Mathema.me on Growth: Why Human Accountability, Better Data, and AI Matter More Than Another EdTech Playbook

Education is supposed to be one of the industries AI changes beyond recognition. If a model can explain algebra, generate exercises, prepare students for exams, and answer questions at any hour, paying another human to teach you can start to look inefficient.

But that assumes teaching is mainly about transferring information. Mathema.me is making a different bet: sometimes people already have access to everything they need to learn. What they are missing is another person who makes them actually do it.

In the latest episode of the Retention Podcast, Oleg Lesov talks with Pavlo Pedenko, Co-founder at Mathema.me, about building and scaling a one-to-one tutoring business, expanding across markets, choosing between subscriptions and lesson packages, fixing lifecycle communication, and finding the parts of education where AI is genuinely useful.

Pavlo’s path to Mathema.me went through engineering, marketing, MacPaw, his own failed startup, one of Preply’s early growth teams, and Wise. Mathema.me itself started while he was still at Wise, and Pavlo didn’t treat leaving a successful corporate career as a romantic leap into entrepreneurship. He waited until the business had enough evidence behind it. When monthly sales reached roughly 300K, he felt confident that the company was likely to survive.

Today, the scale provides more evidence. Mathema.me conducted around 600,000 one-to-one mathematics lessons in 2025, and in one recent month alone delivered about 70,000 lessons and paid 1,500 tutors.

The more interesting question, though, is why customers keep choosing human tutoring when AI can explain the same mathematics almost for free.

People Aren’t Always Paying for Knowledge

A motivated student has never had more ways to learn independently. Books, courses, YouTube, educational apps, and now AI can explain almost any school-level mathematics problem. For Pavlo, that is precisely why content itself is not the strongest argument for one-to-one tutoring.

The scarce part is motivation.

A student can ask Claude to break down a difficult equation step by step. But doing that consistently requires the student to decide to sit down, focus, ask the questions, and keep going when the work becomes frustrating. For children who already struggle with mathematics, that is often the hardest part.

Pavlo sees one-to-one tutoring as a commitment mechanism. A child has a specific person waiting for them, a scheduled lesson, and an expectation that they will show up and participate. It works much like paying for a personal trainer even when you already know how to exercise.

“One-to-one tutoring… has never been about content delivery. It’s always been about this real commitment that one individual could form in front of another individual.”

That changes the product being sold. Mathema.me is not only selling mathematics instruction. It is selling accountability, attention, and the relationship that makes regular learning more likely.

This distinction becomes important as AI enters more educational products. If the core customer problem is access to information, AI can be extremely disruptive. If the problem is motivation, trust, accountability, or human connection, replacing the human may also remove part of the value customers were paying for in the first place.

Don’t Quit Your Job Just Because the Startup Story Sounds Better

Startup culture likes the clean version of entrepreneurship: identify a problem, quit your job, commit completely, and figure everything else out later.

Pavlo took almost the opposite approach.

He describes entrepreneurship as moving from the resources and talent of companies such as Wise into an environment where money, people, and time are suddenly scarce. Instead of treating that uncertainty as something founders should simply embrace, he tried to remove as much of it as possible before committing fully.

Mathema.me already had customers. Tutors were willing to work through the platform. The team had evidence of demand. By the time monthly sales reached roughly 300K, Pavlo saw enough proof that the business was likely to stay alive.

That doesn’t eliminate startup risk, but it changes the decision from “I believe this should work” to “there is enough evidence that people want this.”

It is a useful growth principle beyond entrepreneurship. Companies often confuse conviction with validation. The stronger approach is to keep reducing the number of assumptions until the remaining bet is one you can reasonably afford to make.

Subscriptions Aren’t Automatically the Better Business Model

For a recurring service, subscriptions sound like the obvious monetization model. Mathema.me currently does something different: it sells packages of lessons and offers another package when those lessons run out.

The reason comes down to cash flow and customer acquisition economics.

Pavlo explains that a monthly subscription might reasonably cover around eight lessons, while a package can include ten or twelve. That larger upfront transaction gives the company more room on CAC and lets it reinvest the cash earlier, while already having a clearer view of expected LTV.

Subscriptions have their own advantages. Revenue becomes more predictable and cohorts can become more stable. Pavlo saw those benefits at Preply, which moved from packages toward subscriptions. But he also points out a less discussed consequence: if average transaction value drops when the business switches models, CAC takes longer to recover. For a company without a large cash reserve, that transition can create a real cash-flow gap.

Mathema.me hasn’t ruled subscriptions out. In the UK, customers have already expressed interest in weekly subscriptions because they fit household budgeting better. In Poland, the business has found another solution: roughly 15% of GMV comes through buy-now-pay-later instruments. Mathema.me receives the money upfront while customers spread their payments over time.

The broader lesson is that monetization shouldn’t be selected because a model is fashionable in your category. The right setup depends on customer behavior, cash flow, CAC, LTV, and even local payment habits.

Entering a New Country Means Relearning the Customer

Mathema.me has customers across 75 markets, although its core audiences are primarily Ukrainian and Polish. When the company expanded from Ukraine into Poland, the team initially expected many of the same mechanics to transfer.

They didn’t.

Payments were one of the first surprises. Polish customers strongly preferred local payment rails such as BLIK and Przelewy over cards. In Ukraine, some customers still wanted to pay by invoice or bank transfer, and Pavlo even recalls cases where customers were uncomfortable entering card details online and needed a sales consultant to help complete the transaction.

The marketing triggers differed too. Based on Mathema.me’s own campaign data, Ukrainian parents responded strongly to messaging around their child falling behind academically. Polish campaigns performed better when they addressed how the child felt about struggling at school. Pavlo is careful not to present this as a universal cultural truth; it is simply what their own creative performance showed.

Channel behavior changes as well. Email retention flows work particularly well with Polish customers and appear useful in the UK, while they are harder to make work in Ukraine. Phone sales show almost the reverse pattern: personal calls can be important for Ukrainian customers, less so in Poland, while early UK experience suggests people are particularly resistant to being called.

The product may still be mathematics tutoring, but the commercial system around it changes market by market. Localization therefore goes much deeper than translating a landing page. Payment methods, creative hooks, lifecycle channels, sales processes, and customer expectations can all need to be rebuilt.

International Expansion Is an Organizational Problem Too

For a marketplace, launching a new country means getting both supply and demand right at the same time. Tutors need to be available, customers need to arrive, payments and payouts need to work, acquisition needs to make economic sense, and operations need to support the entire process.

The difficult part is that those responsibilities rarely sit neatly inside one function.

Large marketplace businesses can have dedicated launch teams whose only job is to enter a new city or country and coordinate all the pieces. Mathema.me is still a relatively small company, which means much of that cross-functional work falls to Pavlo and his co-founder.

This creates a different scaling bottleneck. The company may know that demand exists, but entering another market still requires someone who understands enough about marketing, product, supply, payments, sales, and operations to make the system work as a whole.

That is why adding more people to one department does not necessarily solve expansion. Mathema.me has around 50 sales consultants, for example, but those people are there to handle customer conversations. They are not a substitute for someone who can design the business for a new market.

Lifecycle Marketing Gets Hard When Your Data Isn’t Ready

Mathema.me has three distinct audiences moving through the platform: parents, students, and tutors. That already makes lifecycle communication more complicated than a standard subscription funnel.

The bigger problem, according to Pavlo, is data.

The company has customer-level attribution and analytics around first- and last-touch campaigns, but moving from basic CRM coverage to more sophisticated recovery, upsell, and personalized journeys requires clean enough data to know what should happen and when.

His advice for early-stage founders is unusually direct:

“Make sure you’re taking care of your data from day one, because it will bite you in your ass.”

The problem becomes clearer when acquisition and lifecycle are viewed together. Mathema.me ran a successful Winter Camp campaign in Poland, effectively repackaging its existing tutoring product around keeping children engaged during the school break. Yet customers acquired through that specific proposition could still enter the same generic post-signup communication as everyone else.

The technology can send the message. The harder question is whether the company has the data, ownership, and processes to know what message belongs there.

That is the difference between having CRM infrastructure and having lifecycle marketing.

The Customer Journey Should Remember Why the Customer Came

Pavlo points to another gap that many growing companies eventually encounter: nobody owns the journey from the first creative through the post-purchase experience.

Acquisition teams optimize creatives. Product teams optimize the product. CRM teams send emails. Sales teams handle leads. Each node can work reasonably well while the complete journey still feels disconnected.

Imagine that an ADHD-focused creative performs particularly well in Poland. The customer clicks because that specific problem resonates, signs up, and then immediately receives generic messaging about mathematics lessons. The acquisition promise disappears as soon as the conversion happens.

A stronger lifecycle model would preserve that context. The campaign that brought the customer in could inform the funnel, onboarding, follow-up messaging, and eventually the upsell or retention strategy.

For app businesses, this becomes particularly interesting because the same core product can support multiple journeys. Once monetization is established, different acquisition messages can feed into different onboarding and lifecycle paths without requiring teams to build a completely separate product for every segment.

The opportunity is not necessarily to send more messages. It is to stop throwing away the context you already paid to acquire.

AI Probably Won’t Replace the Tutor — But It Can Remove a Lot Around Them

Mathema.me’s position on AI is more nuanced than either “AI will replace teachers” or “education needs humans forever.”

Pavlo believes parts of tutoring will absolutely be disrupted. Structured exam preparation is an obvious example. Motivated students who mainly need repetition and revision can already get significant value from AI without paying for a tutor.

Where he is more skeptical is the idea that the entire tutoring relationship can be reduced to content delivery. Mathema.me’s customers are often paying precisely because motivation and accountability are difficult to automate.

That doesn’t mean the company is avoiding AI. Quite the opposite.

Mathema.me uses it to help tutors prepare homework, write study reports, generate interactive exercises, and provide feedback to parents. The objective is not to remove tutors but to understand what makes the strongest tutors effective and help more tutors work closer to that standard.

AI therefore sits around the human relationship rather than replacing it. The tutor keeps the part where human presence creates value; automation handles more of the repetitive work around that interaction.

The Biggest AI Gains May Be Behind the Product

Some of Mathema.me’s most practical AI use cases have little to do with teaching mathematics.

The company has automated parts of customer support while continuing to grow without increasing support headcount at the same rate. Sales quality assurance, previously handled manually, is now automated through an AI system that listens to calls, evaluates them against the script, and provides feedback at the individual salesperson level.

That matters when you have a sales team of around 50 people.

Another internal tool may be even more revealing. Mathema.me built a Slack bot with access to its database, allowing employees to ask basic analytical questions directly inside the place where they are already working. During one CRM discussion about the right timing for an email nudge, Pavlo expected the team would need engineering help to pull the relevant data. Instead, he asked the internal AI analyst in Slack and received an answer before he had finished the conversation.

This is where AI starts to change operating speed rather than simply content production. The value comes from removing the delay between having a question and getting enough information to make a decision.

The caveat is important, though. Pavlo points out that asking the right question still requires experience. AI can process the data quickly; deciding what is worth asking, interpreting the result, and knowing whether it makes sense still requires judgment.

The Best AI Interface May Be Where Work Already Happens

The Slack example also says something broader about AI adoption.

Companies can build sophisticated AI features and still struggle to get employees to use them if those features live somewhere people have to deliberately visit. Put the same capability inside a conversation where the work is already happening, and the behavior changes.

Instead of “using an AI analytics tool,” an employee can ask a colleague a question, bring the bot into the thread, get the answer, and continue the discussion.

Pavlo describes this kind of experience as particularly powerful because Slack is already where the work happens.

For marketing and lifecycle teams, this can be particularly useful because decisions often require context from several systems at once: campaign performance, CRM data, user behavior, revenue, and previous experiments. The closer AI gets to those workflows, the less time teams need to spend moving between tools simply to assemble the context for a decision.

In a Nutshell

Mathema.me’s story is interesting because it sits directly inside one of AI’s most discussed industries while refusing the simplest AI narrative. The company isn’t betting that technology disappears. It is betting that technology and humans will become valuable for different reasons.

AI is already taking over repetitive analysis, sales QA, parts of customer support, tutor preparation, and other operational work. It may also take a meaningful share of structured learning and exam preparation. But where the customer is paying for commitment, accountability, trust, and another person’s attention, replacing the human can mean replacing the product’s strongest value proposition.

The same pragmatism appears elsewhere in the business. Mathema.me doesn’t use subscriptions simply because recurring revenue sounds attractive. It considers transaction size, CAC recovery, LTV, and cash flow. It doesn’t assume a funnel that works in Ukraine will transfer to Poland or the UK. And it has learned that sophisticated lifecycle marketing depends less on sending capability than on data quality and clear ownership of the customer journey.

Perhaps the clearest takeaway is that growth becomes harder as the system becomes more connected. Acquisition affects lifecycle. Local culture affects payments. Monetization affects CAC. Data quality affects personalization. AI changes operational economics without necessarily changing what customers value.

The companies that handle that complexity well won’t be the ones that automate everything. They’ll be the ones that know which parts of the experience should become faster and cheaper — and which parts customers still want another human to provide.