Boyko Karadzhov, Co-founder and CTO at Payhawk

Boyko Karadzhov is Co-founder and Chief Technology Officer at Payhawk, a leading spend management platform he helped establish in 2018. He oversees the technical development of next-generation corporate card and automated spend management solutions and has been instrumental in building the company’s AI-powered financial infrastructure, including the recent launch of the “AI Office of the CFO”—a suite of specialized AI agents designed to automate routine finance operations while maintaining enterprise-grade security and control.

How are you currently integrating AI into your organisation’s core operations, and what strategic goals is it helping you achieve?

We’ve integrated AI directly into our financial operations platform through what we call the “AI Office of the CFO”—a suite of specialised agents that autonomously handle routine finance tasks. Unlike bolt-on AI solutions, these agents are built on our existing financial infrastructure, understanding company policies, workflows, and permissions from day one.

Our strategic goal is fundamental: transforming finance teams from operational handlers to strategic advisors. The Financial Controller Agent eliminates manual receipt chasing, the Procurement Agent streamlines purchase approvals, and our Travel and Payments agents handle their respective workflows end-to-end. This isn’t about incremental efficiency gains—it’s about reimagining how finance operates, freeing professionals to focus on analysis, forecasting, and strategic decision-making that drives business growth.

What lessons have you learned from early AI deployments that have reshaped your approach to implementation or innovation?

Three fundamental lessons have shaped our AI philosophy at Payhawk:

First, we learned that we must not pursue any type of AI if forms, simple statistics or algorithms do the job better. This principle saved us from over-engineering solutions and helped us focus AI where it truly adds value—handling complex, context-aware tasks that require understanding of financial workflows.

 Second, we discovered that the key ingredients to complement AI are data definition, workflows setup, and integrations—not the AI models themselves. Our early attempts taught us that conversational AI works best as a new interface to existing data models and capabilities rather than a replacement system. 

Third, we realized that trust in AI is fragile, requiring us to only release features with proven quality and accuracy. This led to our approach where every agent works out-of-the-box with great defaults and no configuration required. We also implemented comprehensive performance dashboards that measure quality and speed for on-demand tasks, and helpfulness for proactive tasks initiated by the agents.

In what ways has AI changed how your teams work together—across departments, disciplines, or decision-making processes?

AI has fundamentally shifted our organizational dynamics according to our core principle that “AI is not owned by one team, but one team can build foundations for everybody else.”

We’ve established cross-functional AI literacy where teams across engineering, product, customer success, and sales and marketing now interact with AI daily. This isn’t just about using AI tools—our manifesto emphasizes that “we cannot build it if we don’t understand it, we need to encourage usage of AI every day.” This has created a culture where every team contributes insights about AI applications in their domain.

Decision-making has become even more data-driven through our performance dashboard approach. We now have real-time visibility into the quality, speed, and helpfulness of each AI agent, allowing teams to make informed decisions about feature development and optimization.

Cross-departmentally, AI has unified our focus around measurable outcomes. Engineering teams optimize for quality and speed metrics, while product teams focus on helpfulness scores for proactive features. Customer success teams use agent performance data to guide strategic conversations with clients.

Perhaps most significantly, AI has extended beyond our product into operational excellence. Teams now leverage AI for internal processes—from customer support to financial analysis—creating an organization-wide understanding of AI’s practical applications and limitations.

What criteria do you use to determine whether an area of the business is ready—or worth—being automated or enhanced by AI?

We evaluate potential AI applications against four criteria:

First, volume and repetition—tasks that consume significant time with predictable patterns are ideal candidates. Receipt management exemplifies this: high volume, rule-based, and time-consuming.

Second, data availability and structure—AI needs quality, structured data to function effectively. Areas where we already capture comprehensive data are natural starting points.

 hird, risk tolerance—we prioritize areas where automation errors won’t cause significant business disruption. Financial controls remain important, but receipt processing mistakes are less critical than, say, investment decisions.

Fourth, human value displacement—we ask whether automating this task genuinely frees humans for higher-value work. If AI just makes people slightly faster at routine tasks, the impact is limited. When it eliminates entire categories of operational work, the transformation is meaningful. 

Our philosophy is that AI should handle what computers do best—pattern recognition, repetitive processing, and consistent rule application—while humans focus on judgment, creativity, and strategic thinking that drives business forward.

How do you balance the need for rapid AI adoption with the risks of bias, inaccuracy, or overdependence on automated systems?

While it’s challenging to completely eliminate hallucinations due to the inherent nondeterministic nature of generative AI, there are numerous strategies to significantly mitigate these risks. 

AI Large Language Models in public use are more susceptible to hallucinations because of their broad application scope. In contrast, our AI agents operate within a controlled infrastructure, specifically designed to minimize such risks. Each agent is specialized to excel in a defined set of tasks, continuously learns from its work to improve performance, and is regularly monitored and refined by our product team. Additionally, these agents rely on human oversight for sensitive or critical actions and decision-making, ensuring that accuracy and reliability are maintained.

How are you managing the organisational and cultural shifts that come with embedding AI into legacy systems and structures?

Payhawk is actively hiring for engineering talent to support its AI-first vision, rather than cutting headcount. Our rationale is simple: AI cannot deliver on its huge potential without expert engineers to guide it. 

For complex, interconnected, and regulated industries like finance, engineers are essential for command and control. They need to translate AI into banking-grade systems, embed evolving compliance rules into decision making and ensure ethical guardrails and governance. 

What do you believe will separate AI leaders from laggards over the next five years in your industry or sector?

The separation will come down to three critical differentiators: integration depth, security-first architecture, and organizational AI fluency.

Integration depth will be the primary divider. Laggards will continue adding AI as bolt-on features or standalone tools that create new complexity. Leaders will build AI directly into their existing operational infrastructure, where it understands domain-specific contexts and workflows. In finance, this means AI that comprehends company policies, approval hierarchies, and compliance requirements from day one—not generic tools that require extensive configuration.

Security-first architecture will separate enterprise-ready solutions from consumer-grade experiments. The leaders understand that finance teams need AI that operates within existing permission structures and audit trails. Laggards will struggle with security concerns that prevent adoption, while leaders will have built trust through transparency, explainability, and rigorous performance measurement from the outset.

Organizational AI fluency will become increasingly crucial. Leaders won’t treat AI as a technology project confined to engineering teams. Instead, they’ll cultivate company-wide AI literacy where every department understands and contributes to AI applications. This cultural shift enables faster innovation and more practical AI implementations across all business functions.

What’s particularly important in financial technology is that leaders will focus on measurable business outcomes rather than AI capabilities for their own sake. They’ll have robust performance dashboards, clear quality metrics, and continuous feedback loops that prove AI value to sceptical finance professionals.

The companies that survive the AI transformation will be those that make their customers more successful, not just more automated. In finance, that means elevating human judgment while handling operational complexity—a balance that requires deep domain expertise, not just advanced algorithms.