Everyone in finance software now claims to have "AI-powered collections." Very few mean the same thing by it, and fewer still can show it working. Over the past two months we mapped the vendors building AI agents for collections and dunning, scored them on confirmed evidence rather than marketing copy, and plotted where each one actually sits. This is the result: our market map of AI-powered collections agents, and our first read on the state of play.
What we did
The map is built from desk research. We ran a structured sweep of the market drawing on company websites, API documentation, security and compliance pages, funding records, named case studies, and third-party review sources. Where a vendor claimed a capability, we looked for evidence it was live and at scale before crediting it.
Capabilities are evolving very fast. Things that were not possible 12 months ago are now live in production. Each new model release expands the realm of the possible. Newcomers appear every month if not every week. This paper is a snapshot as at June 2026. Depending on demand and feedback, we’ll update it in due course to make it a living market map. See also our comments at the end of this article for further details on what is in- and out-of-scope.
We scored each vendor across eight dimensions, each on a 1 to 10 scale, using confirmed public evidence only. Roadmap promises and marketing language did not count. The dimensions: company substance, compliance robustness, depth of customer base, geographic scope, scope of activities, scope of AI capability, maturity of AI capability, and integration ecosystem.
Two of those dimensions deserve a word, because they are the heart of this category.
- Scope of AI capability measures the ceiling of what a vendor's AI is designed to do, from rule-based automation (level 0), through predictive scoring and “co-pilot” message-drafting, up to genuinely agentic behaviour (autonomous multi-step outreach, inbound reply handling, payment-plan negotiation) and autonomous voice calls.
- Maturity of AI capability measures something entirely separate: how reliably and verifiably the vendor delivers what it claims; what evidence there is of reliable outcomes and sustained performance.
A high ambition with no live evidence scores well on scope and poorly on maturity…and of course there are plenty of exaggerated claims in this market right now!
The map itself organises vendors on two axes: AI Capability and Scope of AR services (specialist collections tool through to broad O2C platform), while Bubble size reflects company maturity (substance in terms of employees and capital; compliance; and depth of customer base).
AI Collections & Dunning Market Map, Liquidity Lab, June 2026
The state of play: early, and led from the edges
The space is still early in its development. The defining feature of a true collections agent, the ability to read an inbound debtor reply, understand it, and act on it without a human, is still limited to a handful of providers. Many products that market themselves as AI collections are running predictive scoring or generative message-drafting on top of fundamentally rule-based dunning sequences. That is useful, but it’s a long way from realising the full potential of AI.
The pace is being set by young specialists rather than incumbents. The vendors pushing hardest on genuine agentic capability are mostly companies founded in the last three or four years that do one thing: Cleavr, Recombine, Stuut and AgentCollect (previously operating as Respaid) all score at or near the top of our scope-of-AI-capability dimension. These businesses treat the collections agent as the product rather than a feature. Stuut in particular stands out for pairing high ambition with relatively mature, verifiable delivery; while Recombine stands out for the most mature voice-agent capability.
It is worth being clear-eyed about what "leading" means at this stage. These are small companies, several at seed stage, and often with narrow geographic reach. They are setting the direction of the category without yet having the substance to dominate it. The interesting question for the next eighteen months is whether they can build the trust and distribution to convert technical lead into market position before the larger players wake up.
Among the larger players
Two more established names are worth calling out, for opposite reasons.
InDebted is one of the most advanced operators in the entire landscape: strong on company substance, compliance, customer depth, geographic reach, and the maturity of its AI. But it plays in a genuinely different space. InDebted is an enterprise, consumer-debt (B2C) recovery business operating as a licensed collections agency across multiple markets, sharpened further by its acquisition of Receeve in December 2024. It is a useful benchmark for what a mature AI collections operation looks like, but it is scarcely present in the B2B accounts-receivable SME and mid-market ground where most of the specialists are fighting.
Kolleno is the more telling case. Among the accounts-receivable automation tools built for finance teams, it leans into AI and agentic features more than most of its peers and backs it with notably strong compliance credentials. Kolleno shows what the AR-automation incumbents could do if they committed to the agent model.
The incumbents are mostly cautious, and some are absent
By and large, the established AR and collections players are taking a more cautious path. Upflow is the clearest example. The platform, which already processes over $1bn a month for SME and mid-market finance teams across Europe and North America, started out with portfolio scoring and email drafting. They are clear that the destination is autonomous agents, but they are deliberately restricting themselves to the “copilot” autonomy levels until they are convinced the tools are ready.
Others have effectively sat the question out. Clearnox, the French SME invoice-tracking and light-dunning tool, is now a product line inside Wolters Kluwer, following an acquisition in September 2024. Clearnox scores near the bottom of the market on both scope and maturity of AI capability. A strong, stable, widely used product, now owned by a large listed corporate, with little visible engagement in the agentic shift happening around it. It is a clean illustration of the incumbent's dilemma: the installed base and the corporate parent are precisely the things that make moving quickly hard.
What's next?
This map is a starting point not a verdict. Over the coming weeks we will publish deep-dive profiles of the key players named here, looking past the marketing at what each vendor's AI actually does, how mature it is, and where it fits. We will keep the map updated as the market moves, and it is moving fast.
If you're in this market, we'd like to hear from you
- Think we missed you? If you're building in AI collections and you're not on this map, tell us: contact@liquidity-lab.com.
- On the map but disagree with where we placed you? We'd genuinely like to talk. Same address: contact@liquidity-lab.com.
- Considering AI agents in your own collections or AR function? This is exactly what we do. Get in touch and we'll help you cut through the noise: contact@liquidity-lab.com.
What is the scope of our study?
This study focuses on two specific types of platforms:
- First, standalone AI agent platforms which integrate into existing ERPs, CRMs or dunning platforms.
- Second the dunning and accounts receivable platforms who have implemented agentic features.
There are other types of AR platform which are outside the scope of this study, such as Order-to-Cash platforms (High Radius, Esker…), the collections platforms used by banks and other lenders (e.g. Qualco, C&R, Katabat… ). We also don’t cover the agentic features introduced by the neobanks and the ERPs themselves. All these will be the subject of future publications.