Practical Quantum Data Loading for Financial Intelligence
Working with HSBC, quantum hardware firm Oxford Ionics (an IonQ company), and the UK’s National Quantum Computing Centre (NQCC), Haiqu's team took on the challenging problem of spotting new trade corridors early. In explorative trials, the quantum-assisted models beat their classical counterparts, enabled by a Haiqu loading method that fit nearly four times more data onto quantum hardware than standard approaches.
Spotting trade corridors before they take off
What is the problem? In global trade, a trade corridor forms when trade between two countries climbs well above its usual level and keeps growing. Banks want to spot these early, because a corridor caught as it emerges shows where business is heading. A bank that sees one forming can position itself before the trend is widely visible, shaping where it lends and builds its presence.
The trouble is that new corridors are quite hard to predict. They grow out of many tangled forces, from currency moves to shifts in demand, and labeled examples are rare. There is
also no universally agreed definition of what counts as one, in mathematical terms, so a model cannot simply be told what to look for. Evidence that a trade corridor is emerging appears as subtle patterns across many variables, rather than in any single measure.
Where a quantum computer comes in
Quantum computers are, in principle, well suited to finding patterns like these, rare signals hidden in how many variables relate to one another, so corridor prediction is a good test case. The project used UN Comtrade figures, the United Nations’ public database of world trade, covering 2017 to 2025. Guided by a novel model developed by HSBC, it focused on semiconductors and some of their downstream products, commodities of clear commercial interest in recent years that were also well covered in the records. Months with unusually high trade volumes and strong year-on-year growth were marked as corridors for the models to learn from.
Quantum computers are still early in their development; today’s machines are small and noisy. Getting the most from them means working within those limits, and the most basic limit is how much data fits. A quantum computer stores information in qubits, the quantum version of bits, and standard loading methods manage roughly one variable per qubit. Even on an advanced machine like IonQ’s 36-qubit Forte Enterprise 1, that would mean about 36 variables per data point, a fraction of what years of trade history contain.
To get around this, the team employed Haiqu’s technique, Entangled Manifold Embedding (EME). It squeezed in 139 variables per data point, almost four times the usual limit, with almost no information lost. The quantum computer could then do its part, working out how all those variables interact, a job that can rapidly overwhelm ordinary computers, and handing an enriched version of the data to conventional prediction models.
What the tests showed
Prediction quality for the identification of emerging trade corridors was scored from 0 to 1 using a measure called F1 score, which rewards a model for catching real corridors and penalizes false alarms. For every type of model tested, adding the quantum-based data processing produced a better score.
The two simulation rows come from classical computers mimicking a quantum machine, standard practice for refining an approach before running it for real. One assumed perfect conditions. The other added realistic noise. Simulation cannot be the end goal, though. Ordinary computers can only imitate a quantum machine while the problem stays small, and the promise of quantum computing lies at scales they cannot reach. That is why the last row matters most, and it is encouraging. Even on today’s noisy hardware, a clear gain over the best conventional model came through.
These are promising results from an explorative study. Further validation is required; a determined classical effort could close the gap, and further quantum work could widen it.
For a bank, the point of a higher score is practical. Fewer missed corridors and fewer false alarms mean attention can land on the right trade relationships earlier. That value would ultimately have to be proven on live trade flows.
What this means for near-term quantum applications
The quantum computer did not work alone. It performed one step in a larger process, reshaping the data before conventional models took over, and the project showed that today’s hardware can perform that step on a commercially relevant problem. Many early practical uses of quantum computing are likely to look like this, with quantum processing applied in a modular manner at the point where it can help the most.
Interestingly, the same approach could extend to other problems where a weak signal is buried in a mass of highly correlated data, such as fraud detection and risk monitoring. Larger datasets and tougher classical comparisons would be the natural next steps. These are promising results for an early proof of concept, and a glimpse of how quantum computing could help with real problems in global trade.