About Netesis
Engineers, not a research shop.
Netesis is a boutique IT consulting firm that does one thing: the unit economics of AI workloads. The deliverable is instrumentation running in your stack and a model your CFO will sign, not a strategy deck.
Who we are
A boutique firm with one subject.
The firm
What we do, and what we do not
We measure whether an AI workflow earns its spend. Cost per task, cost per accepted output, payback, breakeven acceptance, and breakeven token price, per workflow, from your data. If your question is not one of those, we will say so and point you elsewhere.
We are engineers. The work is SDK-level tagging, warehouse queries, a model with every formula visible, and a written verdict. We do not write strategy decks, run innovation workshops, or draw three-year roadmaps.
You will not find a headcount or a founding year here. Neither changes whether your workflow pays. What you are buying is how we work and what we refuse to do, and both are below.
Team
Who does the work
Scott Damgaard
AI workloads raise an old infrastructure question with a meter attached: what does this cost to run, and what breaks when it scales. Scott spent fifteen years answering it as an infrastructure architect, on estates from 50 seats to 500,000 — cloud and hybrid architecture, automation and disaster recovery, much of it under financial-services compliance and uptime obligations. Netesis points the same discipline at tokens.
How we work with clients
A small team, inside your stack, in writing.
An engagement is code and queries, not meetings. This is the shape of one, from the first week to the handover.
-
01
A small team
Sized to the workflows in scope, not to a bench. The people who scope the work are the people who do it.
-
02
Inside your stack
We work in your repo and your warehouse, under your access controls. Nothing is copied out, and no proxy sits in your request path.
-
03
Weekly, in writing
Every week you get a written update: what was measured, what changed, what is blocked, and what we now believe. No slide deck.
-
04
Yours to keep
Instrumentation, schema, dashboards, the model, and the runbook stay in your stack when we leave. Your people are trained to rerun the verdict without us.
What we refuse to do
Six things we will not do.
These are conditions of the engagement, not preferences. If one of them is a problem, we are the wrong firm, and it is better to know that now.
-
01 · Refusal
Report a saving we cannot reproduce from your data.
Every number in the verdict traces back to a query on your warehouse. If we cannot rerun it in front of you, it does not go in the report.
-
02 · Refusal
Hide a workflow that does not pay.
If a workflow loses money, the report says so, on the first page, in the same table as the ones that pay. That sentence is what you are paying for.
-
03 · Refusal
Resell tokens, models, or seats.
Our fee is the engagement. Nothing we recommend earns us a margin on a vendor's invoice, so we have no reason to see you buy more of anything.
-
04 · Refusal
Take referral fees from vendors.
No affiliate deals, no partner tiers, no kickbacks. A recommendation to switch models or providers is a number, not a relationship.
-
05 · Refusal
Run your traffic through our proxy.
Instrumentation lives in your code and the data lives in your warehouse. We never sit in the request path, and your prompts, outputs, and volumes stay in your stack.
-
06 · Refusal
Publish your numbers.
No case study, no logo wall, no anonymised benchmark that is not quite anonymous. If you choose to be a reference, that is your call, in writing.
Where we come from
A point of view, not a founding story.
The pattern
Funded on token price, killed by review time
We have watched pilots get funded on token price and die on review time. The deck quoted a per-call price, the demo looked good, and nobody counted the minutes a person spent checking each output before it could be used.
In the example scenario in our model, a task costs $3.74 all in. Tokens are $0.052 of that. The other $3.68 is a person reviewing and reworking the output. Double the token price and net savings fall by $621 a month. Double the review time and the same workflow runs at −$6,454 a month.
Figures from the example scenario in the ROI model, not a client result.
That is the point of view. Quality and review time decide whether an AI workflow pays. Price is the lever everyone argues about and the one that matters least. We built a firm around measuring the first thing instead of debating the second.
Token price is the number on the invoice. Cost is what happens after the output arrives.
Start with one workflow and last month's token bill.
A 30-minute scoping call, then a written note on whether an engagement makes sense. Sometimes it does not, and we say so.