No upfront fee. No retainer. You pay from what we can prove we saved you.

The short version

• One metric, agreed upfront. Before any build starts, we agree in writing on the single business metric the system exists to move: cost per inspection, nightly revenue per available listing, churn rate, tickets resolved without a human.
• A documented baseline. We measure the current state from your own data before we build anything, so "improvement" has a number attached, not an anecdote.
• You pay a share of the measured lift. Our fee is a pre-agreed percentage of the measured savings or revenue increase over the baseline, calculated from your data, with a cap so the fee can never become unreasonable.
• No improvement, no performance fee. If the metric doesn't move, you don't owe a performance fee. That risk sits with us, on purpose.
• Work that can't be outcome-measured is quoted fixed-scope. Infrastructure, migrations, and exploratory research get a fixed scope and price upfront — we don't disguise hourly billing as performance pricing.

We concentrate on operations where the saving is verified by somebody other than us: a carrier who decides whether to pay a freight claim, a control store the model never touched, a settlement price published hourly. That is what separates this from a marketing slogan. In practice that means freight and logistics, retail and grocery, and energy and utilities - though we will work anywhere the outcome is genuinely measurable.

The Founding Client Program

Here is the part most consultancies would hide: we do not have case studies yet. We are new, and we are not going to invent client results or dress up industry averages as our own track record.

That is the entire reason these terms exist. A firm with ten published case studies can charge a retainer, because its portfolio is the proof. We do not have that, so we offer the only other honest form of proof: we carry the risk ourselves until the results exist. What you get:

• A free profit-leak audit. We review your data and hand you a written measurement plan naming the metric, the baseline, the attribution method, and what we believe we can move it by. It is yours to keep whether or not you hire us - including if you hand it to another firm.
• No upfront build fee, and no retainer. We carry our own build cost. You cover only hard infrastructure costs such as cloud spend, disclosed in writing before you commit to anything.
• Payment from measured improvement only, as a capped share, computed from your own systems, against a holdout or control group wherever volume allows.
• You own everything we build - model, code, and documentation - at handover. No black box, no hostage situation.
• Ninety days of monitoring and retraining included after launch, because a model that quietly decays is not a result.
• Our ask in return: if it works, a named case study and a reference call. That is what we are actually buying with these terms, and we would rather say so plainly than pretend it is generosity.

We run three performance-based engagements at a time. That is not a marketing device - it is the honest capacity of a small team that has to carry delivery risk on every project it accepts.

Who this is for, and who it is not

We turn down projects in the second column, and we would rather do it on a free call than after taking your money. Under performance pricing, a project that cannot work costs us more than it costs you.

What it costs you to find out

One call and a look at your data. If the honest answer is that you should buy an off-the-shelf tool, fix a process, or use a spreadsheet, we will tell you that and you will have lost an hour. If the answer is that there is real money in it, you will have a written measurement plan and a build that costs you nothing until it works.

Why we price this way

Most ML consulting is billed hourly or on retainer. Under that model, the consultant gets paid the same whether the model ships and works or dies in a notebook — and industry surveys have put the failure rate of ML initiatives well above half. Hourly billing means the client carries all of that risk.

Performance pricing inverts the incentive. We only earn well when your metric actually moves, which changes our behavior in three concrete ways:

1. We say no more often. If your data can't support the model, or the honest answer is "use a spreadsheet," we tell you at scoping — because taking the project would cost us money, not make it.
2. We optimize for the metric, not the deliverable. A slide deck and a trained model are not outcomes. A measured drop in your cost per unit is.
3. We stay after deployment. Models decay. Since our fee depends on sustained measured improvement, monitoring and retraining are in our interest, not an upsell.

How an engagement runs

1. Scoping call (free). We identify the metric, look at whether your data can plausibly move it, and say so honestly — including when the answer is that you don't need us.
2. Baseline and measurement agreement. Before building, we document the current value of the metric, the measurement window, the attribution method, the fee percentage, and the cap. This is a short written document both sides sign.
3. Build and deploy. Typically 6–12 weeks depending on scope. You see progress against the metric, not just status reports.
4. Measurement window. The system runs, and improvement is measured on your data with the method agreed in step 2.
5. Settlement and review. The fee is calculated from the measured lift, capped as agreed, and reviewed at a regular cadence. If the lift isn't there, the performance fee is zero.

How improvement is measured

Attribution is where outcome pricing usually gets slippery, so we fix the method before the build, in writing:

• Holdout or A/B comparison where volume allows — for example, pricing a random half of a rental portfolio with the model and half with the incumbent method. This is the cleanest evidence and our default.
• Before/after against the documented baseline where a controlled split isn't practical — adjusted for seasonality and for named external factors (a market-wide demand collapse is not the model's fault, and a market-wide boom is not its achievement).
• Your data, your dashboards. Every number in the fee calculation comes from systems you control and can audit. If you can't verify it, we don't bill on it.

When performance pricing applies — and when it doesn't

Performance pricing is honest only when three conditions hold:

• The outcome is measurable in money or a metric that converts to money (revenue per available night, cost per inspection, support cost per ticket).
• There is enough volume to separate the system's effect from noise within a reasonable window.
• The improvement is attributable — the system is the main thing changing, or we can run a controlled comparison.

When those conditions fail, we say so and quote fixed-scope instead. Typical fixed-scope work: MLOps infrastructure and deployment pipelines, data platform migrations, exploratory feasibility studies, and compliance-driven projects. Pretending that work is performance-priced would just hide an hourly rate inside a formula.

An illustrative example

How risk is shared