Zenit Auto: Agentic Inventory Decision Intelligence Platform

Direct answers

What did Perfectory AI build for Zenit Auto?

Perfectory AI built an agentic inventory decision intelligence platform that turns sales, stock, competitor, and pricing data into reorder, markdown, stock-gap, price-gap, and assortment recommendations.

What business problem does this AI case study solve?

It solves slow manual inventory analysis by giving operators prioritized actions, configurable recommendation logic, forecasting evidence, and an AI assistant that explains why each recommendation was generated.

Measured business impact

SKU scale: 100k+

Recommendation + agent workflows: 6 engines

Forecasting benchmark: 6+ metrics

The challenge

The client needed to move from manual inventory decisions and static spreadsheet analysis to a data-driven recommendation system for automotive parts purchasing, pricing, and assortment planning.

Business and product requirements:

• Identify which products should be reordered before stockouts create lost sales.
• Detect overstocked items and recommend markdown opportunities without destroying margin.
• Compare internal availability against competitor stock and pricing behavior.
• Surface new assortment opportunities based on competitor coverage and market signals.
• Give business users a clear recommendation dashboard instead of raw operational data.
• Let users ask natural-language questions about SKUs, products, sales behavior, similar products, and why a recommendation was generated.

Why AI mattered

AI was critical because inventory decisions could not depend only on static rules or simple daily averages.

AI was critical to business differentiation and scale:

• Converted fragmented sales, stock, pricing, and competitor data into actionable recommendations.
• Replaced manual SKU-by-SKU review with automated prioritization.
• Improved replenishment decisions by estimating demand and stock coverage risk.
• Supported smarter markdown decisions by balancing overstock risk with margin protection.
• Created a foundation for future forecasting models using seasonality, lifecycle, holidays, and external demand drivers.
• Added an agentic explanation layer so operators could inspect recommendations, validate selected actions, and understand the evidence before approving purchasing or pricing decisions.

Our role

Perfectory AI delivered the end-to-end AI automation and decision intelligence layer.

Perfectory delivered the end-to-end AI execution layer:

• Designed the data architecture for inventory, sales, competitor snapshots, and recommendation outputs.
• Built automated ETL workflows for processing internal and competitor data.
• Implemented recommendation engines for reorder, overstock markdown, competitor stock gaps, competitor price gaps, and new assortment opportunities.
• Connected business UI parameters directly to backend notebook pipelines.
• Added configurable controls for sales period, overstock threshold, competitor stock requirements, margin limits, and safety stock logic.
• Created a separate forecasting backtest notebook to compare multiple demand prediction approaches against real sales data.
• Rebuilt and containerized backend and ETL worker services for local Docker execution.
• Added an agentic assistant experience that can analyze selected recommendations, answer product and SKU questions, explain reorder or markdown logic, and surface evidence from stock, demand, sales, and competitor data.

What we delivered

We delivered a production-oriented AI recommendation platform with an agentic assistant that turns operational data into prioritized inventory actions and explainable business decisions.

Agentic analysis assistant

• Lets users select recommendations and ask the assistant to validate whether the action looks logically supported by the available evidence.
• Answers natural-language questions about a product, SKU, sales behavior, similar products, stock status, demand logic, and recommendation rationale.
• Explains why a reorder, stockout recovery, markdown, or manual-review flag was generated, including the business rules and data points behind the verdict.

Reorder recommendation engine

• Detects products at risk of stockout using demand estimates, lead time, target cover days, current stock, and safety stock buffer.
• Applies business safety rules so suggested order quantities do not fall below the minimum acceptable demand coverage threshold.

Overstock and markdown engine

• Identifies products with excessive stock coverage.
• Recommends markdown candidates while respecting maximum markdown share of margin.

Competitor stock gap analysis

• Finds products where competitors have stock and the client has weak or zero availability.
• Uses configurable thresholds for minimum competitors with stock.

Competitor price gap analysis

• Compares client pricing against competitor pricing signals.
• Helps detect opportunities for price adjustment and market positioning.

New assortment opportunity engine

• Identifies products competitors carry that may be missing or underrepresented in the client catalog.
• Uses configurable competitor count and typical stock thresholds instead of hardcoded rules.

Forecasting and backtesting notebook

• Compares current demand logic against rules-based, statistical, and ML forecasting approaches.
• Measures accuracy, WAPE, MAE, RMSE, F1 score, under-order risk, over-order risk, and floor violations.
• Gives the client an evidence-based way to decide which forecasting model should power future recommendations.

Results and value

Business outcomes delivered:

• Moved inventory decisions from manual review to automated recommendation workflows.
• Reduced dependency on static spreadsheets and hardcoded business assumptions.
• Gave business users configurable controls for recommendation logic.
• Added an agentic assistant that improves trust by explaining recommendations, validating selected actions, and answering product-level questions in context.
• Improved visibility into reorder risk, overstock risk, competitor pressure, and assortment gaps.
• Created a measurable forecasting benchmark using real historical sales.
• Established a scalable AI foundation for seasonality, lifecycle trends, holiday effects, category-specific behavior, and external demand drivers.
• Created faster purchasing decisions, better stockout prevention, more disciplined markdown strategy, clearer competitor-aware assortment planning, lower manual analysis effort, and a reusable AI automation platform for future inventory intelligence.

Case study FAQ

Can this approach work for other distributors or retailers?

Yes. The same pattern fits distributors, retailers, and catalog-heavy businesses that need to combine sales, stock, margin, competitor, and product data into controlled inventory actions.

How does Perfectory AI reduce risk in inventory automation?

Perfectory AI uses configurable business rules, backtesting, explainable recommendation evidence, and human approval so teams can validate recommendations before changing purchasing or pricing decisions.