Markets Technology Awards 2026: March of the riskbots

Will a new wave of AI-powered assistants live up to expectations?

Will a new wave of AI-powered assistants live up to expectations?

Many of us have become accustomed to using artificial intelligence assistants at work, with tools such as ChatGPT, Claude, Gemini and Microsoft’s Copilot now common on our workplace desktops. They are generic tools – useful for basic writing and coding tasks, finding information or summarising blocks of text.

Their next evolutionary step might be to specialise – taking on a wide range of tasks, but within a specific domain. In risk management, some technology vendors are taking this step, rolling out assistants that promise to help users manipulate data, analyse information, explain patterns and visualise complex information within an existing risk system.

But not all vendors. In this year’s Markets Technology Awards, a trio of firms trumpeted their gleaming, eager-beaver assistants; many others did not. And those that have taken the plunge are doing it in slightly different ways, and emphasising different opportunities and challenges.

There is still some scepticism within the industry over what these new AI assistants can actually achieve

Opensee built its application using Amazon’s Bedrock large language models (LLMs); SS&C Algorithmics uses Anthropic’s Claude; ActiveViam is working with the Mistral LLM. All three, though, note they are not tied to these models – they can be swapped for others.

The assistants promise to save time and money for risk managers – and open the door to other users without a risk background – enabling them to ask in plain English (or German, Japanese or any other language) for information that would otherwise take multiple steps or complicated Structured Query Language queries to access. They can also offer new insights and ways to slice data that has sat on financial services firms’ servers for years.

But there is still some scepticism within the industry over what these new AI assistants can actually achieve. Some argue that development time is better spent elsewhere – on narrowly focused AI tools that are designed to tackle defined operational challenges, rather than jack-of-all-trades assistants.

The generative AI (GenAI) leap

Technology vendors have been experimenting with natural language processing (NLP) for the past decade, but the widespread availability of commercial LLMs has turbocharged that process.

Stephane Rio, chief executive of Opensee, says the company spent many years investigating NLP as a way to soup up the firm’s data quality, aggregation, analytics and reporting product.

“Then GenAI arrived, and it simplified our lives because it’s obviously much more intelligent than what you could build yourself. We could leap straight away to a very powerful tool,” he says.

The company’s AI assistant, Agensee, took a five-person team about 18 months to develop, with a proprietary semantic layer on top of off-the-shelf Amazon LLMs.

Rio says these types of AI assistants can be particularly useful for senior managers who want to gain access to complex calculations quickly, but that many day-to-day risk managers also want to use this type of tool.

“The number of people who don’t want to get their hands dirty goes far beyond the senior managers,” adds Rio. But this claim still needs to be tested; currently, Agensee is being used by a handful of power users at its clients. “People are still playing with it and trying to figure out where it’s going to be useful.”

So far, the company has seen more interest from hedge funds than large banks because banks have much bigger compliance hoops to jump through.

“I’ve had many conversations with banks. Everyone is super excited, but they had a lot of blockage internally,” says Rio. “The constraints you find in banks from a regulatory, governance and security perspective are just bigger. So it takes longer.”

A network of agents

While off-the-shelf LLMs may have given risk tech vendors a leg up in their attempts to build new AI assistants, substantial behind-the-scenes engineering and integration work is still required before they can do the job.

SS&C Algorithmics has spent the past 18 months developing its AI assistant, known as Algorithmics Risk Intelligence Assistant (Aria), which aims to augment, rather than replace, human judgement.

Curt Burmeister, chief technology officer and co-head of the company, says: “I don’t think it’s going to be completely unique forever. But, right now, this whole idea of having an agent that is like a personal system, it’s like somebody who is helping you do your job better.”

At Aria’s core is a network of specialised ‘agents’ or digital experts, each trained for a distinct task such as financial reporting, portfolio analysis or interpreting market news. Overseeing them is a co-ordinating chatbot that assigns work to the relevant agent, bringing their insights together into a single, coherent response.

Each agent is powered by Anthropic’s Claude LLM (although – again – this could be switched out), underpinned by the SS&C Algorithmics semantic layer, which provides structure and context to the system’s responses.

We don’t actually pass any client data to the LLM. We only pass metadata

Curt Burmeister, SS&C Algorithmics

Burmeister’s team spent around a year developing the semantic layer – giving the model detailed instructions on what information to provide, how it should be presented and what assumptions to make when a request is ambiguous.

For instance, if a user asks: ‘What is the risk to my portfolio?’, then the semantic layer will fill in the gaps, making decisions about the relevant time period, the best way to visualise it and whether to show exposure, value at risk or expected shortfall, for example.

“That’s where it gets really interesting and challenging”, says Burmeister.

The team went through thousands of iterations of tests for each of its AI agents during this process, he says. It took around six months to make the agent “pretty good”, and six more were required for another step-change in quality.

The tasks handled by the AI agents could all be carried out manually through the company’s user interface but, by automating these steps, the agents reduce the number of clicks and screens needed, streamlining work that would otherwise be fragmented across multiple tabs.

Clients have shown a lot of interest in Aria, Burmeister says, but uptake has been limited. This is partly because of caution around AI guard-rails and concern from clients around what data can be fed into the LLM.

“Everybody understands the power [of it], the adoption is a bit slower than we had expected, but I think it’s going to start to pick up,” says Burmeister. “We spent quite a bit of time working on that aspect to help increase adoption. So, for example, today, we don’t actually pass any client data to the LLM. We only pass metadata. We might pass portfolio names, but not the portfolio holdings.”

Open architecture

At ActiveViam, Chris Horril, director of product marketing, says his company has had similar issues with getting clients comfortable with the new AI-powered chat function on its Atoti platform. In response, the company is stressing that clients can customise it themselves.

“A big aspect of everything we are doing with Atoti Intelligence is to make the framework an open framework. From our research and the conversations we have had with our customers, that is pretty much the top concern,” says Horril.

The chat interface is underpinned by Mistral LLM, and the firm stresses that clients control their LLM, manage data lineage and model versions, ensuring data provenance and compliance. This allows companies to align Atoti AI with their own AI compliance and regulatory rules.

Horril says: “Customers are very worried about AI agents leaking information or hallucinating, so they go through this exhaustive internal process to ensure that any AI that they bring into the organisation is trustworthy.”

Using the chat interface, clients can access the more specialised ‘auto explain’ tool, which enables them to understand why different metrics, such as VaR or profit and loss, are moving in certain ways – a usually time-consuming and involved process. So far, this kind of application is attracting most interest from ActiveViam clients.

Horril says: “When we explain [auto explain] it really captures their imagination. It is saving time for people who are probably on quite high pay grades. I think these AI-powered tools solve harder and more time-consuming problems.”

Not sexy, but practical

Many companies are not yet adding AI assistants into their technology offerings. Other firms are taking a more cautious approach and are sceptical about the technology being useful to clients.

“It’s the buzzword at this point”, says Daniel Finn, head of risk solutions for North America at Conning. “We’ve seen a lot of our competitors rebrand old crap they were doing with an AI badge,” he says.

Conning has decided to stick to its knitting – for now, at least – by keeping its models stable and trusting clients understand the firm has a high-quality product. It will add AI components if and when it finds something that is genuinely useful.

“As with everybody, we are looking at AI. But we have taken a conscious approach to not use it as a crutch. We’re not going to chase things,” says Finn. “What do our clients want? What are we doing that’s hard? And then saying, ‘well, is AI the right approach? Or is there something else out there that is maybe not as sexy, but that gets the job done?”

Other tech vendors believe more targeted tools – not AI butlers – are a better bet. These sorts of tools set out to solve one specific problem, making it easier to demonstrate a saving of time or money.

Ashish Doshi, senior investment consultant at Ortec Finance, says the firm has spent the past four years developing a machine learning tool that enables clients to rejig their portfolios using simulated data – and learns which portfolios offer the best risk/return trade-offs under user-defined constraints. The tool will be rolled out to insurance clients by the end of the year and other institutional clients next year.

“It’s looking for a problem that cannot be solved through traditional tools,” says Doshi. “As a provider we need to be able to demonstrate the benefits of using such tools. That’s why we like using AI as a supplementary tool to existing solutions rather than as a replacement, so the benefit is quite clear.”

Risk Markets Technology Awards 2026: roll of honour

Traded risk technology

Counterparty risk product of the year: SS&C Algorithmics

Electronic trading support product of the year: TransFICC

Market liquidity risk product of the year: Bloomberg

Market risk management product of the year: SS&C Algorithmics

Market scenario generator of the year: Conning

XVA calculation product of the year: SS&C Algorithmics

Front-office regulation

FRTB (SA) product of the year: Bloomberg

Regulatory reporting product of the year: Nasdaq

Buy-side technology

Best execution product of the year: Abel Noser Solutions, a Trading Technologies company

Buy-side ALM product of the year: Ortec Finance

Buy-side market risk management product of the year: Orchestrade

E/OMS provider of the year: Trading Technologies

Back-office

Clearing house support product of the year: Nasdaq

Collateral management and optimisation product of the year: CloudMargin

Clearing member support product of the year: Nasdaq

Pricing/trading technology

Pricing and analytics: commodities: Orchestrade

Pricing and analytics: cross-asset and structured: Murex

Pricing and analytics: digital assets: Talos

Pricing and analytics: fixed income: Quantifi

Data and other specialist categories

Alternative data vendor of the year: Cardo AI

Best vendor for system support and implementation: Murex

Market data vendor of the year: 360T

Risk data repository and data management product of the year: Opensee

Innovation

Best use of cloud: ActiveViam

Best of use of machine learning/AI: ActiveViam

Best user interface innovation: J.P. Morgan

Methodology

Technology vendors were invited to pitch in 31 categories by answering a standard set of questions with a maximum word count. More than 150 submissions were received and 103 firms shortlisted.

A panel of nine industry experts and Risk.net editorial staff reviewed the entries, with judges recusing themselves from categories in which they had a conflict of interest or no direct experience.

The judges individually scored and commented on the shortlisted entrants before meeting in October to review the scores and – after discussion – decide the winners.

In all, 26 awards were granted this year. Awards were not granted if a category had not attracted sufficent entrants or if the judging panel was not convinced by any of the pitches.

The judges

Dipak Chotai, founder, JD Risk Solutions

Sidhartha Dash, chief researcher, Chartis Research

Vishnupriya S Devarajulu, software engineer, American Express

Jenny Knott, founder, FinTech Strategic Advisors

Becky Pritchard, contributor, Risk.net

Peter Quell, head of portfolio analytics for market and credit risk, DZ Bank

Navin Sharma, head of risk and portfolio management partnerships, Fitch Ratings

Jagat Singh, director of software engineering (risk and pricing), Ice Clear Credit

Duncan Wood, global editorial director, Risk.net