See how we help enterprises turn AI into measurable business results: from recommendation engines to data infrastructure

Home /

Contacts

Recommendation engine for a major grocery retailer


Personalized product recommendations across website and mobile app, powered by ML models trained on customer behavior

Results:

+25% conversion

4× CTR

+18% avg order

Industry: Retail/E-commerce

AI-powered first-line tech support automation


AI copilot that classifies tickets, searches the knowledge base, and drafts responses for operators

Results:

31% faster time to resolution

24/7 with no headcount growth

100% coverage of all ticket sources

$300 average drop in op-cost per ticket

Industry: Cross-industry

Omnichannel AI chat operator


AI chat platform unifying Telegram, Instagram, WhatsApp, and web chat with 24/7 automated responses

Results:

-70% load on human operators

≤15s average response time

35% leads from chats

Industry: SaaS/Cross-industry

TextToSQL, natural language data access

AI agent that lets business users query company data in plain language: no SQL required, with result validation and source tracing

Results:

90% accuracy when retrieving the right data

5× faster access

-70% analyst load

Industry: Cross-industry

AIDoc - corporate intelligence over documents

Conversational access to corporate knowledge base with source-verified answers and on-premise deployment

Results:

Faster knowledge access

Lower search costs

On-premise ready

Industry: Cross-industry

Multi-channel media planner connecting ad spend to store traffic and conversions: budget optimization across digital, print, and local channels

Results:

+13% revenue impact

1500 stores

3 forecast horizons

Industry: Retail/Supply Chain

Demand forecasting
for a retail network

Modern data platform replacing legacy infrastructure and separating analytics from production systems for speed and reliability

Results:

-75% report prep time

-40% manual costs

10× faster queries

Industry: Retail

Enterprise data warehouse modernization

Dynamic pricing engine
for self-storage


Revenue management tool replacing manual pricing with data-driven rate optimization across 85+ US facilities

Results:

ML-driven pricing

Competitor monitoring

Demand forecasting

Industry: Real Estate · Self-Storage

AI agent that lets business users query company data in plain language: no SQL required, with result validation and source tracing

Results:

+227% margin growth

+85% MAU

60% retention

Industry: Retail/E-commerce

ML-Powered loyalty program relaunch (RecSys)

Support insights discovery for an online gaming platform

ML classification and anomaly detection on player support tickets, replacing manual tagging with structured analytics

Results:

AI classification

Real-time anomaly alerts

Industry: Online Gaming

AI screening of children's speech

ML classifier on short audio clips of children's speech, auto-flagging cases that need a speech therapist


Results:

90%+ accuracy

↓ screening load

12K+ students

Industry: EdTech / HealthTech

increase in purchase conversion


click-through rate on recommendation widgets

growth in average order value


Industry: Retail/E-commerce

1 500 stores

400K  daily transactions

47% online sales

The retailer faced a classic e-commerce scaling problem: as the product catalog grew, customers increasingly struggled to navigate the assortment. Conversion rates remained flat despite growing traffic, and average order values plateaued. The existing product suggestions were rule-based and generic, meaning the same recommendations for everyone, regardless of purchase history or preferences.

Client:

One of the largest supermarket chains

Challenge

Solutions used:

Recommender systems, Data Science & ML

Zentavor built a personalized recommendation engine for the client's website and mobile app. The system learns from each customer's behavior: what they buy, browse, and add to cart and suggests products tailored specifically to them in real time.

Recommendations work on three levels: products picked for you personally, items that go well with what you're already looking at, and smart ranking that balances relevance with business goals like margin. The models were validated through A/B testing before full rollout.

Solution

Results

reduction in overall time-to-resolution

average cost reduction per ticket

coverage across all ticket sources

operation without headcount growth

Industry: Cross-industry (applicable to retail, telecom, banking, insurance)

Data engineers were drowning in business requests. Up to 70% of the team's time went on routine, repetitive tickets — questions about data definitions, pipeline status, report fixes. Answer quality varied by shift; overnight there was no coverage at all. Hiring more engineers was slow and expensive, the team needed a systemic way to scale service quality.

Client:

Large enterprise data engineering team handling high volumes of internal business requests

Challenge

Solutions used:

AI Agents, LLM, Support Gen AI

Zentavor built an AI copilot into the team's existing tech-support ticketing system. When a request arrives:

1. The ticket is auto-classified by type and priority
2. The system analyzes context and searches the knowledge base and prior tickets;
3. A draft response is prepared with links to docs and runbooks;
4. The engineer reviews or edits the reply, and the system learns from every correction.

Solution

Results

load on human operators


average response time

leads from chats

automated support

Industry: SaaS / Cross-industry

Businesses receive customer messages across Telegram, Instagram, WhatsApp, Messenger, and web chat but there's no single system to handle them all. Owners either hire separate operators for each channel (expensive, inconsistent) or miss messages and lose sales. Response times are slow, there's no unified analytics, and scaling means hiring more people.

Client:

Challenge

Solutions used:

AI Agents, LLM

Zentavor built an omnichannel AI chat platform that brings all messaging channels into one window. The AI bot answers customer questions 24/7 using a company-specific knowledge base: it handles inquiries, books appointments, upsells, and qualifies leads — all within one conversation. When the case is too complex, the bot hands it off to a live operator with full context preserved. Businesses set up their knowledge base by importing documents, website content, and FAQs. Conversation scenarios are configured without code.

Solution

Results

Industry: Cross-industry (finance, insurance, logistics, retail, legal, public sector

Enterprises sit on massive volumes of internal documents: regulations, contracts, policies, knowledge bases, technical documentation but employees spend hours searching through them manually. A lawyer reads 200 pages of a contract. An analyst hunts for a number across five reports. Teams ask the same questions repeatedly, and answers depend on who you ask.

Client:

Zentavor product

Challenge

Solutions used:

LLM, AI Agents, Smart AI Search

Zentavor built AIDoc which is an intelligent system that lets employees have a conversation with their company's documents. Instead of searching and reading, you ask a question in plain language and get an answer with references to the exact source. The system builds a corporate knowledge base using RAG and agent-based retrieval, integrates with existing CRM, ERP, and document archives, and can be deployed fully on-premise so sensitive data never leaves the company's infrastructure.

AIDoc dramatically reduces the time employees spend searching for information: what used to take hours of reading through documents now takes seconds. This lowers operational costs across departments, ensures consistent and source-verified answers regardless of who's asking, and scales across teams without adding headcount. The on-premise deployment option keeps sensitive corporate data fully within the company's infrastructure.

AIDoc is designed for high-stakes environments where accuracy matters, the system is built to minimize hallucinations and always cite its sources.

Solution

Results

Industry: Real Estate/Self-Storage

The client operates 85+ self-storage facilities across the United States. As the portfolio grew, the team recognized an opportunity to take direct control of their revenue strategy. Pricing had been managed by a third-party operator, and the client wanted full visibility into rate-setting, the ability to react to local competitive dynamics in real time, and a data-driven framework for optimizing revenue per available square foot across all locations.

Client:

Major US self-storage developer and operator

Challenge

Solutions used:

AI-prediction, Data Science & ML, BI Systems

Zentavor is building a Revenue Management Tool: an ML-powered platform that replaces manual pricing with data-driven rate optimization across the client's entire portfolio. The tool automates competitive monitoring, sets optimal rates per unit type and location, and forecasts demand at the facility level to maximize revenue.

Project is in active development. The platform is designed to increase revenue per available square foot by giving the client direct, data-driven control over their pricing strategy for the first time.

Solution

Results

margin growth

monthly active users

purchase frequency

average check size

total subscribers

retention rate in the loyalty program

Industry: Retail/E-commerce

1 500 stores

The retailer's subscription-based loyalty program ("Abonement") had been paused due to underperformance. The original mechanics relied on static rules: predefined product lists, blanket discounts which failed to drive meaningful engagement. The business needed to relaunch the program on an entirely new foundation to justify the investment.

Client:

Same major grocery retailer

Challenge

Solutions used:

Recommender systems, Customer analytics, AI-prediction

Zentavor rebuilt the loyalty program from scratch using ML-based personalization. Instead of one-size-fits-all offers, the system now generates individual product selections for each subscriber daily, using three specialized models: one for product newcomers relevant to the customer, one for previously purchased items likely to be repurchased, and one for discovery — products the customer hasn't tried but is likely to enjoy based on behavioral similarity to other shoppers.

Version 1
We replaced algorithmic rules with ML recommendations, delivering +2% revenue per user group (validated in a 6-week A/B test).

Version 2
We introduced the three-model architecture, adding another +1% — for a cumulative +3% revenue impact.

Retail network:

Solution

Results

Case: ML-powered loyalty program

Case: ML evolution in loyalty program

Total subscribers

MAU

Purchase frequency

Average check size

60%

Margin

RECSYS LAUNCH

Version 1

47% online sales (400K online transactions daily)

1,500 stores

Retention rate in loyalty program

-3%

Cumulative revenue impact

Impact on business metrics

Personalized recommendations: +2% revenue per user group

Revenue per user group

Test duration after recommendation launch — 6 weeks

Approach: algorithm-based recommendations

YoY, %

Algorithm-based recs

ML-based recs

Version 2

3 specialized models for personalized recommendations:

Revenue per user group

Test duration after recommendation launch — 6 weeks

ML-based recs

RECSYS LAUNCH

Improved ML-based recs

cumulative revenue impact (treatment vs control group)

in stockouts and excess inventory

into the existing supply chain and replenishment workflows

Retail network, 1,500 stores, 47% online sales (400K transactions daily)

Improve demand forecasting accuracy to optimize inventory and reduce losses

Cumulative revenue impact

Industry: Retail/Supply Chain

1 500 stores

400K daily online transactions

The retailer's demand planning relied on a mix of spreadsheets and basic statistical models. Forecast accuracy was insufficient to prevent stockouts in high-demand periods or avoid excess inventory in slow seasons. The planning team couldn't account for complex seasonality, promotional cannibalization, or location-specific demand patterns at scale.

Client:

Retail chain

Challenge

Solutions used:

AI-prediction, Data Science & ML

Zentavor built an integrated ML platform for multi-horizon demand forecasting: short-term (daily/weekly for replenishment), medium-term (monthly for inventory planning), and long-term (quarterly for assortment strategy). The system generates hierarchical forecasts across SKUs, categories, and individual store locations, incorporating external signals such as weather, local events, and promotional calendars.

The forecasting models were deployed alongside the retailer's existing planning process as an A/B test: a treatment group of stores used Zentavor's forecasts, while a control group continued with the client's existing method. Revenue performance was tracked over several months.

Solution

Results

Case: Demand forecasting for a retail network

Our forecast

Revenue

Revenue (moving average)

Client forecast

Gross Margin

Profit

Profit (moving average)

reduction in reporting preparation time

execution of analytical queries

lower costs for manual data processing

from analytical workloads

for the entire organization

Industry: Retail

All company data lived on numerous Microsoft SQL Server instances with no unified schema. The situation created four critical problems: the architecture couldn't scale with growing data volumes; production systems were at risk because analysts queried live databases directly; large analytical queries timed out or ran unacceptably slowly; and calculation errors were common due to the absence of a standardized data model.

Client:

Large retail chain

Challenge

Solutions used:

Corporate Data Warehouses, BI Systems

Zentavor replaced the legacy SQL Server infrastructure with a modern data platform built for scale. The new architecture separates analytical workloads from production systems, so analysts no longer query live databases. All data from CRM, ERP, POS, and other systems flows into a centralized warehouse with a unified data model and automated quality checks.

The platform combines several specialized engines: Greenplum for heavy analytics, ClickHouse for fast dashboards and BI, and S3 + Trino for archival data — giving each team the right tool for their workload without duplicating data.

Solution

Results

accuracy when retrieving the right data


data access for business users

reduction in analyst workload on routine queries

Industry: Cross-industry

Business users across the organization depended on the analytics team for every data request  from simple metrics lookups to complex cross-table reports. This created a bottleneck: analysts spent most of their time fielding ad-hoc queries instead of doing deep analysis, and business users waited days for answers that should take minutes.

Client:

Enterprise with complex data landscape

Challenge

Solutions used:

LLM, AI Agents, BI Systems

Zentavor deployed TextToSQL — an AI agent that lets business users ask questions about company data in plain language, without knowing SQL. You type something like "What was the revenue by region last quarter?" and the system figures out where to find the data, pulls the answer, and shows it with a visualization. It understands business terminology, validates results before showing them, and explains where the numbers came from.

Solution

Results

improvement in ROAS (return on ad spend)

additional store traffic

payback period

Industry: Retail

The retailer managed advertising campaigns across multiple channels and geographies, but lacked tools to optimize budget allocation, measure cross-channel performance, or connect ad spend to in-store traffic and conversions. Campaigns were planned manually, with limited feedback loops.

Client:

Multi-format retail chain

Challenge

Solutions used:

Customer Analytics, AI-prediction, BI Systems

Zentavor built an integrated platform for planning, launching, optimizing, and analyzing advertising campaigns. Key components include a multi-channel media planner for budget optimization across digital, print, and local channels, and a performance tracking system that connects ad spend to store traffic and sales.

Solution

Results

accuracy in detecting speech issues


drop in manual screening workload


students on the platform, beyond manual reach

12k+ students

B2C speech therapy

Industry: EdTech / HealthTech

Speech therapists on the platform manually listened to and assessed thousands of children's recordings to identify cases that required professional intervention. The process was slow and became a bottleneck: instead of working with children, therapists spent hours on primary screening.

Client:

Logopotam, online speech therapy platform

Challenge

Solutions used:

ML, Audio classification, Data Science & ML

Zentavor built an ML classifier that analyzes short audio clips of children's speech and automatically detects cases that need a speech therapist. The system handles the primary assessment, filtering out clean recordings and surfacing the ones that need professional attention.

Solution

Results

Industry: Online Gaming

The client's support team handled thousands of chats daily and needed better visibility into what was happening across all conversations. Existing tagging covered the basics but didn't provide the granularity or consistency needed for trend analysis. Managers wanted to understand patterns: what are the most common topics this week, how do issues differ across user segments, are there emerging problems worth escalating early?

Client:

International online gaming platform

Challenge

Solutions used:

AI Agents, LLM, Customer Analytics

Zentavor built an analytics layer that sits between the support channels and the client's dashboards. Every incoming chat is automatically classified by topic, enriched with user context (segment, value, activity status), and fed into a structured pipeline. The system also monitors ticket volumes and flags anomalies — for example, if a specific topic suddenly grows. All data flows into Tableau-ready format with pre-built views by topic, user segment, and time period.

Project is in active development. The platform is designed to increase revenue per available square foot by giving the client direct, data-driven control over their pricing strategy for the first time.

Solution

Results

Tell us about your use case — we’ll propose architecture, timeline, and a delivery model