Forward Deployed Engineer - AI Transformation

We're looking for a Principal, AI Transformation to join AI Foundry (our internal team dedicated to making Tipalti AI-fluent through education, equipping, and end-to-end process redesign). You will serve as a key driver of Tipalti's AI-led operational transformation across General & Administrative functions — Legal, People, and Financial Planning & Analysis (FP&A) — working as a full-time embedded partner within their organizations.

Why join Tipalti?

Tipalti is the AI-powered platform for finance automation, elevating how finance teams operate in the global economy. We empower our customers to scale faster and smarter by removing the complexities of doing global business and accelerating their finance operations efficiency. Our platform provides a comprehensive suite of finance automation solutions designed for mid-market businesses across accounts payable, global payouts, procurement, employee expenses, corporate cards, supplier management, tax compliance, and treasury. Tipalti partners with leading financial institutions such as Citi, Wells Fargo, J.P. Morgan, and Visa, enabling over 5,000 global companies to efficiently and securely pay millions of suppliers and payees across 200+ countries and territories, in 120 currencies.

At Tipalti, we pride ourselves on our collaborative culture, the quality of our product and the capabilities of our people. Tipaltians are passionate about the work they do, and keen to get the job done. Tipalti offers competitive benefits, a flexible workplace, and an environment where diverse individuals can thrive and make an impact. Our culture ensures everyone checks their egos at the door and stands ready to reach for success together. 

Founded in Israel in 2010, Tipalti is a global business headquartered in the San Francisco Bay Area (Foster City) with offices in Tel Aviv, Plano, Toronto, Vancouver, London, Amsterdam, Tbilisi and Medellin.

In this role, you will:

• Understand the customer and identify gaps - Find the pain points within Legal, People, and Finance; learn the constraints they can’t change, the workarounds they’ve created, and find the problem that actually costs them time.
• Map and solve end-to-end workflows — Follow work across functions and systems to understand where the handoff failures are and where AI can create leverage across boundaries.
• Build and ship the fix yourself, fast. Prototype and deploy using approved LLMs (Claude and other AI tools), agent frameworks, MCP-based tooling, automation platforms, APIs, data integrations, and code, where code is the right answer. Follow Tipalti’s security, privacy, legal, compliance, architecture, and change-management requirements, including required approvals before production deployment.
• Know what we can build in-house — and be honest about what we can’t. Maintain a live, evidence-based view of what is buildable internally, both with code and without it, including what it takes to run and maintain. Your build-vs-buy guidance should address security, privacy, intellectual property, regulatory, data residency, vendor concentration, resilience, auditability, maintenance, and exit risks. It should still hold up six months after you give it.
• Advise the Divisional Teams on their respective tech stacks (what we keep, consolidate, and retire): Have a point of view on how the stack fits together - where the data lives, where the integration seams are, who owns the data and tool, and where an agent or automation can sit across tools instead of inside one. 
• Design for reliability, not for the demo. Think in failure modes: where the model is wrong, how you’d know, what evaluation looks like, where meaningful human oversight belongs, and how sensitive data is handled. Build in access controls, testing, bias and accuracy assessment, logging, audit trails, monitoring, escalation, rollback, incident response, and business continuity.
• Think of each solution as a product for internal customers. Define success, risk, control, and compliance metrics before you build. Track adoption, performance, control effectiveness, and incidents on a regular cadence. Iterate based on how people actually use it, not what you initially planned.

About You

• You build things, hands-on. You’ve shipped AI or automation solutions that real people depend on — agents, workflow automations, integrations, internal tools. You’re comfortable in a terminal, reading API documentation, and writing and debugging your own scripts (Python, TypeScript, or similar). You’ve felt the difference between “demo’d once” and “actually used.”
• Fluent with Claude. You’ve built with Claude or an equivalent frontier model in earnest (agentic coding, sub-agents and skills, MCP, context and prompt design, and evaluating output quality). You can adapt quickly as approved tools change. You know what these systems are bad at, not only what they’re good at: non-determinism, context limits, cost and latency, and when retrieval, tooling, or a plain script beats a model call.
• Think in end-to-end workflows, not silos. You trace work across functions and systems. You notice handoffs, redundancy, and instances where different parts of the org solve the same problem twice. You see inefficiency as a systems problem, not a function problem.
• Proactive and self-directed. You find problems before they're on your list. You're comfortable setting your own priorities and making the day-to-day call yourself. You keep moving in between syncs.
• Comfortable at all levels of the org. You sit with frontline teams and actually understand their workflow. You sit with the CFO, CLRO, and CPO and speak credibly to strategy. You’ll sit with engineers and translate business problems into technical specs.
• Translator and connector. You turn vague asks into working solutions. You spot where one function has solved a problem that another hasn't. You tell leaders "not that, but here's what would actually help." You see patterns across domains and connect the dots.
• Curious about unfamiliar territory. You are genuinely interested in how these functions work, comfortable asking "stupid questions" to understand the actual workflow, and able to pick up domain language quickly.