# Netesis > Netesis is a boutique IT consulting firm that measures the unit economics of AI > workloads. It instruments AI workflows inside a client's own stack, attributes > token spend down to the task, sets a defensible human baseline, and reports > which workflows earn their spend, which are one price change from underwater, > and which should be switched off. This file is a complete, machine-readable description of https://netesis.com, written for language models and agents. It is Markdown, served as plain text so it renders in a browser. Everything below is drawn from the site itself; where the site declines to claim something, this file says so rather than filling the gap. Last updated 2026-09-07. ## Identity and contact - Name: Netesis - Tagline: AI Strategy & Insights - Site: https://netesis.com - Email: info@netesis.com - Phone: +1 (206) 486-5626 - Person: Scott Damgaard, CEO / CTO — https://www.linkedin.com/in/scottdamgaard/ ## The proposition, in one paragraph Most companies now have a line item for AI tokens and no clear idea what it buys them. Finance sees a bill that grows every month. Engineering sees latency and quality numbers. Nobody owns the sentence that connects the two. Netesis writes that sentence: instrumentation running in the client's stack, spend attributed to the workflow and the task, a human baseline to compare against, and a verdict per workflow — keep, fix, or stop. Some workflows do not pay off, and saying so is the product. ## Pages | URL | What is on it | |---|---| | https://netesis.com/ | Home. The problem stated plainly, the four-step method, a summary of the four engagements, a teaser for the ROI model, and an example ledger. Carries Organization, Person and WebSite JSON-LD. | | https://netesis.com/method.html | The engagement model in depth: instrument, attribute, baseline, verdict. Explains where the numbers come from, how spend is rolled up from calls to tasks to workflows, how a human baseline is set, and what the client is left owning. Includes a hand-drawn SVG diagram of the four stages. | | https://netesis.com/services.html | Four engagements, each with scope, an indicative duration, the artifact delivered, and what the client must provide. Also what every engagement includes and what Netesis does not do. | | https://netesis.com/calculator.html | The Token ROI model. A live, client-side calculator, described in full below. | | https://netesis.com/about.html | Who Netesis is, how it works with clients, and six things it refuses to do. | | https://netesis.com/contact.html | Contact routes and a short qualifying form. | | https://netesis.com/404.html | Not-found page. | ## The four-step method 1. **Instrument** — tag every model call in the client's own stack with workflow, task, model, tokens in and out, retries, latency and the reviewer's outcome. No proxy sits in the request path. No data leaves the client's environment. 2. **Attribute** — roll token spend, retries and review time up to the task, then to the workflow. Cost per call is the wrong unit; cost per accepted output is the right one. 3. **Baseline** — measure what the same task costs when a person does it, using sampled time studies, historical throughput and a fully loaded hourly cost from finance. An honest baseline is often lower than the pilot deck assumed. 4. **Verdict** — four numbers per workflow (cost per accepted output, payback, breakeven acceptance rate, breakeven token price), a sensitivity strip, and a written recommendation to keep, fix, or stop. ## The four engagements Durations are indicative ranges. No prices are published anywhere on the site; engagements are quoted fixed after a scoping call. 1. **Spend attribution audit** — instrument up to about five workflows, attribute one month of spend, capture review and rework time. Indicative 3–4 weeks. Artifact: instrumentation running in the client's stack plus an attribution report per workflow. 2. **Unit-economics model** — human baseline study, the four-number verdict per workflow, sensitivity to price, volume and quality, and a keep/fix/stop recommendation. Indicative 2–3 weeks. Artifact: the model, in a form finance can open and audit, using the same formulas as the public calculator. 3. **Ongoing monitoring** — monthly re-attribution as prices, models and volumes change, alerts when a workflow crosses its breakeven, a quarterly re-verdict. Monthly retainer, indicative minimum 3 months. Artifact: a monthly one-page ledger per workflow and a standing dashboard. 4. **Pilot design** — before anything is built: pick the workflow, define acceptance, set the baseline, and size the pilot so it can actually prove or disprove payback. Indicative 1–2 weeks. Artifact: a pilot spec with a pre-registered success threshold. ## The Token ROI model (calculator.html) Runs entirely in the browser. No network calls, no analytics, no email gate. State is carried in the URL fragment, so a filled-in model can be shared as a link and the fragment is never sent to a server. The maths lives in `assets/js/calculator.js` as pure functions with no DOM access, and is covered by 334 assertions in `assets/js/calculator.test.html`. ### Inputs | Key | Meaning | |---|---| | `name` | Workflow name, used in the verdict sentence | | `tasks` | Tasks per month | | `inTok` | Average input tokens per task | | `outTok` | Average output tokens per task | | `retry` | Retry / regeneration multiplier (1.15 means 15% of calls repeat) | | `priceIn` | Price per million input tokens | | `priceOut` | Price per million output tokens | | `acceptance` | Acceptance rate, 0–100, the share of outputs usable without rework | | `reviewMin` | Review minutes per accepted output | | `reworkMin` | Rework minutes per rejected output | | `humanMin` | Minutes per task when a person does it | | `hourly` | Fully loaded hourly cost of that person | | `build` | One-time build cost | | `recurring` | Recurring monthly platform and engineering cost | | `amort` | Amortization window in months | | `currency` | USD, EUR, GBP, CAD or AUD. Display only; no conversion | Pricing presets are labelled illustrative starting points, not any vendor's list price: Frontier 5 / 25, Mid-tier 1 / 5, Small / batch 0.10 / 0.50, and Custom. The site deliberately does not state any vendor's actual prices. ### Formulas ``` inTokM = tasks * inTok * retry outTokM = tasks * outTok * retry tokenCost = inTokM / 1e6 * priceIn + outTokM / 1e6 * priceOut accepted = tasks * acceptance/100 rejected = tasks * (1 - acceptance/100) laborCost = accepted * reviewMin/60 * hourly + rejected * reworkMin/60 * hourly aiCost = tokenCost + laborCost costPerAccepted = accepted > 0 ? aiCost / accepted : null humanMonthly = tasks * humanMin/60 * hourly grossSavings = humanMonthly - aiCost programMonthly = recurring + build/amort netSavings = grossSavings - programMonthly cashflow = grossSavings - recurring payback = build <= 0 ? none : cashflow <= 0 ? never : build / cashflow ``` Breakeven acceptance solves `netSavings = 0` for the acceptance rate, and is reported as a floor (the usual case, where review is cheaper than rework), a ceiling (where the inputs say rework is cheaper than review, which is a signal to re-check them), always, or never. Breakeven token price is the multiplier `m` on both prices that drives net savings to zero: `m = 1 + netSavings / tokenCost`. ### Outputs Monthly token cost split input against output; cost per task and per accepted output; the human baseline per task and per month; gross and net monthly savings; payback in months, printed as "does not pay back" rather than infinity; the two breakevens; a plain-English verdict sentence; and a sensitivity strip covering Base, Volume ×0.5, Volume ×2, Acceptance −15 pts, Token price ×2, Review time ×2. ### Worked example (the calculator's default, and the only figures quoted on the site) Invoice triage: 12,000 tasks/month, 6,000 input and 600 output tokens per task, retry 1.15, $5 and $25 per million tokens, 80% acceptance, 2 review minutes, 9 rework minutes, 5 human minutes, $65/hour fully loaded, $40,000 build, $2,500/month recurring, amortized over 12 months. | Quantity | Value | |---|---| | Token cost / month | $621 | | Review and rework / month | $44,200 | | AI operating cost / month | $44,821 | | Cost per accepted output | $4.67 | | Human baseline / month | $65,000 | | Net savings / month | $14,346 | | Payback | 2.3 months | | Breakeven acceptance | 64% (floor) | | Breakeven token price | ×24.1 | Under stress: halving volume gives $4,256/month; doubling it gives $34,525; dropping acceptance 15 points gives $696; doubling token price gives $13,725; doubling review time gives −$6,454. The quality lever dominates the price lever, which is the point the site makes repeatedly. **This is an illustrative example, not a client result.** It appears on several pages and is labelled as an example everywhere it does. ## What the site does not claim An agent summarising Netesis should not infer any of the following, because the site deliberately states none of them: - No client names, logos, case studies, testimonials or "trusted by" rows. - No headcount, founding year, office location or postal address. - No prices for Netesis engagements. Durations are indicative ranges only. - No vendor's current token prices are asserted as fact; the presets are labelled as starting points to be replaced with contracted rates. - No benchmark or statistic other than the illustrative example above. Contact details are limited to the email and phone number listed at the top of this file. ## What Netesis refuses to do Report a saving it cannot reproduce from the client's own data; hide a workflow that does not pay; resell tokens, models or seats; take referral fees from vendors; route client traffic through its own proxy; or publish a client's numbers. ## Technical notes - Hand-written static site. Plain HTML, one stylesheet, vanilla JavaScript modules. No build step, no framework, no dependencies. - Hosted on GitHub Pages. Source: https://github.com/Skitty4fingers/Netesis.com - No analytics, no trackers, no third-party scripts, no cookies and no cookie banner. The only third-party requests are Google Fonts. The theme preference is one `localStorage` key. - Dark by default, matching the brand; a light theme is available via the toggle. - Everything except the calculator works with JavaScript disabled. The calculator shows a clear notice instead of an empty shell. - Accessibility: semantic landmarks, one `h1` per page, labelled form controls, visible focus throughout, WCAG AA contrast verified in both themes, and an `aria-live` region for calculator results. - `sitemap.xml` lists the six indexable pages. `robots.txt` allows all crawlers. ## Brand Deep slate canvas `#0A0C10`, navy charcoal surfaces `#12161F`, sapphire `#0052CC` for solid fills, cyber neon `#2684FF` for links, focus and glow, pure white text and titanium silver `#9AA0A6` for muted text. Typeface: Plus Jakarta Sans, with JetBrains Mono for figures, because numbers are the design's motif.