AI Development for Business: How Not to Overpay and Get Results | Tantal

Business owners love the statistic "97% of companies are already implementing AI." But there's a second one that's less often mentioned: only 36% can prove that AI has brought in revenue. The problem isn't with the technology – it's how they choose, implement, and measure it.

The market grew fivefold in a year, but most budgets went to waste

The Russian market for generative AI reached 58 billion rubles in 2025, up from 13 billion a year earlier – a five-fold increase over 12 months (Just AI and Onside, data from Kommersant). The money poured in quickly.

According to the Generation AI report for 2025, 97% of large Russian companies are implementing or planning to implement AI – but only 36% can measure its economic effect. The increase from 2023 was just 9 percentage points. Almost two-thirds of companies spend money without knowing whether it's been returned.

The growing market doesn't guarantee that a specific project will be profitable. Budgets are allocated where AI has been implemented "just because" – without linking to the process and without measuring results. The right question for an owner is: "How not to end up in those 64% who don't see returns?"

Chatbots for business – the fastest way to enter AI automation

A chatbot has stopped being a "leave a message, we'll respond during working hours" widget. According to Tantal's AI assistant cases, up to 95% of routine inquiries are automated – humans only intervene where they're really needed.

For marketers or targetologists, it's a straightforward story about money. You pour traffic onto a landing page, applications come in around the clock, and managers respond from nine to six. Overnight leads cool off by morning. The bot qualifies an application at the moment of contact: asks for budget, service, city, filters out non-targeted ones – and passes on a warm lead with filled-in fields to sales. Conversion from click to conversation grows not because of a new budget, but because no applications are lost.

Three scenarios where a chatbot will pay off first:

• Inbound lead processing. The bot responds in seconds, collects contact information, and qualifies it before passing it on to a manager. Response speed directly affects the percentage of deals reached.
• Recurring questions. Prices, terms, conditions, availability – 24/7 without operator involvement. Support is relieved of dozens of identical dialogues per day.
• Qualification in ad campaigns. The bot separates target clients from random ones at the entrance – only those who are really ready to talk make it into sales.

According to Gartner's forecast (press release, May 2026), by the end of 2026, 40% of corporate applications will include specialized AI agents – up from less than 5% in 2025. The technology will go from "early adopter" to standard in a year. Those who set up automation now will have an advantage while their competitors are still thinking.

Custom development vs. off-the-shelf: where the line is drawn

There are two fundamentally different approaches, and confusion between them leads to wasted budgets.

The first one – "a wrapper over ChatGPT." A ready-made service where you plug in a prompt and connect it to your site. Quick, cheap, covers simple tasks: answering frequent questions, generating text, suggesting products from the catalog. Many SMBs are satisfied with this.

The second one – custom development tailored to a specific business process. Three signs that you've hit the ceiling of a typical solution:

• Integration with your systems. The bot needs to read and write in amoCRM, Bitrix24, 1C, your order base. A box can't do this – it lives separately from your data.
• Your own knowledge base. The assistant answers according to your regulations and prices, not general knowledge from the internet. This requires setting up under your data.
• Specific scenario logic. Your qualification logic, calculations, escalations. A universal bot can't cover this.

The effect's magnitude is visible in a specific case. MMK (Magnitogorsk Metallurgical Combine) received an economic effect of over 4.5 billion rubles from AI implementation – one of the largest documented cases in Russian industry (SberPro, TAdviser). The system was written under specific production processes, not taken off the shelf. Such a return is only possible when AI is integrated into real processes, not just placed alongside them.

The same "under process" principle works for computer vision – object recognition and video analysis where there's no ready-made solution on the market – and in mobile apps, when AI is integrated into a product for clients.

How much does AI development cost: what factors affect the price

The range is wide – and it depends on the scope of work under your task, not a contractor's greed.

The lower bound is a pilot chatbot for a typical scenario: one channel, a limited set of questions, simple qualification. Launched in a few weeks and costs like one to two months' worth of manager time, which it relieves. The pilot's goal is to test the hypothesis, not cover the entire process.

The upper bound is a full-fledged AI assistant with integrations into several systems, its own knowledge base, and complex logic. Here, work goes on for months.

What factors affect the price:

• Process analysis. Before writing code, the team breaks down where applications come from, what typical scenarios are, and where bottlenecks are. Skipping this step – and getting those wasted dollars.
• Integrations. Each link to an external system (CRM, 1C, telephony) is separate work. One integration or five – the difference in price is many times over.
• Setting up under your data. Knowledge base, training on your regulations, testing on real dialogues.
• Support. An AI solution isn't set and forget: scenarios are added, models retrained on new cases.

The market grew fivefold, demand for development is growing, and the average project cost is increasing. A pilot launched now will be cheaper than an analogous project a year from now – and it gives you the numbers to decide on scaling up. If an AI assistant increases conversion from application to deal, and campaigns are run under KPIs, the effect accumulates: target traffic meets a system that doesn't lose any leads. More about this combination in the Tantal performance marketing section.

Why 253 specialists on staff – it's not about presentation, but risks

A typical risk of a small contractor: the project is handled by one or two developers. One leaves – work stops. Nobody knows the context, nobody knows how the integration works, support falters. For you, this isn't just an inconvenience – it's a frozen budget and process that's already dependent on AI.

A team of 253 people eliminates dependence on one specialist. There are those who can take over tasks, there's adjacent expertise: if during implementation it turns out you need computer vision or a mobile app, you won't have to look for a new contractor. Over 10+ years in the market and 281+ implemented projects have accumulated a library of ready-made solutions – your task is rarely solved from scratch.

According to Yakovlev and Partners' assessment (AI-2025 study), AI implementation will bring Russia's economy between 7 and 13 trillion rubles by 2030, with over 60% of the effect concentrated in five industries: e-commerce, telecom, IT, construction, and healthcare. Strategic projects don't leave a team that can disintegrate mid-project.

Three signals: when it's time to hand your task over to a contractor

Signal 1: The task goes beyond one tool. While everything is solved by setting up a ready-made service – do it yourself. As soon as you need to connect several services, write your own logic, and handle non-standard scenarios – this is development, not setup.

Signal 2: Integration with multiple systems is needed. Connecting a bot to CRM, telephony, and order base so that data flows in both directions without errors – work for a team with experience in such integrations. An error here costs lost applications and corrupted data in the base.

Signal 3: No internal resource for support is available. AI solutions live as long as they're maintained. If nobody can retrain the model or add scenarios – the solution will degrade over months. A contractor covers support as a separate service.

What to prepare for the first meeting

To have a meaningful assessment from the very first conversation, and not "we'll send you a quote in two weeks" – gather four things:

• Description of the process. How an application goes from the first touch to a deal, who's involved at what stage, where time is wasted.
• Volume of applications. How many come in per day, during peak hours, on average – this is the basis for calculating returns.
• Current stack. CRM, telephony, messengers, 1C. What needs to be integrated.
• Desired KPIs. Response speed, percentage of automated applications, conversion from application to deal. Without metrics, the project will end up in those 64% who don't see returns.

With these four points, a contractor names the order of numbers and deadlines on the first call.

Where to start

You need to start with process diagnostics, not technology selection.

Tantal's team will break down your process, show where AI can give measurable effect in the first three months, and name the order of numbers before signing a contract. 253 specialists on staff, 281+ implemented projects – if the task turns out to be more complex than a chatbot, expertise for it already exists within.

Request process diagnostics – without obligations, a meaningful conversation about your task, and honest assessment of returns.