Institutional trading operates in an environment where small delays can have significant financial consequences. Newly appointed Just2Trade CMO Vsevolod Smirnov believes machine learning will increasingly determine how effectively institutions manage those delays.
“Institutional trading systems are extremely sensitive,” Smirnov said. “Friction can appear when firms access data, interpret liquidity, route orders, or execute large positions without revealing too much of their intention to the market.”
For decades, institutions have worked to reduce these inefficiencies through faster infrastructure, better algorithms, and closer access to liquidity venues.
Machine learning could now change the way these systems make decisions.
“The important change is not simply that systems will become faster,” Smirnov explained. “They will become better at deciding when speed is useful and when waiting produces a better result.”
Just2Trade is a global brokerage brand operating since 2007, providing access to more than 128,000 trading instruments to clients in over 130 countries. Its services include solutions for retail traders, professional market participants, introducing brokers, and institutional clients.
Third-party trading systems can also connect through the company’s Open API, allowing professional users to integrate external software with the trading environment.
Smirnov believes the next phase of institutional automation will involve machine-learning models playing a more direct role in execution decisions.
“Traditional algorithms follow instructions,” he said. “Machine-learning systems can evaluate the environment in which those instructions are being carried out.”
Execution Is a Decision, Not Just a Speed
One of the most important applications of machine learning is optimal execution modelling.
Large investors rarely send an entire institutional order to the market at once. Doing so can move the price of the asset, reduce available liquidity, and reveal the institution’s intentions to other participants.
Instead, orders are commonly divided into smaller pieces and executed over time.
The difficulty lies in determining when, where, and how aggressively each part should be placed.
A machine-learning model can analyze factors such as:
- bid-ask spreads;
- order-book depth;
- available liquidity;
- short-term volatility;
- recent execution quality;
- market impact;
- correlated asset movements;
- time of day;
- news and macroeconomic events.
“All the old assumptions about execution are being challenged,” Smirnov said. “A fixed rule may work under normal market conditions and fail completely when liquidity changes.”
Instead of relying only on static instructions, advanced systems can continuously reassess whether an order should be placed immediately, delayed, redirected, or divided differently.
This does not eliminate execution risk, but it can make the response to that risk more precise.
“The goal is not to predict every market movement,” Smirnov added. “It is to improve the probability of completing an order at an acceptable cost.”
This distinction matters. Claims that AI can reliably forecast markets are difficult to support, while improving execution quality is a narrower and more realistic task.
Timing Without Fixed Rules
Machine learning is also changing the concept of market timing.
Traditional execution systems often depend on predefined rules. A strategy may trade at a specific time, respond to a set price movement, or execute a fixed percentage of an order during a given interval.
ML systems can make these rules more flexible by responding to live conditions.
“Instead of asking whether an order should be executed at 10:30, the system can ask whether liquidity, spreads, and volatility at 10:30 make execution sensible,” Smirnov said.
This creates a more adaptive approach.
A model may wait when liquidity is weak, execute more aggressively when volume rises, or change venue when another market offers a better combination of price and available depth.
Smirnov believes this will eventually change how institutions evaluate execution performance.
“Execution speed by itself is an incomplete metric,” he said. “An order can be executed quickly and still produce a poor result.”
The more useful question is whether the order was completed at an appropriate price, with controlled market impact and acceptable transaction costs.
“Institutional investors do not need every trade to be instantaneous,” he explained. “They need every trade to be intelligently timed.”
Quantamental Decision-Making
The development of machine learning is also contributing to the growth of quantamental investing.
Quantamental strategies combine quantitative analysis with traditional fundamental research.
Rather than relying exclusively on company earnings, valuations, or management commentary, investors can use models to process larger datasets and identify patterns that may not be visible through manual research alone.
These datasets may include:
- financial statements;
- earnings-call transcripts;
- economic indicators;
- market prices;
- alternative data;
- analyst revisions;
- news sentiment;
- supply-chain signals.
Large language models may support this process by organizing and summarizing unstructured information. Machine-learning systems can then connect those insights with market data and execution conditions.
“The strongest systems will not treat research and execution as separate worlds,” Smirnov said. “They will connect what an institution believes with how and when it enters the market.”
This could create a more continuous decision-making process, where new information affects not only whether an institution trades but also how the order is carried out.
AI Built Into Hardware
Another development is the movement of AI processing closer to the hardware level.
High-frequency trading systems already work within extremely narrow timeframes. Even minor delays caused by software, operating systems, data transfer, or processor workload can affect execution.
Some firms are therefore exploring ways to run specialized neural-network models directly on chips and programmable hardware.
The objective is to reduce the time between receiving market data and placing or modifying an order.
“In high-frequency trading, firms compete in margins that are almost invisible to a human observer,” Smirnov said. “A system may need to interpret a change and respond before a conventional application has finished moving the relevant data through its software stack.”
Hardware-level processing can reduce this delay, particularly for narrowly defined tasks such as signal classification, anomaly detection, or order-routing decisions.
However, Smirnov warns against treating speed as an unlimited advantage.
“Faster is useful only when the decision is correct,” he said. “An incorrect response delivered in a microsecond is still incorrect.”
For this reason, institutions must balance latency improvements with testing, controls, explainability, and risk management.
Managing Large Orders
Machine learning may be especially valuable when institutions need to execute large positions without creating excessive market impact.
A large order can be divided into smaller sub-orders, but the system must still determine when those orders should enter the market.
Models can assess the limit-order book and estimate whether liquidity is likely to appear or disappear.
They can also evaluate whether passive execution is likely to complete the order or whether the institution needs to trade more aggressively.
“This is where continuous analysis becomes important,” Smirnov said. “The system can review the market after every partial execution and adjust the remaining plan.”
One example of this approach is LOXM, an execution model based on deep reinforcement learning and trained using large volumes of historical and simulated trading activity.
Models of this type can test different order-placement decisions and learn which approaches tend to reduce market impact and transaction costs under various conditions.
They do not remove uncertainty. However, they can evaluate more possible execution paths than a human trader could compare manually in real time.
The Need for Intelligent Speed
Just2Trade supports professional and institutional trading through modern infrastructure and access to international markets.
Its technological environment includes a high-capacity network spanning Europe, the Americas, and Asia, designed to support efficient order execution for clients operating across different regions.
Smirnov believes the future of execution will depend on combining this infrastructure with increasingly adaptive decision systems.
“Within five years, execution algorithms that cannot learn from changing market conditions may begin to look outdated,” he said.
This does not mean that all institutions will use the same models or that automation will remove the need for human oversight.
It means that execution systems will increasingly be expected to interpret context rather than merely follow predefined instructions.
“The next competitive advantage will not come from speed alone,” Smirnov concluded. “It will come from understanding when to use that speed, when to protect liquidity, and when not to trade at all.”
Machine learning is unlikely to make institutional markets predictable. But it may make the process of operating within those markets more responsive, precise, and cost-efficient.
Disclaimer: This article contains sponsored marketing content. It is intended for promotional purposes and should not be considered as an endorsement or recommendation by our website. Readers are encouraged to conduct their own research and exercise their own judgment before making any decisions based on the information provided in this article.
