Independent AI and cloud cost work for companies spending $2M–$50M a year on cloud, and for AI-native teams whose inference bill just became the scariest line on the P&L. Nine years inside AWS solutions architecture; now on your side of the bill. Fixed fee, agreed before we start — never hourly, never a percentage of savings.
Bring three months of cost data. I'll show you three things you didn't know about your own bill. If there's nothing worth fixing, I'll say so.
Four measurements, each from the primary source. No vendor surveys, no numbers I can't point at.
The uncomfortable part: those two facts sit in the same room. Per-token prices really are collapsing, and bills really are climbing — because token volume per task, model-tier drift, and agentic workloads grew faster than unit prices fell. Nobody has published that decomposition with real data. That's what I'm working on.
Four ways in, in the order people usually take them.
A screen-share over your actual cost data. Three findings, live. No deck, no follow-up sequence.
Two to three weeks, read-only access, built on CUR-level billing data rather than dashboard screenshots. Every opportunity quantified in dollars and ranked, an inference unit-economics baseline, a commitment-strategy review, and a 90-day roadmap your team can execute without me.
I run the roadmap with your team: forecast reviews, commitment timing, AI spend governance as models and prices shift monthly, and a quarterly review your CFO will actually sit through. Cancel any month.
Timed to your enterprise discount renewal. Baseline analysis, commitment sizing, and negotiation support from someone who spent nine years inside AWS watching how these deals actually get made.
Full scope and operating principles on the advisory page.
There are four kinds of people who will offer to cut your AI bill. Here's the honest comparison.
| Instead of | What you get here |
|---|---|
| Percentage-of-savings firms | They are paid per dollar of attributable savings, so they rationally take the easy, machine-measurable wins and leave. I charge a fixed fee to go after the hard 30% — architecture, model right-sizing, the habits that regenerate waste every quarter. |
| FinOps analysts who aren't engineers | They read dashboards and write decks. Most AI waste lives in the architecture — context that grows quadratically, a frontier model doing a classifier's job, a batch that never batches. I ship the queries and the fixes during the engagement. |
| Cost tooling vendors | Software shows you what. Somebody still has to decide so what and drive now what. I don't compete with the dashboards — I'll tell you which one to buy, and then do the judgment layer they can't. |
| Your cloud provider's own cost team | Structurally conflicted: they optimize within their own platform and will never suggest a different model provider. I'm independent — and I know their playbook, because I sat next to the team that runs it. |
"Firms that get paid from savings are incentivised to find the easy 10% and leave. I charge a fixed fee to find the hard 30% — and to teach your team so it never comes back."
Fixed fees, in writing, before work starts. Never hourly; never a percentage of savings. Read-only access — I analyse, your team implements, and capability transfer is a deliverable, because a cost function that depends on my continued presence is a failed engagement. Realistic savings identified on an unoptimised account run in the high teens to roughly 30 percent; anyone promising 50%+ up front is selling you a baseline dispute. The full argument is here.
I'm Hussain Sehorewala. I spent nine years at Amazon Web Services in solutions architecture, where I built and ran a 40-person organisation serving Fortune 500 accounts — which means I've seen both how enterprise cloud gets architected and how it gets priced, discounted, and renewed from the inside. I hold an Executive MBA from UW Foster. I now work independently on AI and cloud cost.
The reason I publish my methods is simple: this field is about two years old, most of what's written about it is vendor marketing, and the fastest way to be trusted with a $10M bill is to show the working. Everything here follows one rule — never cite an aggregator when the primary source is one click deeper.
I never disclose confidential information from my time at AWS; what I bring is how the market mechanics work, which is public knowledge that almost nobody has assembled. How engagements work · Get in touch.
No client logos on this page, because I won't invent them. Here is the actual evidence instead.
One real cost pattern per issue, with the numbers and the query to check whether you have it. Start with "Why I don't do pay-from-savings" — it explains the whole operating model.
A GPU TCO model, a live price-per-intelligence tracker, and ai-bill-doctor — a CLI that scores an LLM provider bill. Released free under MIT as they're finished, because a working model is a better credential than a case study.
One teardown a week — real numbers, primary sources, the occasional spreadsheet nobody else will publish. No sponsors.
No sponsors, no list-sharing. Unsubscribe in one click, any issue.