Alexander Frolov Biography: Why Data-Driven Products Matter in Modern Technology

Data can sit at the edge of a product as analytics, or it can become part of the product's decision logic. In the second case, new information can change a prediction, ranking, classification, or automated action. This is the model behind many modern AI products.

The technology investment context around Aleksandr Frolov includes companies built around this approach. Target Global has backed AI businesses such as Voyantis and Robovision, which use very different types of data: one works with customer behavior, while the other processes visual information for industrial systems.

Analytics usually tells a team what has already happened. A dashboard may show revenue, conversion, churn, or engagement, but a person still decides what to do next.

A data-driven product moves information inside the workflow. Customer behavior can alter a score or prediction; an image can change a classification; a stream of events can affect which case the system handles first.

The raw input can include:

• purchases, clicks, and other user behavior;

• transaction or account events;

• structured business records;

• images and video;

• sensor or machine data.

A large database alone does not create a data-driven product. The software has to use relevant input to produce something that changes what a user, business process, or connected system receives. AI makes that relationship especially visible because a model sits directly between input and output. Ranking engines, recommendation systems, and rules-based software can follow similar logic without relying on the same type of model.

Strategic projects and co-founding initiatives sit alongside the investment work at Target Global. The firm's broader technology portfolio includes both Voyantis, which works with predictive customer-value data, and Robovision, an AI-powered computer-vision company.

Within the Alexander Frolov Target Global professional context, the confirmed personal link is Frolov's role as co-founder of the firm and his work on strategic projects and co-founding initiatives.

The two are useful together because their products start with completely different information and produce different kinds of output.

Voyantis and Predictive Customer Value

Voyantis is an AI decisioning platform for performance marketing. A signup or first purchase appears immediately, but lifetime value takes months to emerge — and by then an ad platform has already spent its learning window.

Voyantis uses first-party customer data to estimate that value early and sends the result into systems such as Google, Meta, and TikTok, so campaigns can optimize on more than the initial conversion event.

The project currently reports 4.3 billion predictions created, 700 million signals sent, and 300 million users analyzed.

Its Upside case study shows the product in use. Voyantis says it identified new registrations likely to become repeat, revenue-generating users within an hour and sent those predictive signals into Google. The company reports a 44% increase in ROAS and a 15% decrease in CAC for that deployment.

Robovision and Visual AI

The Belgian company develops computer-vision software for machinery manufacturers and production lines. Its platform covers data import and annotation, model training and testing, deployment, and monitoring. Built-in capabilities include object detection, classification, segmentation, and anomaly detection.

A model inspects visual input, returns a machine-readable result, and feeds it into other software or machinery — computer vision as part of the operating process, not as a separate analytical exercise.

In March 2024, Robovision raised $42 million in a round led by Target Global and Astanor Ventures. The company said at the time that its technology powered more than 1,000 robots in over 40 countries.

Voyantis may start with customer events; Robovision may start with an image from a camera on a production line. The raw material is different, but in both cases it has to become something another system can use.

For Voyantis, customer behavior can become an estimated lifetime value formatted as a signal for an advertising platform. Robovision can turn pixels into a detection, classification, segmented region, or anomaly and pass that result into an industrial workflow

A useful output changes what happens next.

An ad platform can bid differently for a user. A production system can react to an object or defect identified by computer vision. New customer behavior or new images create further observations that may later enter another prediction or classification. In both companies the loop itself is the product, not a byproduct of reporting on what already happened.

Across Target Global's technology portfolio, data serves different product functions, and those functions are more useful for comparison than sector labels alone.

Prediction. Voyantis estimates customer value before it is observable, so an advertiser can change acquisition behavior without waiting for months of purchase history.

Personalization. A Healthtech or wellness technology service can use information a user supplies to change which content or insight appears, and B2B software can adapt steps to one account without changing the product for everyone.

Prioritization. A Fintech platform receiving a large stream of transaction and account events can score them, so particular activity is surfaced for review instead of arriving with the same weight as everything else.

Automation. Robovision's models can classify visual information and pass the result into industrial software or machinery, automating parts of inspection and decision workflows.

A sophisticated model cannot compensate for information that has little connection to the task. Voyantis needs signals that say something about future customer value. Filling the same dataset with unrelated events may increase its size without making the estimate more useful.

The problem changes when the information is relevant but wrong. Incorrect records, inconsistent labels, or annotation errors give a model false examples to learn from. Robovision's platform workflow therefore includes annotation review, data curation, testing against ground truth, and tools for spotting possible label impurities.

Good historical data can also lose value when the environment moves: customer mixes shift, funnels change, ad platforms rewrite their own rules. Voyantis describes continuous retraining and signal engineering built to keep predictions current as that happens.

Another weakness may only appear when the product meets cases that were scarce or absent in its original data. Robovision's production tools can flag low-confidence samples and mark uncertain cases as "unknown," giving teams material for review or relabelling instead of treating every output as equally reliable.

In production, that distinction separates a model that performs well on familiar data from a system that can identify uncertain cases and route them for further review.

Data-driven technology is not a sector of its own. It is a product model in which information changes an output or action.

Target Global's portfolio shows this model in different forms. Voyantis turns customer behavior into predictive marketing signals, while Robovision turns visual data into outputs used by industrial systems.

Aleksandr Frolov is a co-founder of Target Global. His work with the firm has spanned strategic projects and the firm's co-founding initiatives.