The Measurement Agreement

What this document is

Most website and marketing spend is unauditable: when results move, nobody can say whether the work caused it. This one-page agreement exists so that both of us will be able to tell. Before any work begins, we agree — in writing — on one metric, its current baseline, the window we will compare, how we will compare it, and what each of us will conclude if the number comes back flat. It is signed by both parties before the build, and neither party edits it afterwards.

The one metric

[e.g. New booked appointments from new clients, counted in the practice booking system]

Why only one, and why this one

One primary metric, at the money end of the business — not a dashboard of nine. The test we apply: you can log into your own system and count it, without any tool of ours in the way. The metric is chosen for decision-relevance to your business, never for how good it would look published.

The baseline

[e.g. 38, 41, 35, 44 … per month over the last 12 months, from the booking system — monthly counts listed in full]

Whose numbers count

Measurement uses your records, in your accounts, readable without us: your Google Business Profile, your phone records, your booking or CRM system, your revenue — in that order of preference, and ranked by trust rather than convenience. Never analytics we installed: anything running on infrastructure we configured is our number, not yours, and grading our own work with our own instrument would reproduce the exact problem this document exists to solve.

The window

[e.g. The six months from launch, compared year-over-year to the same six months]

The comparison method

[e.g. Year-over-year, same window — removes seasonality. Where history allows: organic/local channel segmentation to isolate the surface changed]

The confounder log

• [e.g. A second practitioner joins in March]
• [e.g. The competitor on Main Street has announced closure]

Why we log confounders now

A new hire, a competitor closing, a planned price change — things both of us can already see coming that would move the number regardless of the work. We write them down before the work starts so that the after-measurement conversation is defensible instead of arguable, in both directions. Discovering a confounder later is allowed; retro-fitting one to explain an inconvenient result is what this log prevents.

The lift commitment

The subscription carries a performance term: a 15% year-over-year improvement in the agreed metric over the agreed window. At three months we produce a full measurement report against the baseline — direction of travel and leading indicators — with no fee consequence. At six months the test binds: if the lift is not shown, the monthly fee stops until it is. While the fee is paused, all work continues — hosting, maintenance, and the ongoing distribution work — because that is the only way the lift arrives and the fee resumes. At twelve months, if the lift still has not been shown, both parties review together: continue, change the metric, or part ways.

An honest note on the threshold

The 15% figure is a commercial commitment, not a statistical claim. For a single business measured in monthly counts, a shift of this size cannot always be distinguished from ordinary variation with statistical confidence — anyone who tells you otherwise is overclaiming. We state this plainly here because the alternative — pretending the measurement is more certain than it is — is exactly the practice this document exists to replace. The commitment stands regardless: if the number is not shown, the fee stops.

If the number comes back flat

[e.g. If flat at six months: the fee pauses; sg presents what it would change and does that work without a fee until the lift shows. Neither party treats a flat number as proof the work was bad — the confounder log is reviewed first, in both directions.]

Publication

[e.g. Results may be published as a case study naming the practice / anonymised as “a Halifax dental practice”]

Why we ask now

Publication rights are agreed at signature — with a named anonymised fallback — because asking after the results are in puts one party in a negotiation the other controls. Agreeing now, either way, costs nothing.

Document version

This agreement was produced from fixed template version 1.0.0, published at sheigoldberg.com/measurement-agreement. Acceptance records the version, so both parties can always establish exactly which terms were agreed.