Electronics Era: Agentic AI workloads introduce highly volatile compute costs. From a CFO’s lens, how do you balance innovation velocity with financial discipline?
Ankit Sarawagi: Agentic AI looks exciting right up to the moment costs spike without warning. A small workflow change or one customer pushing usage harder than expected can move the bill overnight. From a finance lens, speed only works when limits are clear. At Verloop, teams move fast, but nobody moves blind. Every experiment has an owner, a budget, and a check-in date. That clarity does not slow people down. It actually removes hesitation. Engineers know where the line is, so they stop second guessing.
Unexpected overruns change behaviour in the wrong way. People become defensive and decisions slow down. Clear guardrails avoid that. When costs are visible and expectations are set early, teams stay confident and momentum stays intact. My role is to keep curiosity alive without letting costs drift quietly. When both happen together, speed feels sustainable rather than stressful.
Electronics Era: How different is AI infrastructure cost management compared to traditional SaaS hosting models?
Ankit Sarawagi: Traditional SaaS costs follow a rhythm. Usage grows, infrastructure scales, and spend moves within a range you can plan for. AI infrastructure behaves very differently. Costs react directly to how people use the product. Prompt quality, retries, edge cases, and product shortcuts all show up on the bill. Teams are often surprised by how fast this adds up.With AI, cost decisions start inside product design. Model choices, retry limits, caching, and batching are not just technical details. They decide margins long before finance sees an invoice. That is why finance needs to be part of product conversations early. Once everyone understands how usage behaves in real conditions, cost discussions feel shared instead of restrictive. That shift changes the relationship between product and finance completely.
Electronics Era: What financial guardrails or internal metrics do you put in place to prevent margin erosion as AI usage scales unpredictably?
Ankit Sarawagi: No AI feature goes live without clear numbers attached. Cost per action, cost per outcome, and cost per customer over time are mandatory. If the economics are unclear, the feature stays on the drawing board. We track margins at the feature level. Blended views hide problems. One inefficient flow can quietly eat into margins if you do not spot it early.
