Behavioral History Won't Tell You What Someone Will Buy. Intent Will.

01 /The Concept

Replenit has built what it calls an "AI brain" for improving buyer retention in e-commerce.

There is no shortage of platforms making similar promises. But this startup has a conceptual angle that feels more compelling than most – and, more importantly, early pilots are already backing it up. One retailer increased its visitor-to-buyer conversion rate 12x. Another grew repeat revenue from existing customers by 235%. A third found that Replenit-driven sales now account for 12.7% of total revenue.

So what exactly is the concept?

Cognitive psychology has a concept called "theory of mind" – the human ability to understand that other people have their own unique thoughts, intentions, desires, beliefs, and emotions that may differ from our own perception.

Replenit applies this idea directly to e-commerce. It creates a dedicated AI agent for each individual shopper – an agent whose job is to model that buyer's "theory of mind" by studying their purchasing habits and behavioral patterns.

The result: a specific AI agent can predict what a specific person is likely to want to buy at any given moment. That prediction then determines which products the store surfaces when that visitor returns to the site.

Instead of showing "frequently bought together" recommendations based on the current item, the store shows products that actually make sense for this particular person – perhaps because they buy that item on a regular cycle and the timing is right, or for some other reason tied specifically to them.

The same logic applies to promotions. Rather than pushing "the most popular item with the best discount," the platform identifies a product the customer would likely have bought anyway – maybe later, maybe eventually – and surfaces it precisely when a relevant promotion is running.

Obviously, this opens up a wide space of personalization signals. But above all the individual buyer agents sits a coordinating layer – an AI agent called Maestro. Maestro's job is to filter the list of products that could reasonably be shown to a given person and select only those that also align with the store's current business priorities.

Those priorities might include raising average order value, increasing purchase frequency, growing total revenue, improving margin, or reducing churn. In an ideal world, a store wants all of these at once – but that's not how it usually works. So the store ranks its priorities, and Replenit's recommendations are then optimized to serve that ranked list for each individual buyer.

Replenit was founded last year in Poland – interestingly, by a team of Turkish founders. In February of last year it raised an undisclosed pre-seed round, and has now followed that with a new €2.1 million round – still described as pre-seed. Notably, one of the investors in this round is a co-founder of ElevenLabs, another Poland-based startup.

02 /Why It Matters

In its blog, Replenit argues that the next phase of e-commerce will be won by stores that make better decisions. But what distinguishes a better decision from a worse one?

Today, every store treats shopper activity as data – something to be recorded and used as the basis for choices. Replenit goes one level deeper. It treats behavior as an external signal of internal desire – the outward expression of a person's "theory of mind." So instead of using behavioral signals as-is, the platform tries to decode them: why did this happen? What was actually going on in this customer's head, and why did it result in this specific action?

Only by decoding that deeper layer of intent can you reliably predict what a person might want to buy next – with significantly higher accuracy than any surface-level behavioral model.

This means the winner in e-commerce's next chapter won't be the store that collected the most behavioral data. It will be the store that best translates that data into individual intent.

And that, in turn, makes segmentation obsolete. Every customer is their own segment. Every decision should be based on that customer's individual intent model – which is exactly what Replenit enables.

This connects directly to a [related review](/review/psihologicheskoe-segmentirovanie-auditorii) of Psympl – a platform for improving financial services sales through psychological segmentation. Psympl doesn't go all the way down to the individual level, but it identifies five distinct financial psychotypes and uses a short questionnaire to assign users to a segment.

From there, Psympl's Psymplifier tool helps marketers craft pitches tailored to the language and values of each segment.

Today's Replenit also echoes a concept [covered previously](/review/chto-nuzhno-delat) in the context of Pomo – a platform from a completely different domain that promises to bring Fortune 500-level marketing capabilities to any small business.

Pomo's key argument is similar: the real differentiator for large companies isn't that they execute better – it's that they have infrastructure for making better decisions. Small companies lack that infrastructure, which, as Pomo puts it, makes roughly 90% of their strategic decisions wrong.

The thread running through all three startups is the same: the companies that win are those that have learned to decide well.

03 /Opportunities

The broader trend is clear: the next generation of AI platforms needs to "have brains"

That means not just executing – and not just analyzing surface-level data. The platforms that matter will interpret data at a deeper level and make decisions based on that deeper understanding rather than on what's visible on the surface.

For Replenit, that deeper level is the individual buyer's theory of mind. For Psympl, it's the psychological archetypes of financial services customers. For Pomo, it's a real-time reading of market trends and competitor behavior.

What could that new "deeper level" look like for your product?