- Nexus turns financial statements, ESG disclosures, identity files and other untidy documents into structured data with an audit trail.
- Its customers are banks, asset managers, insurers, governments and other organisations where a plausible answer is not enough.
- The unusual move is commercial: its Capacity Unit model prices completed, auditable work rather than access to another software licence.
- The useful lesson is to automate a bounded workflow, preserve the evidence and name the human who approves the result.
A bank can forgive a chatbot for suggesting the wrong restaurant. It cannot shrug when the same sort of machine misreads a borrower’s liabilities. This is the awkward little fact sitting beneath the enterprise AI boom: the most impressive answer in the room may still be unusable if nobody can show where it came from, reproduce it or accept responsibility for it.
Nexus FrontierTech has made a business from that awkwardness. Founded in 2015 and headquartered in London, with teams across Tokyo, Singapore and Hanoi, the company builds AI for the least photogenic parts of finance: spreading statements, parsing ESG disclosures, checking identity documents, assessing risk and moving data from paper-shaped chaos into the systems where work actually happens.
There are no charming robot mascots here. The raw material is a mortgage form, a bond prospectus or a sustainability report with a table stranded on page 137. The promise is not that the machine will have an opinion. The promise is that it will do the repetitive reading quickly, show its work and hand the consequential judgment to a person.
The first thing to fail was the demo
This sounds backward because demos usually succeed. The model extracts a field, answers a question and produces a neat dashboard. Then the pilot reaches a risk committee. Where did the number come from? Which version of the document supplied it? What happens when the format changes? Who is accountable if it is wrong?
The cleverness survives. The deployment does not. Nexus calls this the accountability gap. Its answer is “machine-first, human-final”: automation performs the high-volume preparation, deterministic controls hold the centre, evidence stays attached and a named professional signs off on the work that matters. It is less cinematic than autonomy. It is also much closer to how a bank is allowed to operate.
“Executives don’t need more tools; they need outcomes.”Danny Goh, founder and CEO
The machinery underneath has changed names as the company matured. Podder was Nexus’s platform for configuring, fine-tuning and deploying reusable models. Financial Services Extraction packaged more than 50 modules for statements, IDs and other documents, later distributed through Microsoft Azure Marketplace and Temenos Exchange. OneNexus Capacity Cloud is the current proposition: a controlled execution layer that converts a client’s procedures and rulebooks into observable work.
From selling the machine to selling the shift
The more interesting change happened on the invoice. In January 2026, Nexus introduced the Nexus Capacity Unit, or NCU. Instead of asking a bank to buy seats and hope employees create value with them, Nexus offers a standard unit of completed, auditable work. A client subscribes to baseline capacity, then pays for more by task, by role-equivalent capacity or against a business measure such as loan volume or assets under management.
That is a meaningful transfer of risk. A conventional software vendor gets paid when access is provisioned. Under an outcome model, Nexus has more reason to care whether a credit file is actually processed to the agreed standard. It is the difference between selling a dishwasher and selling clean plates.
One published deployment, in basis points
The company also says its credit agents can handle repetitive spreading and verification in under ten minutes, compared with one to two hours for a human analyst. Those are company-reported figures, not universal laws. Their value is in making the promise falsifiable. “AI transformation” is fog. Ten minutes, a complete audit trail and an agreed error threshold can be tested.
The wedge is paperwork, not prophecy
Nexus sits between several familiar categories. It looks like an intelligent-document-processing vendor when it reads financial statements; a regtech firm when it builds KYC and client-risk workflows; an AI platform when it deploys models in private clouds or on premises; and a services firm when engineers adapt the system to a client’s rules. That overlap is deliberate. Banks rarely need a naked model. They need the integration, controls, process redesign and support wrapped around it.
The APS Asset Management collaboration shows the shape of the work. ESG evidence across Mainland China, Hong Kong, Macau, Taiwan and Singapore was scattered, inconsistent and difficult to compare. Together, the firms built ANAFES, a system that extracts disclosure data and maps it into an investor’s own framework rather than forcing every question through a generic score. The project received proof-of-concept support from the Monetary Authority of Singapore. Three model groups averaged 80% accuracy in the published technical results - useful enough to reorganise research, not magical enough to eliminate judgment.
The customer list is broader than finance alone. Nexus and gap personnel co-developed a candidate-registration system that the recruitment firm says reduced administration, accelerated onboarding and cut applicant drop-off. Kaizen Compliance Solutions worked with Nexus on client-risk assessment. Azure and Temenos provide distribution into existing technology estates. The common denominator is not an industry label. It is a queue of documents, expensive specialists and a rulebook that cannot be ignored.
The part worth copying
Most companies cannot copy Nexus’s model library, nor should they try. They can copy the sequence. Begin with one costly, repetitive decision process. Define the unit of finished work before choosing the model. Attach every extracted fact to its source. Let experts correct the system. Measure cycle time, exception rate and cost per completed case. Only then expand.
A five-part field guide
- Pick a workflow with high volume, clear rules and an expensive backlog.
- Write down what “finished” means in operational, not technical, language.
- Keep the source evidence beside every machine-produced field.
- Send uncertainty and exceptions to a credentialed human.
- Price and measure the outcome - not logins, prompts or model prestige.
The conditions matter. This approach is strongest where the input repeats, the desired output can be specified and errors can be caught before they escape. It is weaker when source data is inaccessible, decisions are fundamentally ambiguous, case volume is too small to justify integration, or nobody inside the customer owns the final judgment. A traceable system also cannot rescue a bad policy. It will merely execute the bad policy more neatly.
That restraint is the point. Nexus does not need the machine to be the boss. It needs the machine to make the boss faster, better informed and able to defend the answer later.
A decade, reduced to a unit of work
Nexus begins by connecting AI research to enterprise workflows.
The AI Factory formalises a route from process audit to pilot and deployment.
ANAFES receives MAS support; Nexus joins Tech Nation Fintech 5.0 and launches on Azure.
More than 50 extraction modules reach the Temenos banking ecosystem.
ISO 27001, a Tokyo entity and work delivered across more than 20 countries.
The Capacity Unit turns finished, auditable work into the product.
The company’s arc is a quiet correction to the way technology is usually sold. First came expertise. Then a platform. Then reusable modules. Finally, a unit that the operations department can count. The progression suggests that the hard part of enterprise AI was never getting a model to produce something. It was turning that something into work the institution recognises, budgets for and accepts.
A signature seems almost comically old-fashioned beside a generative model. Yet that may be why it matters. In finance, the future arrives only after someone is prepared to put their name at the bottom.