Speech analytics for banks is an AI platform that transcribes 100% of contact-centre calls and automatically checks every conversation for script adherence, service quality and risk. Manual quality assurance covers less than 5% of conversations, according to McKinsey, so the remaining calls go unchecked, and the compliance risks go with them. For a bank in Uzbekistan, two more conditions matter: recordings and transcripts stay inside the bank's perimeter (on-prem), and the handling of recordings is designed with Law No. ZRU-547 “On Personal Data” in mind. The ZRU-1125 amendment of 26 March 2026 narrowed mandatory localisation to biometric, genetic and telecom subscriber data, yet for CIS banks in-country storage and on-prem remain the default requirement. Zukko.AI is an AI agent platform for business: its speech analytics deploys inside the bank's perimeter and keeps call data in Uzbekistan, and an enterprise rollout with a security review takes up to 30 days. Below: the law, data localisation and why banks choose on-premise.
speech analytics and AI call analysis
What speech analytics for banks is
Speech analytics for a bank is an AI platform that turns 100% of contact-centre calls into text and automatically checks each conversation for quality, script adherence and risk. Manual review manages to score less than 5% of calls (McKinsey estimate). The platform analyses the rest.
How is that different from ordinary call recording? Recording simply stores the audio. Analytics reads its meaning. For a bank, that comes down to specific things:
- Transcription of the conversation into text in Uzbek, Russian and English.
- Script control: whether the agent greeted the customer, gave the mandatory recording notice and offered the product according to procedure.
- Risk markers: promises of returns, mentions of fees, rudeness, sensitive data disclosed during the conversation.
- Tone and emotion: customer irritation, raised voices, signs of a complaint to come.
In banking, speech analytics has a layer that retail lacks: the high cost of an agent's mistake. A loan condition stated incorrectly is a potential claim from the regulator, not just a lost sale. That is why banks treat analytics as part of compliance rather than as a marketing tool.
The contact centre is not the only team that uses the results. The table shows which banking tasks speech analytics covers and who in the bank needs the output.
| Banking task | What speech analytics shows | Who in the bank uses the result |
|---|---|---|
| Contact centre: service quality | A score for every call on a single scale, the tone of the conversation, reasons for repeat contacts | Head of the contact centre, supervisors |
| Selling products by phone | Whether the product was offered according to procedure, which objection ends the conversation, which wording leads to an application | Head of retail sales |
| Collections (soft collection) | Whether the debtor conversation scenario was followed, whether the tone was appropriate, whether a promise to pay was recorded | Collections unit |
| Script compliance | Whether the mandatory phrases were said (recording notice, product terms), plus stop phrases and promises that must not be made | Compliance and internal control |
| Agent training | Typical mistakes of each agent, examples of strong and weak calls for review | Trainers and the training team |
how speech analytics monitors 100% of calls
Why a bank needs on-premise speech analytics
On-premise means the platform is deployed on the bank's own servers and call recordings never leave its perimeter. This is critical because, even after the 2026 relaxation, some personal data in Uzbekistan remains subject to mandatory localisation (gazeta.uz, 2026). If a customer's identity is established by voice, that recording counts as biometric data.
Why do banks in particular choose on-prem over the cloud? There are several reasons, and they add up.
- Regulatory pressure. The financial sector is regulated more tightly than the average business; “data inside the country” is a default expectation here, not an option.
- Security perimeter. Recordings of conversations that mention cards, balances and passports must not leave the bank's infrastructure.
- Audit and access. On-prem gives the bank full control: who accessed which recording and when, all inside its own audit trail.
- Risk management. The fewer external places sensitive data is stored, the smaller the surface for a leak.
It is worth being precise about two terms that banks ask about first. On-prem is the bank's own hardware in its own data centre. Private cloud is a dedicated, isolated environment hosted by the provider. For the financial sector in Uzbekistan, on-prem remains the gold standard where call recordings are concerned.
Is it legal to record customer calls in Uzbekistan
Uzbekistan has no separate “call recording law”. Recording is governed by the personal data law, Law No. ZRU-547 “On Personal Data”, which came into force on 1 October 2019. The voice of an identifiable person is personal data. Processing it requires the customer's consent or another lawful basis.
How do banks obtain consent in practice? The most common mechanism is an opening notice.
- The agent or an automated announcement says at the start of the call: “This call is being recorded.”
- By continuing the conversation, the customer gives implied consent.
- The notice is captured in the recording itself and can be verified by analytics.
Here speech analytics works for the bank twice over. First, it helps keep the procedure in place by automatically checking that the mandatory recording notice was actually given in every call. Second, in a dispute the recording and its transcript become evidence. That is the platform's role: an aid to compliance, not a substitute for it.
An important caveat: this is not legal advice. The specific legal bases for processing, the wording of notices and the retention policy are agreed by the bank with its own legal team and compliance officer.
Where call data must be stored by law
Until 2026 the law required the personal data of Uzbekistan's citizens to be stored on servers inside the country. The ZRU-1125 amendment of 26 March 2026 relaxed this: mandatory localisation now applies to biometric data, genetic data and telecom subscriber data (gazeta.uz, 2026). Other personal data may be processed abroad with adequate protection.
You might expect the relaxation to weaken the “server in Uzbekistan” argument. For banks, in practice, it is the opposite. Here is why “data in the country” remains a pressing requirement:
- Voice biometrics. If a bank identifies a customer by voice, that voiceprint stays in the mandatory localisation category.
- Telecom subscriber data. This category also remains under mandatory localisation.
- The industry bar. Banks and regulated organisations require in-country storage and on-prem regardless of the general relaxation; it is their internal risk standard.
A telling detail about the market: the 2026 relaxation helps e-commerce and small business, but changes almost nothing for the financial sector. A bank building a contact centre for the next five years will not put customer data in a foreign cloud to save on servers. So “data in Uzbekistan plus on-prem” remains the durable choice for banking speech analytics.
One clarification, so that two cases are not confused. A recording used only for transcription and quality monitoring, and not to establish identity, generally counts as ordinary personal data. A voice becomes biometric data when it is used to identify a person. Keeping recordings in the country and inside the bank's perimeter settles the question in both cases. For a detailed breakdown, see the article on ZRU-547 and personal data localisation; the text of the law is on lex.uz.
As for violations: the legislation provides for administrative and criminal liability for unlawful processing of personal data. For a bank, the reputational risk often weighs more than the sanction itself.
Cloud or on-prem: which a bank should choose
For a bank in Uzbekistan the answer is almost always the same: on-prem with data stored inside the country. The cloud is cheaper and quicker to launch, but it does not meet the regulatory expectations of the financial sector. Below is a straightforward comparison on the parameters banks check first.
| Parameter | Cloud (vendor cloud) | On-prem + data in Uzbekistan |
|---|---|---|
| Where recordings sit | The provider's data centre, often abroad | The bank's servers, inside the country |
| Voice biometrics | Localisation risk | Inside the bank's perimeter |
| Access control | Shared between provider and bank | Entirely with the bank, its own audit trail |
| Time to launch | Days | Up to 30 days (enterprise rollout) |
| Capital expenditure | Lower | Higher (own hardware) |
| Fit with regulator expectations | Partial | High |
| Best suited to | Small business, e-commerce, retail | Banks, fintech, telecom |
The conclusion from the table is simple. Small businesses and e-commerce gain from the cloud: it is fast, inexpensive and, since the 2026 relaxation, permissible. A bank needs the platform where its data lives, under its own audit and access control. So in banking the cloud is usually kept for testing a hypothesis, while production runs on-prem.
How Zukko speech analytics helps banks with compliance
Zukko is an AI speech analytics platform that is deployed on-prem and stores call data in Uzbekistan. It puts 100% of calls under automated quality monitoring and flags a script violation shortly after the call ends. Enterprise onboarding takes up to 30 days, including the security review.
What the platform gives a bank as an aid to compliance:
- Data isolation. Recordings and transcripts stay inside the bank's perimeter; there are no external storage locations.
- Personal data masking. Phone numbers, names and other sensitive fields are hidden in the analytics interface, so the QA reviewer sees the substance of the conversation but not the raw personal data.
- Data in Uzbekistan. In-country storage in line with the regulatory expectations of the financial sector.
- On-prem deployment. The platform runs on the bank's infrastructure, under its audit trail and access control.
- 100% of calls under QA. Every conversation is checked for script, tone and risk: the whole flow, not a sample.
- Fast detection of violations. A missed recording notice, a promise of returns or rudeness surfaces shortly after the call, not in a sample check a week later.
To be clear: Zukko does not provide legal guarantees and does not replace the bank's lawyer or compliance officer. The platform is a tool that makes adherence to procedures observable and verifiable. The decision on the legal bases for processing always rests with the bank.
CIS banks already apply similar logic: speech analytics becomes part of the compliance perimeter rather than a standalone marketing gadget. The driver is plain to see: Uzbekistan's banks are publicly compared on contact-centre quality (Kursiv, June 2025), and the pressure on QA keeps growing.
the Zukko speech analytics module
AI employee for banks: deployment inside the bank's perimeter
how much contact-centre call analysis costs
The short version
Speech analytics closes a gap a bank cannot close by hand: manual review sees less than 5% of calls, the platform sees all 100% (McKinsey estimate). But for a bank, quality is half the job. The other half is where the recordings sit. A recording of a customer's voice is personal data under ZRU-547, and the 2026 relaxation of localisation barely touches the financial sector: industry security requirements and voice identification keep the bar at “data in Uzbekistan plus on-prem”. Zukko helps a bank keep to its procedures, with data isolation, personal data masking, in-country storage and 100% of calls under review, but it does not replace the bank's lawyers. Want to see how this is deployed inside your perimeter?
Frequently asked questions
Is it legal to record customer phone calls in Uzbekistan
Yes, provided there is a lawful basis. Uzbekistan has no separate law on call recording; recording falls under Law No. ZRU-547 “On Personal Data”, in force since 1 October 2019. A customer's voice is personal data, so consent or another lawful basis is required. In practice, banks use the opening notice “this call is being recorded” as implied consent.
Where must call recordings be stored by law
Following the ZRU-1125 amendment of 26 March 2026, mandatory localisation in Uzbekistan applies to biometric data, genetic data and telecom subscriber data (gazeta.uz, 2026). Other personal data may be processed abroad with adequate protection. A voice used to establish identity is biometric data, which is why banks keep recordings in the country and on-prem by default.
Is customer consent needed to record a conversation
Yes. Under ZRU-547, processing personal data requires consent or another lawful basis. The most common practice in banks is a notice at the start of the call, “this call is being recorded”, after which continuing the conversation is treated as implied consent. The exact wording and the legal basis are agreed by the bank with its own legal team.
Why does a bank need on-prem if localisation was relaxed in 2026
Because the ZRU-1125 relaxation did not touch biometric data or telecom subscriber data, and a voice the bank uses to establish a customer's identity counts as biometric data (gazeta.uz, 2026). On top of that, the financial sector requires in-country storage and on-prem as an internal risk standard, regardless of the general relaxation for other personal data.
Where are a bank's call recordings and transcripts stored
In an on-prem deployment, call recordings and their transcripts are stored on the bank's own servers and never leave its perimeter. The bank configures access itself: staff roles and the access log are kept inside its own environment. The hosting model, whether on-premise, a private environment or a hybrid, is chosen by the bank's information security team.
Does speech analytics handle Uzbek and Russian in the same call
Yes. Bank customers in Uzbekistan often move from Uzbek to Russian and back within one conversation. Zukko transcribes calls in Uzbek, Russian and English and detects the language automatically as the call goes on, so mixed speech is captured in full and the script and risk-marker checks work for both parts of the dialogue.
How long does it take to deploy speech analytics in a bank
An enterprise rollout with on-prem deployment takes up to 30 days, including the security review and audit-trail setup. The exact timeline depends on how ready the bank's infrastructure is, on access, and on the scope of integrations with existing telephony. A test environment on limited traffic takes less time because there is less work involved.