Copyright Fast Company
The debate around AI ROI has gotten loud—and, frankly, a little cyclical. One moment, we’re hearing that AI is the key to exponential growth; the next, that 95% of AI pilots fail. At Addi, we’ve been able to leverage AI to grow 4x faster while operating at ~2x the profitability of BNPL peers. This year alone, we’ve saved more than $500,000 from our AI initiatives. But how have we accomplished such strong AI ROI? The difference between performative AI and AI with returns isn’t in which model or tool you’re using; it’s how your team is using them. Here’s how we’ve driven genuine AI-native team adoption and built a workflow/data pipeline that actually makes sense. Subscribe to the Daily newsletter.Fast Company's trending stories delivered to you every day Privacy Policy | Fast Company Newsletters 1) Hire and grow for fluency We run nationwide, admissions-style assessments to find talent in unexpected places (from the Amazon to the Ecuador border), then teach AI-native workflows from day one. From our intern program through our senior leadership, we design our interview process for the AI age. We assign a relevant project—something candidates could use AI to help with—but then have a panel interview where they present their project, ensuring that candidates actually know the ins and outs of their work without an AI aid. Our interviews additionally probe into potential candidates’ own familiarity with AI tools, while our intern cohorts get hands-on with agents and graduate into teams already expecting that fluency. The pipeline is designed to recruit for an AI era from the get-go, versus being an afterthought once already employed. 2) Codify AI-native rituals into culture When it comes to cultivating an AI-native work culture, AI-native is a learned behavior. We invested in extensive AI onboarding and habit-building, pairing every knowledge worker with the right agent or copilot, and encouraging AI usage as the company default. Today, more than 90% of our engineers are weekly active copilot users and ~80% of AI-generated code is accepted. This translates into efficiency gains of up to 60% without increasing headcount. We’ve kept our core product engineering team flat for three years while shipping more products. The story here isn’t in the savings; it’s in the deep level of AI adoption we’ve witnessed among our employees by securing their buy-in, setting expectations for an AI-friendly environment, and offering targeted training. Rollouts fail when AI is treated as a “here only if you need it” tool. They work when companies rewire rituals around it—e.g., code reviews with AI diffs, CX stand-ups that inspect agent transcripts, legal postmortems that include our AI’s outputs—to normalize the behavior. You might even consider baking AI proficiency into employee reviews. In other words, don’t over-index on tools; over-index on culture. That cultural shift is why AI usage at Addi is voluntary yet ubiquitous. 3) Design AI as a colleague There’s a reason our in-house agents have regular names like “Addri” and “Aegis.” Every agent at Addi is treated like an employee—one with a clear scope, service-level agreements (SLAs), and metrics. Addri’s job is first-contact resolution with target customer satisfaction (CSAT); the merchant agent owns KYP throughput and reactivation; Aegis owns escalation latency and evidentiary completeness. Human owners review outputs and tune prompts like they would a new hire’s playbook, and we always welcome teamwide feedback on how our fellow “agentic employees” can improve before their next review cycle. Moreover, our AI “employees” have the same depth of contextual knowledge and understanding that a human employee would, to help them function side-by-side with our team and minimize the frustration that comes with false or limited context. Our agents are tailored to specific roles, not catchalls from an outside vendor that shoehorns a base agent into a wide variety of situations. We ensure they’re trained with high-quality, high-volume, company-owned data. We spent four-plus years building a world-leading data platform, ensuring more than 40 terabytes of data was instantly available as it began building AI agents, giving our “digital teammates” the best possible training. 4) Invest in the right foundations “AI-first” isn’t what works; “data-first” is. This is how you ensure your “AI colleagues” have that employee-like context. advertisement More than four years ago (pre-LLMs!) we made the decision to invest in a next-generation data engine that would ensure everything that happened on our platform (from a single text message to a full underwriting analysis) would be stored and could be queried by anyone and anything—traditional AI models, human analysts, and, yes, even LLMs via vectorization. With a single monorepo and an event-based system that logs everything, we have nearly perfect context: 50 terabytes of clean, searchable data. If you don’t own your stack (i.e., control your data and event logs) you will rent your advantage to a vendor. Set your AI-native team up for success by logging everything, and reap the benefits of a database that can be read by humans and AI alike. 5) Celebrate adoption Reward employees’ usage of AI by celebrating adoption rates, cycle-time reduction, and defects avoided. This year, our AI initiatives saved upwards of $500,000 in annual operating costs. For lean teams where a startup’s success is their teammates’ success, these metrics (and transparency) matter. That $500K isn’t a bottom-line cut; it’s $500K back into the pockets of our employees in the form of raises, better benefits packages, and profit sharing. Tie budgets to solved tickets, minutes saved, merchants activated—then compound wins into subsequent quarters. That mindset of “AI gains are your gains” is why AI can comfortably power half of our legal and coding throughput, a big chunk of CX, and critical onboarding flows. In Summary Train your people to be AI-native and give them the infrastructure to thrive. The models will change. The muscle you build won’t. This approach is how we’ve been able to launch more products more quickly while maintaining a generally lean team—and it’s why I’m confident the best AI ROI stories are still to come.
