How OLX is using AI to reduce repetitive tasks

At OLX Group, we start with everyday friction and build AI tools that make work simpler for customers, professional partners and OLXers.

Nobody chooses a career because they love filling out forms, writing call notes or looking through a spreadsheet for a number they know they have seen before. Still, a surprising part of the working day disappears into tasks like these. They need to be done, but they are rarely where people do their best work.

A car dealer creates value by understanding customers and selling cars, not by writing descriptions for every new vehicle. A salesperson creates value through conversation, not by trying to remember every detail once the call is over. A data analyst should be able to spend time explaining what the numbers mean, rather than cleaning the same spreadsheet before every meeting.

This is often called the work around the work. It is necessary, but it should not take over the day. At OLX Group, this led us to a simple question: what work would nobody miss?

The answer is not one task or one tool. It is a collection of small moments that create unnecessary effort throughout the day. When those moments become easier, people have more time and attention for customers, decisions and the work they are uniquely qualified to do.

Starting with the blank screen

Imagine you want to sell something. You take a photo, choose the category and add a few details. Then you reach the empty description box and have to decide what to write, which details matter and how much information is enough. A task that seemed simple suddenly starts to feel like work.

At OLX, Auto Description helps people move past that moment. It uses a photo and some basic information to prepare a first draft of the listing. The seller can review it, change it and decide what to publish, but they no longer have to start with a blank screen.

The tool can identify visible characteristics of an item and use the information provided by the seller to suggest a relevant description. It does not take control of the listing away from the person creating it. It provides a useful starting point, while the seller remains responsible for checking the information and deciding what appears in the final advertisement.

It is a small change, but the time saved adds up. Auto Description saves people around two million keystrokes each month. Over 90 days, this amounted to more than 20 days of typing.

The quality of the suggested descriptions also matters. Human evaluators rated Auto Description at 98% for style and relevance and 89% for faithfulness to the facts. These evaluations help our teams understand where the tool is working well and where it still needs to improve.

The aim is not to remove people from the process. It is to make the process easier for them.

Tomek Jamiński, our Director of Data Science, puts it this way:

“From the very beginning, we understood that AI can help us solve important customer problems. We wanted to use all the cutting-edge technologies to give our customers more time to be human. If our AI automation can reduce the friction and hence the time spent on data input, it means our customers won. And so did we.”

Giving people more time to be human sounds like a big ambition. In practice, it often begins with ordinary moments like this one. It can mean getting past an empty description box more quickly, keeping track of an important conversation or finding the information you need without searching through several systems.

What happens after a sales call?

Consider what happens when a salesperson finishes a good conversation with a customer. They understood what the customer needs, answered their questions and agreed on what should happen next. Once the call ends, however, they still need to stop and write everything down.

On a busy day, details can easily be forgotten. Incomplete notes may also mean that the next conversation begins without the full picture. Sales2Text helps capture that context by transcribing the conversation and organising the important information.

The salesperson can move on to the next customer knowing that the details of the previous call have not disappeared. Instead of trying to reconstruct the conversation from memory, they have a clearer record of what was discussed and what needs to happen next.

The same technology can help managers understand what is happening across a much larger number of conversations. In the past, teams could review around 1% of sales calls because listening to every conversation simply was not possible.

Sales2Text can review calls using a shared sales scorecard. This helps teams identify useful examples, recurring problems and opportunities for better coaching. Managers can spend less time searching for a relevant call and more time helping their teams learn from it.

People remain part of the process. If a salesperson disagrees with an assessment produced by the system, they can request a human review. This creates a feedback cycle in which the technology helps organise information, while people provide context and make the final judgement.

The purpose is not to turn every conversation into a number. It is to preserve useful context and make it easier for teams to learn from the conversations they are already having.

Built by the people who understand the problem

Repetitive work looks different in every team. A data analyst may spend hours checking numbers and preparing a familiar dashboard. A legal team may need to search through a long document for clauses that deserve closer attention. A customer support specialist may have to piece together the context behind a familiar question before they can begin to help.

The people doing this work every day often understand the problem better than anyone else. That is the thinking behind our Agentic Workforce, which gives OLXers the tools to build AI agents for problems they encounter in their own work.

Close to 3,000 agents have already been created, each responding to a problem identified by the people closest to it. These agents support different teams and tasks, but they share the same starting point: a real piece of work that could be made simpler.

A data agent can connect information, check important metrics and prepare a summary before a meeting. This gives the analyst more time to explain what changed, investigate why it happened and help others understand what it means. A legal agent can review a document, identify important clauses and highlight areas that may need attention. The legal expert still controls the assessment and decides what action to take, but they do not have to begin every review by searching manually through every page.

A support agent can summarise a customer’s issue and suggest possible next steps. This gives the specialist more space for conversations that require empathy, careful judgement or a deeper understanding of the customer’s situation.

These agents are not there to make every decision. Their role is to reduce the effort required to find information, organise context and prepare the next step. People remain responsible for understanding the situation and deciding what should happen.

Useful does not always look extraordinary

None of these examples sounds like science fiction, and that is part of what makes them useful. This is how we approach AI at OLX: we begin with a real problem and look for a practical way to make it easier.

Sometimes, the result is a form you no longer need to complete. It may be a call you no longer need to summarise or a spreadsheet that is ready before the meeting starts. It may also be a tool created by an OLXer who understood exactly where their team was losing time.

On their own, these moments may seem small. When they happen across thousands of people, customers and working days, they begin to change the experience of work.

The benefit is not only speed. It can mean more attention for customers, more space to think and more time for conversations and decisions that need experience, creativity, care and judgement.

We often ask what AI will make possible, but it may be just as useful to ask what it will make unnecessary. For many people, the best answer may be the work they never wanted to spend so much time doing in the first place.