleagues could access from different locations. Those are the company’s explanations for the shift, rather than a transcript of eight separate purchasing decisions.
Founded by Nentwich and Michael Marconi, Duco had to survive long enough to benefit. In a FinTech Forum interview, Nentwich described the uncomfortable stage when the company had roughly 30-40 employees and a similar number of clients. Customers were already demanding enterprise service; the business lacked the resources to step back and plan.
“We probably under-raised and should have grabbed a bit more capital early on.”Christian Nentwich · FinTech Forum
The lesson he drew was specific. Pursuing the very largest institutions too early could exhaust a young company. Duco also expanded into New York while still small, seeking proximity to customers. There is a useful tension here: resist a buyer’s demand that you abandon the product’s premise, but take the cost of serving that buyer seriously.
$28 million to grow. Then a different kind of owner.
In January 2018, Duco raised $28 million from backers including Insight Venture Partners, Eight Roads Ventures, NEX Opportunities and former SunGard chief Cristóbal Conde. The announced uses were concrete: more staff in Europe and America, an Asian office and a broader product range. Selling software internationally requires people within reach when something goes wrong.
In July 2021, Nordic Capital announced an agreement to acquire a majority holding from existing investors. Nentwich retained a stake. The new owner’s stated agenda included international expansion, product investment and acquisitions. Nordic’s portfolio page now reports €40 million in revenue for 2025, giving a more useful scale reference than treating old fundraising totals as the size of the business.
Duco earns money through SaaS subscriptions and enterprise agreements, with professional services also available. A 2023 announcement offers a glimpse of the bigger deals: two global bank contracts exceeded $27 million in combined total contract value, excluding professional services. That figure covers the contracts, not one year’s revenue and not a standard price for the software.
For the financial year ended March 2023, Duco also reported 40 percent organic growth in annual contract value and 131 percent net revenue retention. The commercial logic is straightforward: win a control problem, then expand into more of the institution’s work. For customers, the relevant calculation includes staff time, implementation effort and systems they can retire, as well as the subscription.
The document joins the queue
Buying Metamaze in February 2024 expanded the territory. The acquired business specialised in intelligent document processing, tackling information that arrives in documents and messages rather than convenient rows. An insurance claim or loan document can contain useful data that somebody would otherwise have to type into another system.
Duco’s July 2024 release made adaptive IDP-enabled reconciliation available on its platform. The release notes describe supervised models trained on a customer’s document types and improved through human interaction. Extracted information can then join structured data in preparation and reconciliation. The point is to connect stages that otherwise require separate tools and manual handoffs.
The customer mix reflects that widening remit. CIBC Mellon’s strategic collaboration focuses on enterprise data automation; Phoenix Group announced a relationship covering reconciliation in asset management in December 2025. These are settings where the operational problem extends across teams and data sources. A narrow matching utility has an opportunity to become part of a larger operating process.
Duco’s own employment pitch emphasises autonomy, pragmatism and responsibility. Its UK benefits include flexible working, unlimited annual holiday and four volunteering days. The careers page also carries an employee’s acknowledgement of long hours and pressure. That combination feels recognisable for a company promising demanding institutions both speed and dependability.
The founder returns with agents on the agenda
Nentwich returned as chief executive in March 2026, succeeding Michael Chin in what Duco described as a planned transition. Chin moved into an advisory role. The announcement tied the change to the next phase of AI investment, putting a technically minded founder back in charge as the product’s ambitions expanded.
In May, Duco launched its Agentic Operations platform. Its approach gives agents access to defined operational tools through Model Context Protocol, including data preparation, reconciliation and exception handling. The announcement reported more than 250 clients, 20 billion transactions processed monthly and ten firms already using Duco agents in production. These are company-reported figures.
One early deployment result deserves a careful reading. Duco said building a reconciliation fell from two days to four hours, with about twenty minutes spent on agent runtime and the rest on human review. The review is part of the result. Fast generation only helps if somebody can examine what was generated and decide it belongs in production.
The less glamorous release notes are revealing too. June 2026 changes made hidden document annotations visible when they blocked new work and increased reconciliation selections to 1,000 records. September notes describe early access to a multi-reconciliation view. Even as agents attract attention, the product still advances through very human frustrations: finding the obstruction, seeing the history, handling more work in one action.
The useful test is the awkward record
Duco operates in a crowded market. Celent’s reconciliation research includes it alongside SmartStream, Gresham, FIS, Broadridge, Xceptor and others. Cloud delivery or an AI label alone cannot settle a comparison. Duco’s distinctive pitch combines business-user configuration, preparation, matching and document extraction within a common workflow.
The transferable idea is to start with a bounded, expensive problem and put the people who understand the data close to the controls. Test a real account, a troublesome document or a recurring exception. Measure the investigation burden as well as the match rate. Currencycloud’s account is persuasive because the first step was small enough to describe.
The same reasoning identifies the limits. If an organisation cannot approve cloud processing, cannot define acceptable matching behaviour or cannot spare knowledgeable reviewers, this operating model faces a constraint software alone will not remove. A permissive tolerance can make a dashboard look tidy while accepting the wrong agreement. The awkward record, and the explanation of what happened to it, remains the useful test.