How an AI analyst answers business questions without SQL (using an online retailer as an example)

Introduction

A large online retailer of tools and equipment: more than 100,000 products, 800+ stores across the country, B2C and B2B sales. Every day, business users — category managers, marketers, logistics specialists, branch managers — ask dozens of questions about data:

"What is the margin for the DeWalt brand in the Ural region for July?"

"Which SKUs are going negative on turnover at the Novosibirsk hub?"

"Compare the conversion of the promo newsletter and the regular one over the last three campaigns."

Today, every such question requires an analyst who writes SQL, consolidates the data in Excel, and sends an answer — an hour later, or even the next day. An AI analyst changes the game.

Why this is painful

The bottleneck

At a company of this scale, the typical picture is:

• 75+ business users need data every day to make decisions.
• 5–7 analysts serve all of them. The queue for a simple report can take 4–8 hours.
• Repetitive requests — 60% of questions are variations of the same thing: "sales of X in region Y for period Z."
• Analysts burn out — they were hired to solve complex problems (forecasting, ABC analysis, clustering), but they write SELECT SUM(price) ... GROUP BY region 20 times a day.

The business loses money

• A category manager wants to understand whether it's time to hold a sale on slow-moving inventory — waits half a day for a report, and in that time a competitor has already lowered the price.
• A branch manager in Yekaterinburg sees sales declining and wants to understand the cause — the analyst is busy, the answer will come tomorrow, and the problem is getting worse.
• A marketer launches a campaign without waiting for data on the previous one — the budget is spent blindly.

What an AI analyst can do

1. A natural-language query → answer + chart

A business user asks a question as if asking a person — the agent generates the SQL itself, executes it, and visualizes it:

User: Sales of rotary hammers by region for July 2026
Agent: [Automatically builds a query to the DWH, after 3 seconds
       returns a table + a bar chart by region.
       Top 3: Moscow — 1,240 units, St. Petersburg — 890 units,
       Yekaterinburg — 670 units.]

Key feature: the agent doesn't just generate SQL, but also checks its correctness, explains the result in Russian, and suggests the next logical question.

2. Trend and anomaly analysis

• Sales dynamics — "How have sales of welding machines changed over the last 6 months? Show the chart and seasonality."
• Anomalies — "In which region did sales drop the most compared to last month?" → the agent finds the outliers itself and highlights them.
• Plan vs. actual — "Compare actual July sales with the plan for each category. Where is the biggest deviation?"

3. Product analytics (for category managers)

• Turnover — "Which SKUs in the 'Screwdrivers' category have been sitting in the warehouse for more than 90 days? Sort by descending stock."
• Margin — "Top 10 products by marginal profit in rubles for Q2."
• ABC analysis — "Break down the 'Power Tools' category into A/B/C by revenue and show the share of each group."
• Cannibalization — "Which substitute products are eating into sales of the Bosch GSR 12V?"
• Price analysis — "Compare our prices for drills with Citilink and DNS. Where are we 10%+ more expensive?"

4. Marketing analytics

• Channel effectiveness — "What is the ROI of contextual advertising broken down by category over the last 3 months?"
• Cohort analysis — "What percentage of customers who bought in January returned for a repeat purchase within 6 months?"
• A/B tests — "Compare the conversion of two versions of a product page — old vs. new."

5. Logistics and warehouse

• Warehouse stock — "Which SKUs at the Novosibirsk hub are below safety stock? Group by supplier."
• Delivery times — "Average delivery time by million-plus cities in July. Where have the times increased compared to June?"
• Returns — "Top 10 products by return rate over the last 3 months. Is there a correlation with price?"

6. B2B analytics

• Customer base — "Which B2B customers reduced their purchase volume by more than 30% this quarter?" → a reason for a manager to call.
• Average order value — "Dynamics of the B2B average order value by industry (construction, manufacturing, utilities)."

User scenarios

Scenario 1: A category manager makes a decision on stock

Manager: Which screwdrivers have been sitting in the warehouse for more than 120 days?

Agent: [After 3 seconds]
       Found 8 SKUs with turnover >120 days:
       1. Makita DF333D — 320 units, 230 days, stock worth 4.8 million ₽
2. DeWALT DCD701 — 180 pcs., 145 days, stock worth 2.3 million ₽
       ...
       Total frozen capital: 12.7 million ₽.
       Show sales dynamics for these items?

Manager: Show me. And suggest a discount to turn it over in 2 months.
Agent: Based on demand elasticity in the category, to
       sell off in 60 days a discount of 18–22% is needed.
       This will generate revenue of 9.1–9.7 million ₽ and free up
       warehouse volume. Launch the promotion KPI calculation?

Scenario 2: Branch manager looks for the cause of a decline

Branch director: Sales in Kazan fell by 15% in July.
                 What's the reason?
Agent: Analyzed data for Kazan for July 2026
       vs June 2026:
       - Overall decline: −14.8% (from 42.1 to 35.9 million ₽)
       - Main factors:
         • Drop in store traffic: −9%
           (facade renovation at Mega mall)
         • 3 hit SKUs dropped from the assortment
           (no supply from Bosch for 4 weeks)
         • Website conversion unchanged (3.2%)
       Detailed report — [link]. Show recommendations
       for recovery?

Scenario 3: Marketer evaluates promotion effectiveness

Marketer: Compare Black Friday 2025 and 2024. What worked,
            what didn't?
Agent: Black Friday 2025 vs 2024:
       - Revenue: +23% (from 180 to 221 million ₽) ✅
       - Average order value: +8% (from 9,200 to 9,950 ₽) ✅
       - E-mail newsletter conversion: −2 pp (from 8 to 6%) ❌
         Reason: the mailing base grew, but 40%
         of subscribers are inactive.
       - Most effective category: "Garden
         tools" (growth +340% driven by 50% discounts).
       Would you like a breakdown by channel and recommendations
       for 2026?

Scenario 4: Logistician checks warehouses

Logistician: Show hub utilization — where is the overflow?
Agent: Current occupancy:
       • Moscow: 92% ⚠️ (intake limit exceeded by 5 days)
       • Novosibirsk: 88%
       • Yekaterinburg: 68% — can redirect flow
         from Moscow.
       • Krasnodar: 95% 🔴 (emergency overflow —
         180 pallets with no space).
       I suggest: redirect shipments from Moscow and
       Krasnodar to Rostov-on-Don (utilization 54%) —
       storage savings ~340,000 ₽/month.
       Create a redirect task?

Scenario 5: CEO at the morning meeting

CEO: Show the top 5 categories by dynamics for the first half of the year
     and the main trends.
Agent: For H1 2026:
       🟢 Growing:
       1. Cordless tools: +34% (trend toward
          mobility, shift away from corded models)
       2. Protective equipment: +28% (tightening of
          occupational safety requirements in construction)
       3. Garden equipment: +22% (boom in suburban
          construction)
       🔴 Declining:
       4. Gas-powered tools: −11% (substitution by cordless)
       5. Hand tools: −5% (high competition
          with marketplaces)
       Monthly chart — [link]. Export
       to presentation?

Implementation economics: calculation for a retailer

Input data

Calculation of current costs

We assume that 6 analysts spend ~70% of their time on standard requests:

Of this, ~70% (8,618,400 ₽/year) is the cost of time spent on standard requests that AI can automate.

Calculation of costs with an AI analyst

The AI analyst handles 70% of requests (simple and medium: queries, aggregations, period comparisons, top-N). Complex analytics (forecasting, clustering, predictive models) remains with humans.

Final savings

Payback

Payback = 500,000 ÷ (8,183,000 ÷ 12) ≈ 0.7 months ≈ 3 weeks

Hidden savings: decision speed

The biggest effect is not in reducing the analyst headcount, but in the speed of decision-making:

With a 15% margin and monthly revenue of ~800 million ₽, every day of delay in deciding on stock reallocation or price adjustments means potential losses of hundreds of thousands of rubles.

Conservative estimate: accelerating decisions by 1 day on average saves ~2–3 million ₽ per year through optimization of stock, marketing, and pricing.

Non-financial effects

• Democratization of data — any manager can get an answer without SQL, without waiting in line for an analyst, without prioritizing requests.
• Analysts are busy with real work — complex models, forecasts, A/B tests, rather than SELECT ... GROUP BY.
• Data-driven culture — when data is available instantly, decisions are more often made based on numbers rather than intuition.
• Onboarding new managers — no need to teach them where to go and how to build reports; they just ask the agent.

Summary

An AI analyst is not just text-to-SQL. It is a bridge between the business and data that removes the intermediary. For a retailer, this means: the category manager optimizes the assortment in real time, the marketer adjusts campaigns on the fly, and the branch director sees a problem on the day it arises, not at the end of the month. And analysts finally do what they trained for — complex models and forecasts, rather than endless data extracts.