Founded by Nicolas Rieul and Nans Thomas, Actionable announces financing of 8.5 million euros led by Hi inov, with Axeleo Capital. The startup wants to help large companies anticipate their customers’ departures, complaints and purchases by bringing together commercial data and operational events. Its deployments at Carrefour, OUIGO and Engie illustrate this ambition. The change in scale will depend on its ability to reproduce these uses and measure the benefit of the resulting decisions.
Six minutes and twelve seconds, according to Actionable, is the waiting threshold at the drive-thru beyond which the Net Promoter Score, the customer recommendation indicator, drops in the data analyzed at Carrefour. The brand would have transformed this result into a monitoring standard, implemented at national, regional and store by store levels.
Precision attracts attention, its operational translation counts more: an analysis result can lead to examining the preparation of orders, the availability of teams or the organization of withdrawals. Customer knowledge then enters into the decisions that determine the quality of the service and its cost.
This is the place that Actionable seeks to occupy. The company announces, on September 9, 2026, a fundraising of 10 million dollars (8.5 million euros) led by Hi inov, with the participation of its historic investor Axeleo Capital. Created in April 2024, it brings together the experiences of Nicolas Rieul, former manager of Criteo, and Nans Thomas, formerly in the product management of Innovorder and founder of Wino. The first knows the business decisions that companies seek to improve; the second brings his experience in building business software.
Going from average to customers to call back
The problem described by the founders starts in front of a dashboard. A drop in satisfaction appears, the teams look for its origin, then must determine which people to contact and what response to give them. Between the collective observation and individual intervention, several decisions remain to be made.
“The measure existed, the granularity necessary for the decision did not. It’s this layer that we built,” explains Nicolas Rieul.
Actionable claims to be able to assign satisfaction, risk of departure, serious complaint and repurchase scores to each customer, along with the factors that explain them. The company has sixteen major accounts, including Carrefour, SNCF, Edenred and Engie, and the journeys of 117 million consumers analyzed.
The business context is built before the prediction
To produce these results, Actionable begins by reconciling the available information. An order, a wait, a return and a complaint can be recorded in different software. You still have to find the same customer, put the events in the right order and understand what each field means.
The platform reconstructs this journey in a “Common Customer Data Model”, a structure adapted to the sector concerned. The time it takes to prepare an order, a train delay or the processing time for a file become variables that can be interpreted in their context.
“The hard work is transforming hundreds of in-house tables and definitions into a customer model that a machine can use without making mistakes,” explains Nans Thomas. The company claims to have spent two years on this construction and reduced work that could take months of engineering to just a few days.
This promise directly engages the economics of the product. Each new contract must benefit from the connectors, definitions and treatments already developed. The reusable work share determines the resources required for deployment. Actionable does not detail the median time observed or the volume of human intervention required. These two measures would make it possible to assess its capacity to increase the number of installations.
A commercial campaign and a service organization to adjust
The uses presented by Actionable show how this preparation can fuel different decisions. At Carrefour, individual predictions are used, according to the company, to determine which customers receive which CRM campaigns. The company claims an incremental return on investment of seven times the stake.
At OUIGO, the model is used to identify travelers likely to be dissatisfied and send them a commercial gesture before a possible complaint. Actionable says the additional revenue generated by an initial campaign covered the cost of the platform within a few weeks. At Engie, customer service teams would rely on the detection of complaints likely to escalate to intervene before escalation.
These situations refer to the same allocation of resources: which customers to call back, for whom to reserve compensation and which incidents to treat as a priority? They also open up a trade-off between repairing a bad experience and investing in the organization to reduce its frequency. A repeated delay may justify revisiting a process; its commercial consequences may call for immediate intervention.
Is the customer who risks leaving the one we can keep?
This question marks an additional difficulty, let’s imagine two clients presenting a comparable initial risk. Compensation may convince the first to return. The second may have moved or chosen an offer that the company cannot respond to. An identical risk score does not necessarily justify the same expense.
Actionable claims the use of “causal AI”. This approach aims to shed light on the cause and effect relationships and consequences of an intervention. It requires distinguishing the factors associated with behavior from those on which the company can act with an expected result.
The distinction is methodological, reference tools in causal inference, such as DoWhy, are based on explicit hypotheses concerning the relationships between variables and on the examination of their robustness. An observed correlation between waiting and dissatisfaction is not enough to quantify what a reduction in waiting would change, all relevant things taken into account.
Actionable does not detail the methods used; nor the validation of the effects attributed to actions. In commercial campaigns, comparable control groups make it possible to assess what would have happened without intervention. For operational changes, the protocol must take into account other factors that may influence the results.
This rigor conditions the interest in large-scale deployment. A good prediction can improve knowledge of the customer portfolio. A well-chosen intervention must produce an additional result, beyond its cost.
A place to be established in the face of already established suppliers
The reconciliation between experience data, operational data and commercial actions is already present in competing offers. Qualtrics describes in particular functions combining this information to predict individual behavior, identify customers at risk and organize their follow-up.
Actionable will therefore have to establish its advantage in the concrete conditions of use: deployment time, effort required of the teams, quality of predictions, explanation of results and integration with existing tools. Large groups with data teams can also compare purchasing a platform to developing their own models.
The space occupied in the client’s budget matters just as much. A solution used by several professions, included in their regular decisions, can become more difficult to replace. This position involves demonstrating its usefulness to marketing, customer service and operations, with results that everyone can understand.
The 117 million consumers analyzed do not, however, constitute a proprietary shared database. Actionable says its customers’ environments are isolated and their data is not shared between them. Its cumulative advantage could therefore be built in the sector models, integrations and deployment methods that it knows how to reuse.
The agents broaden the ambition, the international tests the economy
The company now intends to extend the use of this data model to AI agents; it already offers Actionable Intelligence, an analytical agent capable of producing analyzes from structured information.
This extension could increase the frequency of use of the product and the number of functions concerned. It also adds requirements: tracing calculations, verifying recommendations and determining who validates decisions.
The funding must support recruitment in product, engineering and sales, as well as international expansion, particularly in the United States, through a network of reseller partners. It extends a first round of 2 million euros announced in September 2024, which combined capital and debt.
