Lemons Development began adopting artificial intelligence years before the AI boom. The company’s founder, Ira Melimonadze, tells Forbes how an AI manager helps a 50-person team operate at the scale of a 200-person company while keeping human error to a minimum.
By Anton Kokaia
Lemons offers its partners – mainly construction companies – a full package of services, including web development, marketing, sales, and social media management. What path did your company take before establishing itself in this niche?
At the first stage, when I founded Lemons, we worked exclusively in web technologies. Within about two years we entered the international market, supplying our web products there.
This was 2011–13, when CRM systems were actively starting to enter the Georgian market. Even though more than 300 web studios were registered in the public registry at the time, only 3 or 4 of us were capable of building projects of that scale. We gained a great deal of experience. We built several internet banking platforms, applications, and school systems.
This gave us extensive experience across various directions, which created the need to add marketing services. One thing was technically building a website or application, and another was delivering it properly to the customer. Since I and several of my employees already had marketing experience, we decided to work in that direction as well. That’s how marketing was introduced at Lemons.
Meanwhile, social media was actively taking hold in the Georgian market. The real estate market picked up at exactly the same time. Construction companies started using social media. When we were building their websites, they would ask us to manage their social media as well.
Even before that, when I founded the company in 2011, I was already actively researching artificial intelligence in social media and consumer behavior in general. I had several pages and was studying what Georgian users responded to. With that experience, we took on our first major project – Lisi Green Urban. It was a full contract: technical services, social media, sales. The sales component included monitoring sales, creating content for sales, lead generation, and call monitoring. In fact, that’s where we started incorporating social media into our services.
In 2018, we proactively moved into the construction industry with a full outsourcing service. This covered every process from naming and preparing architectural documents to moving in, including accounting and legal work.
This was already the period of the real estate sales boom. Real estate prices were rising by almost $50 every quarter, and rising prices drove up customer activity. We faced a serious challenge. Basic experience with websites, social media, and even sales was far too limited compared to the ambitious goals we had set. Therefore, we had to use a great many web technologies to find customers.
You mentioned artificial intelligence. It turns out you started working on it years before the AI boom?
Back then, only a handful of people knew anything about artificial intelligence. I remember that if you walked into a BOSS store abroad, for example, BOSS ads would then start showing up on your social media. It was a real sensation if you told anyone about it. That was the first message for me that something was happening in this industry, and I started learning. From then on, artificial intelligence became a subject of research and interest for me.
When we entered the real estate market, I studied every CRM system available. None of them met even the minimum requirements of this field. Therefore, I started thinking that we should build our own CRM system. The first attempt ended in failure, though we gained a lot of experience from it. We worked at it for a year and a half and got tangled up. We were also developing as a company, and by the time we got from writing the logic to its implementation, our business processes had already changed.
After a year and a half, we decided to start from scratch despite the costs already incurred. This was around the end of the pandemic. Anything that might look like a setback has to be used as experience if you want to move forward quickly. We said it was fine that we’d made mistakes – instead, we’d accumulated so much technical knowledge that we rebuilt, in a month and a half, everything we had struggled with for a year and a half, and launched the first version of the CRM at the company. This eliminated the risk of double-selling an apartment – before that, a mistake could slip through in Excel. The main reason we wanted a CRM system was to make sure errors like that didn’t happen.
Eventually, we arrived at a system where today, everything from finding a buyer through to move-in is digital. If any error is possible, it would only come from the servers going down, and we already cover that with backups made at the end of every day – the data is stored on several separate servers and computers.
When the AI boom began, we were fully ready to build our CRM system on AI. We use it, for example, for customer market research. The AI agent we built performs a full analysis of five years’ worth of data – how many inquiries there are per month, what each person wants, what’s selling and how. This is extremely valuable to us, especially since our marketing is already very refined.
This is how we created products such as buying with no down payment – we were among the first to offer it. Also, lowering the monthly payment: recently, $500 was the minimum instalment with which you could buy an apartment. We managed to bring that down to ₾550, because a very large segment of customers were asking for that kind of product.
How does the use of AI show up in the organization’s day-to-day operations?
If Lemons today consists of 50–55 employees, I can say that it operates at the scale of a 150–200-person company. That doesn’t mean at all that our employees are on the verge of burnout – that’s entirely AI’s doing. Before, it took me three days alone for market research alone, processing a database of 200,000 and then figuring out what kind of content to create. Today I do that at the push of a single button.
Also, say we need to circulate new information among employees – for instance, there’s a change in a project. AI generation creates the content; of course, we step in by hand for whatever needs adjusting. You press one button, a survey is created, and it’s sent automatically to every employee at 10 a.m. As a result, everyone is informed before work starts – you can’t start work otherwise without confirming that information. This rules out an employee missing something: when they fill out the questionnaire, if they get a lot of wrong answers, it immediately goes to the manager, flagging that the employee needs help. This isn’t a system of control or punishment – it’s a monitoring system. We need to be sure that the flood of information, which changes almost every 2–3 days, reaches each employee correctly.
We also have a call evaluation system. We upload recordings of calls made by the call center, and an AI model, using a prompt we’ve given, evaluates the entire conversation. It gives employees positive recommendations and advice – telling them what they’re doing right and should keep doing, as well as recommending how to improve their communication and what to take into account based on the call with the customer.
This evaluation system isn’t punitive either, in fact, it’s incentivizing. If you have better scores, you get a bigger bonus. And it’s not just an evaluation – it gives you the kind of advice and recommendations that my partner and I would otherwise have had to spend from morning to night talking through with every single employee.
How acceptable did your employees find it to have human resources replaced by an AI agent?
We directed the AI to deliver this information to employees in the form of advice. When we first rolled it out, it caused a huge protest in the team. Our bot was a bit aggressive, and as a result, the team stalled for three days. Just when we thought we’d get a great result, the results suddenly collapsed to zero.
What was the reason for that? If I or any manager gave feedback to an employee, they could normally respond to us and voice any internal objection right there. That’s why we’re now working to give employees that same ability of feedback to the AI – to tell it they disagree and continue the conversation. We’ve already built in an interactive AI for that.
Another area where we’ve introduced AI is task management. Previously, someone – or you yourself – had to write out the tasks you needed to carry out. But because the whole process, from acquiring a lead to move-in, is digital, we decided the task manager should be automated.
Let’s say a lead comes in – your task list shows you need to call them. If, after the call, you mark their status as “coming to a meeting,” it turns into a task showing you have a meeting. If they come to the meeting and you enter into the system that the buyer will purchase the apartment, you fill out the contract template, and it automatically goes to the legal team’s task manager, so they have contracts to review and confirm. As soon as they confirm it, it goes to the accountant, who has an invoice to issue, and so on through to the handover act. As a result, a huge amount of resources is saved – 80% of the tasks for our 50 employees are already pre-written.
Plus, we’ve built in a function in the task manager where, for instance, if I have the same task to do once a week on a recurring basis, I can register it so that for the next year, every Wednesday I have one such video to make.
The vacation, salary, and bonus system is handled entirely by AI. It calculates the quality of work, which is measured in numbers, and by the end of this year we’ll also build in qualitative data – how well someone wrote a comment, how accurately they described a customer, and so on. Employees can view the bonus system themselves – the dashboard shows how much they could earn depending on which results they achieve.
This has essentially eliminated the need for management in the call. We don’t have a call manager. It’s a 12-person team with no middle or top manager – it reports directly to the administrative unit, and we effectively don’t need auditing.
The only thing we still need is human support. We hired a trainer, and if an employee doesn’t want to talk to the AI, they can switch over, go to the trainer, and discuss how to handle the situation. Even that stems purely from psychological factors, though I think in a few months the need for it will disappear, too.
We’ve talked about how AI helps you run the company’s operations faster and more efficiently. But what does all this mean for your partners and their customers?
It’s important for our partners to know that whatever they’re doing today, if they decide to simplify a process, that doesn’t mean things get cheaper. It might make you think it’s better to bring back human resources, because using AI like this in a company runs to such a high budget. But it’s so much comfort, and so much resource is saved, that the moment your employees start using it, you’ll notice they’re finally found relief. They gain mechanisms that let their minds unwind and actually think, though on the other hand, there’s a risk of swinging to the other extreme and starting to decline.
Companies need to know that things like this exist. I object to anyone claiming they’ll manage to set up an AI system in two days. Because that’s damaging – people get disappointed, switch it off, and never come back to it. It shouldn’t be treated as though a university could be replaced by a three-day masterclass. No – you can’t replace people by uploading three days’ worth of information.
From our side, this is very simple – technically we’ve been through this so many times already. But from the companies’ side, a great deal of human resource has to be dedicated to it, and they need to be prepared for that. They also need to get past that disappointment: when I say now that I got something done with the push of one button, in reality, we pressed a great many buttons to arrive at work that’s done with the push of one button.
As for end customers, their biggest problem – issues caused by human error, such as mistakes in contracts – has been reduced to a minimum through AI.
Digitized processes are especially convenient for emigrants. In a dedicated app they receive information about what stage construction has reached, complete with photo and video material. They can see their payments and transfers, which instantly show up in the app.
On top of that, if an operator isn’t available at a given moment, customers start communicating with the AI assistant and get the same answers a human would give them. Therefore, getting information is faster for them.
Customers should also feel reassured because when a company uses technology this extensively and takes on this much expense to keep the customer as protected as possible, that shows the customer truly is the company’s core value. We wouldn’t need any of this technological equipment if we didn’t have customers.
Let’s touch on future plans too. What will your next step be in terms of AI transformation?
Our future plans include adding a voice AI assistant. We’ve already implemented a dialer. We already have an outbound-call dialer – it makes calls, presents information, and if the person is interested, the call gets connected to an operator, but they’ll still be speaking to a human operator. Now what we want is to replace the operator with an AI assistant. We’re ready to implement this, but we’re not launching it yet. The market is still not ready for that.
I think AI will take on every human function except the act of thinking itself. AI can’t do the intellectual and creative part without you. It’s important to know your field, know your customer psyche, and teach the bot exactly what you want. And teaching is the hardest part of all.
If we want to use our capabilities properly, there needs to be a lot of meetings. We need to talk openly about what experience each of us has had, whether it turned out good or bad. That’s why I’ve shared that sometimes some things don’t pan out.
At the same time, we need to try to have a code of ethics. We don’t contact a customer without their consent. The customer’s willingness must exist and must be respected, despite of technological novelty or the possibility of disregarding their interests. From an ethical standpoint, the customer is and will always be number one priority, no matter what sort of AI technology gets implemented.
S
