More than four years ago, Amir Zare came to Armenia from Iran. Today, he not only lives here—together with two other specialists, he is building his own technology startup.

The company is called DataLayer, and it works with data for small and medium-sized businesses. It sounds like a story about complex engineering, but the problem the team is trying to solve is quite down-to-earth: an entrepreneur may have thousands of orders, hundreds of customers, and multiple sales channels—and still not know which product sells best or why revenue suddenly dropped.

Amir knows this problem not only as an engineer. During conversations with potential clients, his team discovered that some entrepreneurs spend between four and twelve hours analysing their own data. The issue is not a lack of data—on the contrary, more often than not there is too much of it, scattered across different systems. DataLayer helps pull all that chaotic data into a single picture, and then, with the help of AI, transforms it into clear answers and recommendations.

Amir spoke about his startup to a correspondent of NEWS.am Tech at the Seaside Startup Summit Armenia 2026.

A Startup That Started with One Person

Amir moved to Armenia more than four years ago, obtained permanent resident status, and officially registered his company here two years ago. As the product grew more complex, the need arose to bring in specialists in AI and marketing.

The team now consists of three people: Amir handles data engineering, a second specialist is responsible for AI, and a third handles marketing. All three are originally from Iran and now live in Armenia.

The startup is still in its growth phase: it has an MVP, it has its first real clients, but there is still a whole list of technical tasks ahead—from automated integrations with social networks to developing the AI assistant and its own LLM.

When a Business Has Data but No Answers

DataLayer is trying to solve an old problem with new technology. A modern small online store may have numerous sources of information simultaneously. Sales happen through the website, Instagram, Facebook, Telegram, Shopify, and other platforms. Some data is stored in a database, some in Excel or Google Sheets, and some remains in correspondence with customers. As a result, the entrepreneur seemingly has a large amount of data at their disposal, but answering basic questions about their own business can be surprisingly difficult.

Which product sells best? Who is the most valuable customer? When were the peak sales periods? Why did revenue drop in the last week? Which items have not sold for a long time? Which customers should receive marketing attention?

According to Amir, during interviews with potential clients, the team found that some entrepreneurs spend anywhere from four to twelve hours analysing their own data. The cause is often not a lack of information but quite the opposite—an excess of it, in a chaotic state.

"They may have dozens of Excel spreadsheets, and it is extremely difficult to organise everything and figure out who the best customer is, which product is the best, how many orders there were, and when the sales peak occurred," Amir explains. DataLayer aims to remove precisely this barrier.

What DataLayer Does

According to Amir, the platform is built on data processing using data engineering approaches. The team uses ETL and ELT, as well as tools such as DBT and BigQuery, to collect information from various sources, structure it, and turn it into clear analytics.

And the business does not necessarily need to have a perfectly organised database. If the business has a website, DataLayer can work with its database. If sales are conducted through Instagram, Facebook, or Telegram, data can be exported from there. If the entrepreneur has stored information for years in Excel, CSV, or Google Sheets, the system can accept those formats as well. Working with "messy" data is, according to Amir, one of the key parts of the product.

After processing, the system does not merely show numbers—it gives the entrepreneur a clearer picture of what is happening. For example, DataLayer can identify the top product, the top customer, and show changes in sales and other key metrics.

One of the approaches used for customer analysis is RFM analysis. It evaluates buyers based on recency, frequency, and monetary value of their purchases. In this way, the entrepreneur gets not just a list of customers, but an understanding of who is most important to the business and who should be the focus of marketing efforts.

Not Just a Verdict, but Recommendations

For Amir, it was essential that DataLayer not become just another dashboard where the entrepreneur is shown dozens of charts and left alone with the figures. The system should help in making decisions. For example, if a particular blue T-shirt has not sold in the last 90 days, the platform can flag this to the store owner. Perhaps the product should no longer be ordered. Perhaps it needs better advertising. Perhaps the issue lies elsewhere.

In the opposite scenario, the system might show that a red T-shirt is among the best-selling products. In that case, it makes sense for the entrepreneur to increase orders of that item. Recommendations will differ across business types. And this, according to Amir, is the core purpose of DataLayer: to give businesses the ability to make "smarter" decisions, based not on intuition but on their own data.

From an Analytics Dashboard to an AI Assistant

The next level of product development involves artificial intelligence. DataLayer already has an AI assistant, which is being developed by the team's AI engineer.

The idea is that the entrepreneur should not have to search for a specific metric in tables or dashboards manually. Instead, they can simply ask a question in plain language. For example: "Who is my best customer?", "Why did my revenue drop last week?", or "Which product is performing best?"

The system should then query the data itself and generate an answer. Thus, the interface gradually evolves from a standard analytics tool into a kind of conversation with the company's data.

This is why integrations with various platforms are important to the team. Currently, DataLayer is at the MVP stage, with real clients in Armenia already using it. In the next version, the developers want to add automatic integrations with Telegram, Instagram, Facebook, and other platforms.

That would solve another problem. At present, data from social networks has to be exported and then uploaded into the system. In the future, the team wants this process to happen automatically. For example, correspondence with a customer on Instagram or Telegram could become part of the overall picture of sales and business performance, with data updating without the need for manual export.

The Toughest Challenge for Armenia Is Language

But when it comes to working with Armenian clients, a problem arises that cannot be solved by integration alone. Many users in Armenia write in Armenian using Latin letters. And they do so with mistakes, abbreviations, and a mixture of Armenian and English words. For a human, the meaning of such correspondence is usually clear. For an AI system, it is far more challenging.

DataLayer's first clients currently use the platform in English. However, the team plans to add Armenian. And here Amir acknowledges that the task is not as simple as just "adding another language". If a user writes an Armenian word using Latin letters, the system first needs to recognise that it is Armenian, then correctly interpret what was written, and only then use that information in analytics.

According to Amir, this will require additional specialists and dedicated work on the AI model. The team is considering using RAG—Retrieval-Augmented Generation, an approach that allows a language model to access external data and use it when generating a response. But that is only part of the challenge.

DataLayer Wants to Build Its Own LLM

The team's plans include building their own local LLM, and the reason is not purely technical. For DataLayer, security is a particularly important concern.

The platform handles commercial data: information about sales, customers, revenue, orders, and customer behaviour. This is sensitive data that businesses are not always willing to entrust to third-party large language models.

That is why the team wants the AI model to run locally. DataLayer already uses a local LLM, though it currently exists separately from the main application. In the future, the team wants to integrate the model more deeply into the product and train it for the relevant language and business scenarios.

This should also help solve the problem of Armenian and Armenian text written in Latin script. Amir says it is technically feasible, but implementation will require expanding the team.