About — Bluetinto

About

I help companies stop overpaying for AI.

I'm Chris Echevarria — a senior finance controller turned independent AI cost analyst, working where IT, finance, and technology meet. I help companies turn an unpredictable, ballooning LLM bill into something measured, defensible, and dramatically smaller.

My background is 11+ years of IT and finance controlling across Booking.com, Heineken, VodafoneZiggo, T-Mobile, and Ahold Delhaize — leading the financials on global transformation and SAP S/4 programs, building executive investment cases, and steering €120m+ CapEx and OpEx technology portfolios. I've spent that career on automation, RPA, and — increasingly — AI initiatives. That combination is the whole point: optimizing AI spend isn't a pure engineering problem or a pure finance problem; it's both. I stay model-agnostic and vendor-neutral by design, so my only incentive is the lowest real cost at the quality you need.

Every recommendation is grounded in current pricing, your actual token volumes, and reproducible math — not vibes. You get inputs you can re-run yourself.

How I work

Measured, neutral, reproducible.

Neutral

No vendor agenda

Independent of every provider. The recommendation is whatever costs least at your required quality — full stop.

Evidenced

Numbers, not narrative

Every figure traces back to published pricing and your real volumes, with assumptions stated so you can audit them.

Reusable

You keep the model

You leave with a cost model your team can re-run as prices, models, and usage shift — not a dependency on me.

What I believe

The cheapest token is rarely the cheapest task.

A model that costs half as much per token but is twice as verbose — or needs three retries — costs you more per finished task. Headline price is a trap; cost-per-outcome at fixed quality is the number that matters.

That's the lens I bring to every engagement: quality-adjusted, workload-specific, and honest about the trade-offs. Sometimes the answer is a cheaper model. Sometimes it's a smarter architecture on the same one. Occasionally it's self-hosting — but only when the math genuinely supports it.

Work with me → See services