Iftach Nataf - Technical Product Manager (AI, Automation & Platforms)
Technical Product Manager in Israel (Center, Hybrid / on-site) with 4+ years of experience owning stakeholder discovery, PRDs, ADRs, MoSCoW/RICE prioritization, data architecture, and production delivery across AI, RPA, and B2B2C platforms.
Measured Impact: Baselines, Measurement Methods & Delivery Stages
- 30 days to 90 sec review cycle time across 1,000+ sources (Stage: Deployed Enterprise POC at If-Tech): Baseline & Problem: An infrastructure/utilities contractor spent ~30 manual days every month cross-checking 1,000+ Excel sheets across fragmented work-order and partial/final billing workbooks because row status was encoded visually in RGB cell fill and font colors rather than data columns, and Hebrew table headers shifted coordinates across sheets. Measurement & Mechanism: Timed end-to-end execution (~90 seconds) of a deterministic Python pipeline (`openpyxl` RGB hex color-to-status decoder + regex coordinate locator for shifting Hebrew headers + `pandas` cross-workbook reconciliation + `xlsxwriter` auto-filtered review workbooks and `matplotlib` statistical PDF plots).
- ₪1M+ government deal advanced & ~40% department load covered (Stage: Pre-Sale AI POC & Rollout Plan at Ness Technologies): Baseline & Product Re-Scoping: The client originally requested basic coordinate-based invoice RPA. Stakeholder shadowing and document sampling proved template RPA would fail on unstandardized Hebrew/RTL supplier invoices. Measurement & Mechanism: Re-scoped into a Hebrew/RTL AI Document Understanding POC evaluated against an SME-labeled dataset using precision/recall KPIs, a field-level confidence routing threshold (auto-post vs. human-in-the-loop validation queue), and an error taxonomy, covering ~40% of total department volume with ~30% handling-time savings.
- 35% reduction in manual handling hours (Stage: Enterprise Production Rollout at University of Haifa, subcontracted via Ness): Baseline & Measurement: Measured before/after staff handling hours across recurring multi-step institutional operations after deploying an end-to-end UiPath transactional state machine with business vs. system exception routing, automated daily status-file reporting, zero silent failures, and operator training.
- ~50% faster engineering time-to-solution (Stage: Enterprise Production & Internal SDK at Valor Automation Solutions): Baseline & Measurement: Measured recurring issue resolution time across 10 parallel Israel and offshore client tracks after pausing ad-hoc code duplication to ship versioned mail-intake and structured error-logging SDKs, upstream/downstream CRM schema impact checks, and unified onboarding runbooks.
- 20% team delivery velocity lift (Stage: 12-Month Acting Team Lead at Ness Technologies): Baseline & Measurement: Tracked sprint throughput and delivery milestones across 9 concurrent enterprise automation initiatives and 6 clients during a 12-month leadership transition in the UiPath division.
Flagship Deep-Dive Case Study: 01 · YN Business Solutions (Live in Production at https://ynbis.com/)
Delivery Stage: Launched Production Platform · Active Daily Operations | Role: Product Lead, System Architect & Builder (Pro Bono, B2B2C)
- Business Problem: A growing financial, insurance, and pension-operations agency operated on fragmented spreadsheets, manual lead distribution, disconnected external partner agencies, and ad-hoc regulatory tracking, causing lead leakage and operational overhead as volume scaled.
- Exact Responsibilities: Owned the entire product and engineering lifecycle end to end. Conducted stakeholder discovery and shadowing with the agency owner (Mar-Jun 2026), authored PRDs and Architecture Decision Records (ADRs), prioritized the multi-phase roadmap, designed the MariaDB relational schema, directed AI coding agents under stage-gated QA, and shipped continuous releases through CI/CD.
- Product Prioritization (Phased MoSCoW Rollout): 1. Phase 1 (Stop the Bleeding): Shipped the core ~45-route internal Admin CRM (Kanban & list pipelines, duplicate review, lead routing) and no-code Hebrew/RTL Website & Brochure Builder first to eliminate daily lead leakage. 2. Phase 2 (External Collaboration & Comms): Layered query-scoped Partner and Calling-Center Portals, live Voicenter telephony/CDR webhook integration, built-in IMAP/SMTP webmail (with mobile XML autoconfig), and public `/sign/[token]` PDF contract e-signatures. 3. Phase 3 (FinOps, Governance & AI): Shipped the commission ledger, full HR & time-clock suite (with strict self-approval ban), Israel Do-Not-Call (DNC) registry verification, 8 SQL-pushed-down executive analytics suites, and privacy-first multi-provider AI.
- Product Trade-Offs & What I Decided NOT to Build (ADR-01): 1. Cut Microservices for a Modular Monolith: Chose a single Next.js + TypeScript + Prisma + MariaDB modular monolith over separate portal microservices, eliminating distributed-trace overhead while enforcing strict query-level tenant isolation across 4 portals. 2. Cut Autonomous AI Replies for a Mandatory PII Scrub & Human Preview Gate: Refused to let LLMs auto-message financial clients or ingest raw PII. Built a pre-flight regex + entity PII scrubbing gateway with a mandatory human preview step before any prompt reaches Anthropic, OpenAI, or Groq. 3. Cut External Redis Brokers for MariaDB SKIP LOCKED: Implemented asynchronous job queues (bulk imports, DNC registry checks, data retention) directly in MariaDB using `SKIP LOCKED` transactions, keeping infrastructure lean and ACID-consistent. 4. Cut Client-Side BI Transforms for 8 SQL-Pushed-Down Suites: Pushed weighted forecast, funnel conversion, source ROI, renewals, cross-sell, deal aging, partner scorecard, and data-quality aggregations directly to SQL for sub-100ms executive readouts.
Additional Portfolio Case Studies (With Stages & Product Decisions)
- 02 · Our Home OS (2025-2026 · Stage: Launched Production Web App at https://household-routine.netlify.app/): Hebrew-first multi-tenant household & operations OS on Next.js, Prisma, MariaDB, and Vitest (30+ routes, ~60 models). Features official Israeli supermarket XML price-comparison sync across branches, thermal receipt OCR, camera barcode scanner, Children Mode, expiring Babysitter tokens, and AES field-encrypted document vault. Product Decision & What I Cut (ADR-02): Deliberately chose NOT to subsidize centralized LLM token burn or allow silent AI database writes; architected a Bring-Your-Own-Agent (BYOA) model across Claude, OpenAI, Gemini, and Groq where every `Ctrl+K` natural-language command stops at a mandatory human-approval draft gate.
- 03 · If-Tech ('Stanga' · 2024-2026 · Stage: Functional Alpha + Deployed Client POCs): Founder & TPM of Stanga, a hybrid B2B automation orchestration platform (React/Node control plane + outbound-only Electron/Node local desktop agent storing credentials in the OS keychain via `keytar`, FIFO job-queue deduplication, 4 trigger modes, and 7-language polyglot execution across Python, Node.js, C#, C, C++, Java, PowerShell), plus a production OpenCV + multi-pass `heb+eng` Tesseract OCR engine and the 90-sec data consolidation POC. Product Decision & Venture Gate (ADR-03): Rejected cloud-only SaaS runners because enterprise IT blocked inbound firewall ports, designing an outbound-only local agent instead. After proving the alpha and 90-sec POC, executed a disciplined founder kill-gate based on CAC payback vs. ~₪335K legacy RPA incumbents, closing the startup cleanly on data.
- 04 · University of Haifa (2024-2025 · Stage: Enterprise Production Rollout · Subcontracted via Ness): Technical Product & Automation Lead delivering end-to-end UiPath institutional process automation. Product Decision (SPEC-02): Prioritized a deterministic UiPath state machine separating business-rule exceptions from system retries over a brittle linear script, delivering a 35% manual-hour reduction with zero silent failures and daily status-file visibility.
- 05 · Ness Technologies (2023-2025 · Stage: 9 Enterprise Production Rollouts + Pre-Sale AI POC): Acting Team Lead (12 months, +20% velocity) & TPM across 9 concurrent enterprise automation initiatives and 6 clients. Product Decision (SPEC-01): Pushed back on a narrow coordinate-RPA invoice brief after discovery proved layout variance, re-scoping into a Hebrew/RTL AI Document Understanding POC with human-in-the-loop confidence routing that advanced a ₪1M+ government deal.
- 06 · Valor Automation Solutions (2022-2023 · Stage: Enterprise Production + Internal Engineering SDK): RPA Developer, Knowledge & Product Process Leader managing Valor's largest enterprise account and 10 parallel Israel/offshore tracks via MoSCoW prioritization. Product Decision: Paused ad-hoc project duplication to build versioned mail-intake and error-logging SDKs plus mandatory CRM/DB schema impact reviews, cutting recurring engineering time-to-solution by ~50%.
- 07 · PacketPilot (2025 · Stage: PM Team Lead of 6-Person Pod · PRD, Gherkin Epics & Validated MVP): Team Lead of a 6-person PM capstone team in the Product Experts Program solving home network safety for households with 25-40 IoT devices (vs. ~6 perceived). Product Decision & What We Cut (PRD-01): Deliberately cut Deep Packet Inspection (DPI) and router firmware flashing from the MVP, using lightweight metadata and behavioral heuristics to deliver a 3-state Safety Score (Green/Yellow/Red), 1-sentence plain-language explanations, 1-tap actions (`Block device`, `Limit activity`, `Mark as safe`), 3 personas, 4 Gherkin Epics, and 3 mathematical KPI formulas (40% weekly registration conversion target).
- 08 · PainterBot (2025 · Stage: Production Hardware Firmware · Modular C++ Refactor · Pro Bono): Re-architected an automated industrial painting carriage firmware on Arduino Nano from a monolithic `.ino` script with blocking Interrupt Service Routine (ISR) loops into 7 modular C++/H units (`config.h`, `interrupts`, `motion`, `piston`, `ui`, `util`, `globals`). Product & Engineering Decision (SPEC-03): Preserved 100% of factory potentiometer and stepper calibration thresholds (`1000 / 650 / 450`) to require zero mechanical recalibration while replacing blocking ISR loops with immediate valve safe-off, volatile abort flags, 15s stepper / 6s piston safety watchdogs, and 25ms switch debouncing.
Direct Stakeholder & Peer Endorsements (Anonymized)
- Peer Product Manager · Product Experts Program (PacketPilot 6-Person Capstone Team): "I want to warmly recommend Iftach. We completed the Product Management program together - he brings a strong technical background, truly understands the engineering and product details, and has outstanding interpersonal skills. He led our 6-person capstone group as Team Lead, and it was a pleasure working with him."
- Project Stakeholder · University of Haifa (Enterprise UiPath Automation · Subcontracted via Ness): "Champion! Following the rollout of our end-to-end process automation - whenever you need recommendations down the road, share my contact details gladly."
How I Work (5-Stage Product Lifecycle)
- Discover: User interviews, shadowing, JTBD, and workflow mapping to quantify root operational bottlenecks (Outputs: Problem statement, Process map, Opportunity size).
- Define: PRDs, ADRs, and KPIs anchored to client OKRs across RPA, OCR, deterministic APIs, and LLMs (Outputs: PRD & ADRs, Success metrics, Build-vs-buy call).
- Build: Time-boxed POCs and production MVPs with RBAC, data flows, and failure modes designed in from day one (Outputs: Working POC / MVP, Architecture & data flows, Failure-mode review).
- Ship: Stage-gated QA, release planning, CI/CD rollout, and hands-on user enablement (Outputs: Release plan, Rollout & training, Adoption baseline).
- Measure: Telemetry and SQL-backed dashboards tracking adoption, precision/recall, and error taxonomies (Outputs: KPI dashboard, Error taxonomy, Next-iteration backlog).
Principles
- If you do it three times, a robot should do it for you.
- The wrong problem, solved perfectly, is still wrong.
- Security isn't a phase. It's a design constraint.
- An MVP is a proof instrument, not a smaller version of everything.
Contact: iftachn85@gmail.com | +972-53-9280225 | LinkedIn: https://www.linkedin.com/in/iftach-nataf | GitHub: https://github.com/IFTACHNATAF | Full AI context: /llms.txt and /llms-full.txt