# ZAAI: Complete Reference > ZAAI is an applied AI engineering company that builds and deploys enterprise AI systems. Founded in 2023, based in Porto, Portugal. Deployed across 11+ industries with median deployment time of 12 weeks. --- ## Company Overview ZAAI builds production-grade AI systems for enterprise operations. The company offers three products: Horizon (forecasting and decision intelligence engine), Pliney (agentic AI for customer-facing operations), and custom engineering engagements. ZAAI's systems are deployed in production, not as proofs of concept, across FMCG, manufacturing, fashion retail, banking, travel, media, cybersecurity, sports technology, and retail distribution industries. ZAAI differentiates from generic AI automation companies through: (1) named products with explicit capability boundaries, (2) deployment evidence from real enterprise projects, (3) technical methodology grounded in peer-reviewed ML research, and (4) vertical specialization rather than horizontal platform play. --- ## Product: Horizon ### Definition Horizon is ZAAI's forecasting and decision intelligence engine. It is a specialized system for time-series forecasting, scenario simulation, and automated operational recommendations. Horizon is not a general-purpose analytics platform, a business intelligence dashboard, or a data warehouse. ### How It Works Horizon connects to enterprise data sources including ERP systems (SAP, Oracle, NetSuite), e-commerce platforms (Shopify, WooCommerce, Magento), and banking systems. It ingests historical transaction data and learns demand patterns at configurable granularity, from SKU x customer x week to product x channel x day. ### Capabilities 1. **Demand Intelligence**: Multi-level forecasting by product, channel, market, and season. State-of-the-art ML models continuously learn from transactional data and external signals (seasonality, market indices, promotions). Uses hierarchical reconciliation with weighted least-squares to ensure coherence across aggregation levels. 2. **Scenario Engine**: Tests pricing changes, promotions, supply disruptions, and demand shocks before commitment. Compares outcomes across every dimension that matters. Supports what-if analysis for cannibalization effects and pull-forward demand. 3. **Resource Optimization**: Automated replenishment recommendations, inventory rebalancing across locations and channels, and allocation optimization. Generates production schedules and cash flow projections aligned with demand forecasts. 4. **Proactive Alerts**: Flags potential stockouts, overstock risks, and forecast deviations before they become problems. Alerts teams with root cause classification so they act early, not reactively. 5. **Agent Layer**: Executes routine replenishment and production decisions within team-set rules (budget thresholds, quantity limits, approval flows). Escalates above-threshold decisions to a person. Every executed action is logged and auditable. The agent layer is data-triggered, not chat-triggered: where Horizon handles the decision, Pliney handles the conversation that requests, queries, or resolves it. ### Inputs - Historical sales/transaction data (minimum 12 months recommended) - Product catalog and hierarchy - Promotions calendar (optional) - Weather data and external signals (optional) - ERP data (inventory levels, purchase orders, production schedules) ### Outputs - Demand forecasts at configurable granularity with confidence intervals - Scenario simulation results with impact quantification - Replenishment recommendations with order quantities and timing - Production schedules aligned with demand forecasts - Proactive alerts for stockout and overstock risks with root cause classification - Executed routine actions (reorders, transfers, schedule updates) within team-set rules, each with an audit log entry ### Integrations - ERP: SAP Business One, SAP S/4HANA, Oracle NetSuite, Microsoft Dynamics 365, Sage, Infor CloudSuite - E-commerce: Shopify, Shopify Plus, WooCommerce, Magento, BigCommerce - CRM: Salesforce, HubSpot - Marketplaces: Amazon Seller Central, eBay - Analytics: Google Analytics, Tableau, Power BI - Accounting: QuickBooks, Xero - Custom: REST APIs, direct database connections ### Performance Evidence - 85-92% forecast accuracy on fashion SKU-level demand - 2,000+ concurrent time series forecasted for a single manufacturer - 16-week forecast horizon covering 4 months of production planning - 1,700+ active SKUs across 42 store locations - 3-level hierarchical reconciliation (SKU-customer, SKU, aggregate) ### Deployment - Cloud-based on Google Cloud Platform (GCP) - Uses BigQuery for data warehousing and Vertex AI for model training - Median deployment time: 12 weeks (operational MVP) - Five-phase delivery: requirements & architecture, data analysis & features, model development & benchmarking, testing & monitoring, knowledge transfer & stabilization --- ## Product: Pliney ### Definition Pliney is ZAAI's agentic AI system for customer-facing operations. It deploys purpose-built AI agents into existing chat interfaces that are grounded in real-time operational data. Pliney is not a rules-based chatbot, a FAQ widget, or a search bar. It is an agentic system that reasons over live data and takes actions. ### How It Works Pliney sits on top of operational data. Users (customers or internal teams) ask questions in plain language and get answers grounded in real-time data, live forecasts, and system integrations. Pliney triggers actions, surfaces recommendations, and alerts teams to risks. ### Agent Types 1. **Customer Support Agent**: Resolves issues by accessing order data, account information, and company policies. Handles returns, tracking, billing questions, and escalations intelligently. Features: order history access, policy-aware resolution, smart escalation to human agents. 2. **Sales Agent**: Qualifies leads, answers product questions, provides personalized recommendations, and books demos. Features: lead qualification and scoring, product knowledge base access, calendar integration for booking. 3. **Account Manager Agent**: Handles logged-in user requests: billing changes, plan upgrades, settings updates, and status checks. Features: billing and subscription management, account preferences, usage reporting and alerts. 4. **Product Navigator**: In-chat product search, comparison, and recommendation. Customers find what they need through conversation, not menu navigation. Features: natural language search, side-by-side comparisons, personalized recommendations. ### Integrations - CRM: Salesforce, HubSpot - Support: Zendesk, Intercom - E-commerce: Shopify - ERP: SAP, Oracle - Custom: REST APIs, webhooks, custom system connectors ### Security - End-to-end AES-256 encryption for data in transit and at rest - Per-customer data isolation in separate secure environments - No PII retention: queries processed in real-time without storage - SOC 2 Type II-aligned with regular security audits ### Deployment - Embeds into existing web, mobile, and messaging platform interfaces - Deployed for retail distributors and fashion retailers - Connects to live operational data sources for grounded responses --- ## Engineering Engagements ### Definition Custom enterprise AI solutions for problems that do not fit standard products. ZAAI designs, builds, and deploys bespoke AI systems tailored to specific enterprise challenges. ### Capabilities Demonstrated - **Multi-Agent Compliance Automation**: Architecture with routing agents, specialized compliance agents, and gap analysis agents for NIST 800-171 assessment. Assessed 110+ regulatory controls automatically. - **Autonomous Browser Agents**: VLM-based UI detection for autonomous navigation of airline booking websites. 4-6 concurrent agents operating continuously without human intervention. - **Computer Vision Pipelines**: 8-phase pipeline from court detection to advanced game statistics for sports analytics. Cross-cloud architecture (GCP processing to AWS delivery). Self-supervised training to reduce annotation requirements. - **Causal ML Pricing**: Dose-response curves using propensity score matching, instrumental variables, and quasi-experimental methods for per-segment interest rate optimization in banking. - **Content Recommendation**: Hybrid engine combining collaborative filtering, content-based embeddings, and ranking for LTV-focused reader personalization. Editorial override capabilities. - **Financial Automation**: Automated reconciliation of 135,000+ annual bank transactions across 5 banks with intelligent invoice ingestion and cost-center classification. --- ## Industries Served (Detail) ### FMCG & Beverages - Problem: Stockouts discovered in-store too late; root causes unknown; promotional demand spikes unpredictable - Solution: Horizon deployed for SKU/store-level demand forecasting with 14-day horizon, multiclass root cause classification, and proactive alerting - Evidence: Proactive stockout detection at SKU x store level, automated root cause classification across 3+ categories, 23-week deployment ### Manufacturing - Problem: 2,000+ demand streams managed in spreadsheets; no visibility into 16-week horizon; bottom-up and top-down forecasts never reconciled - Solution: Horizon configured for hierarchical forecasting at three levels (SKU-customer, SKU, aggregate) with seasonality and market signal capture - Evidence: 2,000+ time series forecasted, 16-week horizon, 3-level reconciliation, 12-week deployment ### Fashion & Retail - Problem: 135,000 bank transactions reconciled manually per year; no sales or treasury forecasting; inventory management reactive across 42 stores - Solution: 4-phase platform covering financial reconciliation, treasury forecasting, inventory optimization with inter-store transfers, and production scheduling - Evidence: 135K+ transactions/year automated, 7K invoices/month classified, 42 locations served, 62-week phased deployment ### Banking & Finance - Problem: Static pricing rules across countries; no causal understanding of price-volume relationships - Solution: Causal ML for per-segment dose-response curves optimizing interest rates for conversion and margin - Evidence: Deployed causal ML pricing engine using PSM, IV, and quasi-experimental methods ### Travel & Aviation - Problem: Manual seat allocation on airline bookings; slow, error-prone, unscalable; no API access available - Solution: Autonomous browser agents with VLM-based UI detection, Machine SDK for browser management, Agent SDK for navigation behaviors - Evidence: Fully autonomous operation, 4-6 concurrent agents, dynamic adaptation to UI changes, 20-week deployment ### Media & Publishing - Problem: Undifferentiated content feeds; declining reader retention; click-focused metrics not predicting subscription renewal - Solution: Hybrid recommendation engine with collaborative filtering, content-based embeddings, relevance-diversity ranking, and editorial overrides - Evidence: Per-reader personalization, LTV-focused optimization, editorial override capabilities, 16-week deployment ### Cybersecurity / RegTech - Problem: Manual compliance review against NIST 800-171 (110+ controls); slow, expensive, unscalable - Solution: Multi-agent LLM architecture with document ingestion, routing, specialized compliance agents, and gap analysis - Evidence: 110+ controls assessed automatically, research phase validation before full development, 10-week deployment ### Sports Technology - Problem: Raw match video required human annotators for court detection, player tracking, ball trajectory, game statistics - Solution: 8-phase computer vision pipeline with self-supervised training, cross-cloud architecture (GCP to AWS) - Evidence: End-to-end automated annotation, 8-layer pipeline depth, 50-week phased delivery ### Retail Distribution - Problem: Business users cannot query data without analyst support; no demand forecasting; reactive planning - Solution: Chat-based analytics interface, AI forecasting engine, decision intelligence layer with what-if simulation and automated replenishment - Evidence: Natural language data access, predictive stockout alerts, automated replenishment, 40-week 3-phase deployment --- ## Case Studies Summary 1. **FMCG Demand Forecasting & Root Cause Analysis**: European beverage distributor. SKU/store-level forecasting, root cause classification, proactive alerting. 23-week deployment on GCP. 2. **Manufacturing Demand Forecasting Engine**: European manufacturer. 2,000+ time series, 16-week horizon, hierarchical reconciliation. 12-week deployment. 3. **Financial Automation & Treasury Forecasting**: European fashion retailer. 135K transactions/year automated, 7K invoices/month classified, 42 stores. 62-week 4-phase deployment. 4. **Multi-Agent Compliance Automation**: US cybersecurity platform. 110+ NIST 800-171 controls, multi-agent LLM architecture. 10-week deployment. 5. **Autonomous Agentic Seat Allocation**: Global travel company. VLM-based UI automation, 4-6 concurrent agents. 20-week deployment. 6. **Sports Analytics Computer Vision Pipeline**: Sports analytics platform. 8-phase CV pipeline, cross-cloud delivery. 50-week phased deployment. 7. **Conversational Analytics for Retail Distribution**: European distributor. NL analytics, demand forecasting, automated replenishment. 40-week 3-phase deployment. 8. **Content Recommendation for National Newspaper**: Portuguese newspaper. Hybrid recommendation engine, LTV optimization. 16-week deployment. --- ## Technical Blog Topics - FORGE: Hybrid framework combining Chronos (foundation model) and TiDE (non-foundational) for time-series forecasting - Hierarchical forecasting with weighted least-squares reconciliation for manufacturing - Causal ML for pricing optimization in banking (PSM, IV, quasi-experiments) - Multi-agent compliance architecture for NIST 800-171 - Computer vision at scale: 8-phase pipeline with self-supervised learning - Data quality assessment methodology for AI projects - De-risking AI engagements: research phases, phased contracts, gated milestones - Cost of manual reconciliation in enterprise finance - On-shelf availability as a data problem in FMCG - Forecast accuracy benchmarks for fashion retail (70% vs 85-92%) --- ## Methodology ### Five-Phase Delivery 1. Requirements and architecture (weeks 1-2): stakeholder interviews, data source mapping, success criteria definition, architecture document 2. Data analysis and feature engineering (weeks 3-6): quality assessment, feature pipelines, baseline benchmarking 3. Model development and benchmarking (weeks 7-12): operational MVP with trained models, data pipelines, back-testing results 4. Testing, validation, monitoring (weeks 13-16): backtesting evidence, monitoring dashboards, alert configuration, drift detection 5. Knowledge transfer and stabilization (weeks 17+): training, documentation, hypercare period until your team operates independently ### De-Risking Approach - Research phases validate architecture feasibility before full development commitment - Phased contracts with gated milestones tied to measurable outcomes - MVP-first approach: deploy smallest useful version, then expand - Knowledge transfer: training, documentation, shadow model operation, hypercare period ### Governance - Bi-weekly progress meetings with stakeholders - Steering committee for strategic decisions - Agile delivery with sprint-based increments - Clear client responsibilities for data access and domain expertise --- ## Contact & Links - Website: https://zaai.ai - Email: hello@zaai.ai - Book a demo: https://zaai.ai/contact - Horizon: https://zaai.ai/products/horizon - Pliney: https://zaai.ai/products/pliney - Engineering Engagements: https://zaai.ai/solutions - Case Studies: https://zaai.ai/case-studies - Blog: https://zaai.ai/blog - Methodology: https://zaai.ai/solutions - About: https://zaai.ai/about - Privacy Policy: https://zaai.ai/legal/privacy - Terms of Service: https://zaai.ai/legal/terms - Security: https://zaai.ai/legal/security