Business model
We charge our cost and earn a success fee. The client pays less than the army would have charged, and we are paid when the work is accepted.
Pricing mechanics
Assessment, at cost
Six to twelve weeks depending on size, fixed fee equal to our cost including contingency. The client gets an evidence base, a semantic model of how the business runs, a recommended path and a fixed-fee delivery quote, and keeps all of it even if it never hires us for delivery.
Delivery, fixed fee at cost
Our delivery cost (expert time, inference, tooling and 15% contingency) passed through with no margin, billed monthly as incurred.
Success fee
15% of what the client's current path would have cost to deliver: their SI-led SAP-to-SAP program, or the actual SI budget if they share it. Paid only on acceptance, in tranches at agreed milestones such as rehearsal passed, go-live and hypercare exit. This is how we make money.
We can also run as a quiet parallel track beside an existing program. The client keeps its main path, we prove out the alternative, and either way the client ends up with a far better map of its own systems.
Why break-even pricing
- It removes the buyer's main objection. The usual worry about a new delivery model is that it costs more or does not work. There is no margin in the fixed fee to argue about, and the success fee is paid only when the client accepts.
- It aligns us to outcomes. We earn nothing extra from extra hours. Finishing faster and cheaper raises our return on the same success fee.
- It prices below the army before any savings. Our fixed fee for the large scenario is $11.8M against $24M for the current path's delivery services; adding the success fee and assessment still totals $16.2M.
Unit economics per scenario
Rescue path, thesis preset. Our gross profit on an engagement is the success fee, because the fixed fee is at cost.
| Scenario | Fixed fee (at cost) | Success fee | Client total | Client savings | Our gross profit |
|---|---|---|---|---|---|
| Mid-market | $3.5M | $1.1M | $5M | $2.2M | $1.1M |
| Large | $11.8M | $3.6M | $16.2M | $7.8M | $3.6M |
| Very large | $32.5M | $9.9M | $43.7M | $22.3M | $9.9M |
| Global | $154M | $46.8M | $203M | $109M | $46.8M |
Client total includes the at-cost assessment. Savings are measured against the current path's delivery services. Planning estimates, not quotes. For switch and freedom paths the client's up-front savings are smaller or negative; their case rests on five-year savings and strategic value, shown in the estimator.
The fixed-fee lines are not revenue we are relying on to cover overhead. Because they pass through cost, company overhead and the platform build are paid for from success fees, and from the raise until the first one lands.
Inference cost
An earlier internal budget allowed $7,200 of AI tokens against about $1.8M of labor. That understates how agents are used. Our model budgets about 70B tokens per AI-enabled engineer-month (a working figure; our range is 60 to 80 billion), which at a Claude routing mix is about $27K per engineer-month. Agent cost is dominated by re-reading context rather than writing output, so the price of cached input matters most.1 Provider choice moves the figure widely.
| Provider or routing mix | Inference per engineer-month |
|---|---|
| Qwen3-Coder (DeepInfra, open-weight) | $8K |
| DeepSeek V4 Pro (peak) | $10K |
| Opus 5.5 plans and reviews, DeepSeek V4 Pro (DeepInfra) executes | $21K |
| Claude Sonnet 5.5 | $22K |
| Opus 5.5 plans and reviews, Sonnet 5.5 executes (base case) | $27K |
| Claude Opus 5.5 | $43K |
| Claude Fable 5.1 | $92K |
| GPT-6 Astra | $141K |
List prices per million tokens checked 3 November 2026.2 The model applies a cache-heavy token mix; see how it works. Inference is part of the cost we pass through, so a pricier model raises the client's fixed fee and cheaper routing makes our price more competitive. We route by task and meter by agent and engagement.
Sensitivity
The estimator has three presets. Conservative assumes agents absorb less of the work, seniors are 1.5 times as productive as the integrator's pyramid, and review takes more hours. The thesis preset assumes 2 times. Aggressive assumes 3 times. Presets change our cost, and therefore the fixed fee and the client's savings. Our own profit moves far less, because the success fee is set by the client's current path and not by our cost. The exposure is on the downside: under a fixed fee, cost beyond the contingency is ours to absorb, which is why the contingency, scope control and the week-12 and later gates matter. Every assumption is editable in the estimator.
Cash timing
Fixed fees are billed monthly as incurred, so delivery work does not require us to fund a program. Success fees are lumpy, which is why they arrive in milestone tranches rather than a single payment at the end. The raise of $5M carries the company through the platform build and the runway to the first success fee ($1.1M). Receivables and collection terms are agreed in the contract, and are a diligence item.
Longer-term upside (not in the base case)
Every engagement should leave reusable agents, connectors and semantic models that lower our cost per engagement. Where we can reuse non-client-specific assets, a lower cost lowers the client's fixed fee and widens our room to win. A software subscription for the agent workbench is possible later. It is an upside option, and the base case assumes no subscription revenue.
Sources
- GitHub Copilot coding-agent telemetry, 13.5M sessions, June 2026. arXiv 2608.00101.
- Anthropic, pricing; DeepSeek, API pricing; OpenAI, API pricing. Checked 3 Nov 2026.