Sberbank: AI resolves 65% of customer inquiries in the contact center

• ИИ на линии: Сбербанк автоматизировал рассмотрение более 65% клиентских обращений — CNews, 2026-06-15
• Сбер автоматизировал рассмотрение более 65% клиентских обращений — AK&M, 2026-06
• Сбер внедрил GigaChat в контактный центр — CNews, 2024-11-07
• ИИ принёс Сберу 300 млн руб. за счёт оптимизации обработки звонков — ComNews, 2024-03-01
• Сбер сэкономит более 7 тыс. часов в месяц благодаря внедрению ИИ в контакт-центре для бизнеса — PLUSworld, 2024
• GigaChat (обучение на суперкомпьютерах Christofari / Christofari Neo) — Википедия, 2026-07-10

Background

Sberbank is Russia's largest bank with over 110 million retail customers and the developer of its own large language model, GigaChat. This is a rare configuration globally: the bank does not buy AI from a vendor but trains its own models on its own infrastructure — the Christofari and Christofari Neo supercomputers (Cloud.ru) with NVIDIA A100 GPUs.

The bank's contact center serves both retail and corporate customers across voice and text channels. At this scale support automation is not an experiment but one of the main levers of operational efficiency: fractions of a percentage point of automation translate into hundreds of operators.

The AI rollout was staged and documented in the press at every step. In 2023 the bank optimized AI routing of business-client calls — the most measurable and 'boring' scenario, which nevertheless produced the first publicly named money. From November 2024 a GigaChat operator assistant went live in the corporate-client contact center. And for Q1 2026 the bank reported the share of inquiries AI resolves with no human involvement at all — benchmarked against Frank RG market data, a rarity for the Russian market.

That sequence — routing, operator assistant, full automation — makes the Sber case a useful roadmap for any large contact center: the stages run from simple to complex, and each has its own success metric.

Problem

At the scale of the country's largest bank, routine inquiries dominate the contact center flow, and every minute of handling is multiplied by millions of dialogues. Routing calls from corporate clients was a particular pain: before optimization it took 3.5 times longer than after (per ComNews, 18 seconds post-implementation). A business customer bounced between lines is not just irritation but direct cost: every extra transfer consumes operator time.

The corporate segment is objectively harder than retail. Alexander Krushinsky, voice technology director at BSS, told ComNews that 'good' automation of an incoming line is 30–60% — and called the corporate segment one of the hardest: business clients' topics are broader, phrasing is more specific, and the cost of a wrong answer is higher.

The second pain concerned the operators themselves: during a conversation they had to search the knowledge base manually. While the operator hunts for a regulation, the customer waits on the line; answer quality depends on how quickly a person navigates the documentation. That gap — between the knowledge the bank has and the speed of accessing it mid-conversation — is exactly what an operator assistant closes.

Finally, there was the measurability problem: 'we deployed AI' is not a result. The bank needed to show impact in metrics both the business and the market understand: the share of inquiries handled without a human, answer speed, hours and rubles saved.

Solution

Sber built AI in the contact center as three sequential layers, each with its own economics.

The first layer is routing. Predictive analytics routes 87% of corporate client calls, routing accuracy rose 14 percentage points over a year to 77%, and routing time fell 3.5x — to 18 seconds (ComNews data). Speech analytics covers 99% of consultations, and by early 2024 the AI assistant in business text channels was already resolving 23% of inquiries on its own. This layer produced the first publicly named money: 300 million rubles saved in 2023.

The second layer is the GigaChat operator assistant, launched November 7, 2024 for corporate client support. The model analyzes the conversation with the entrepreneur in real time, finds the right materials in the SberHelp knowledge base, summarizes them, and offers the operator ready-made suggestions. A separate function is quality control: GigaChat scores the dialogue's politeness and professionalism with 80% accuracy and gives recommendations when service standards are violated. At launch 40% of operators had access; by the end of 2024 the bank planned to roll it out to all business-support staff in chats, then to the 0321 voice line. Sergey Lekhanov, director of Sber's Corporate Solutions Center, framed the bank's position: artificial intelligence should help business, and GigaChat in corporate client support demonstrates that clearly.

The third layer is full automation: AI resolves routine inquiries in voice and text channels on its own, with no operator involved. This is the layer behind the 65%+ figure the bank reported in 2026.

All layers share the same technological foundation: GigaChat models trained on Sber's own Christofari and Christofari Neo supercomputers (Cloud.ru) with NVIDIA A100 GPUs. For a bank of this scale, in-house training infrastructure is not about prestige but a strategic condition: customer dialogue data never leaves the perimeter, and the model can be fine-tuned to banking specifics without external dependencies.

Result

As of Q1 2026, AI resolves over 65% of customer inquiries: 66% in voice channels and 71% in chats. The market comparison sets the frame: per Frank RG's 2026 estimate, the industry average is 23% automation in voice and 67% in chats. In voice, Sber is nearly three times ahead of the market; in chats, slightly above it. 95% of calls are answered immediately, during the conversation; the remaining 5% require additional analysis (typically 2–3 days). The bank's contact center received two Frank RG awards — for the most stable operator team and the best robotic service on an incoming line.

Publicly named effects by layer: AI routing of corporate calls saved the bank 300 million rubles in 2023; AI in the business contact center saves over 7,000 operator hours per month. The GigaChat assistant added +7% to operator productivity and +2 pp to the CSI satisfaction index; operators use up to 20% of the model's suggestions (up to 45% in some areas), and dialogue quality scoring reaches 80% accuracy. The results were publicly discussed by Elena Levina, Vice President and Director of Sberbank's Customer Care Department.

Frame the numbers correctly. This is the bank's own reporting, not an independent audit; industry experts in ComNews explicitly cautioned that improved routing metrics can have multiple explanations (Evgeny Surkov, Innostage) and that quantitative indicators without customer quality assessment give an incomplete picture (Evgenia Gilenyuk, SKB Kontur). The strength of Sber's reporting is its anchor to the external Frank RG benchmark: the market comparison makes the headline number verifiable.

In our view, the Sber case is valuable above all as an architectural template: three layers with different metrics (routing — seconds and rubles; the assistant — productivity and CSI; automation — the share of inquiries without a human) are not blended into one 'AI effect' but measured separately. That transfers to any large contact center — not least because the first layer (routing) pays for itself before any LLM appears. A separate observation: the share of suggestions operators actually use (20–45%) is a rare example of an honest assistant-usefulness metric; most deployments report that suggestions exist, not whether people take them.

Lessons learned

1. The share of inquiries resolved without a human is the key contact-center automation metric. Sber reports exactly this (65%) with a market benchmark (Frank RG 23%/67%), making the number verifiable.
2. An operator assistant and full automation are different layers with different metrics: productivity (+7%) and CSI (+2 pp) for the former, automation share for the latter.
3. Starting with a narrow, measurable scenario pays off: AI routing of corporate calls saved 300M RUB before any LLM assistant existed — and funded trust in the next stages.
4. The share of suggestions operators actually use (up to 20%, up to 45% in some areas) is an honest usefulness measure — not the mere fact of deployment.
5. The corporate segment is among the hardest to automate (the industry norm for 'good' automation is 30–60%): start there with routing and an assistant, not full automation.
6. AI can control quality, not just answer: scoring dialogue politeness and professionalism at 80% accuracy turns sample-based QA into full-coverage QA.
7. In-house training infrastructure (Christofari, A100) is a strategic requirement at this scale: dialogue data never leaves the perimeter, and the model is tuned to banking specifics.

Frequently asked questions