Before a retained agency engagement, a buyer should be able to measure more than the quality of a presentation. The first phase needs a baseline, a testable hypothesis, explicit access, accepted deliverables, a traceable source-to-outcome path, named owners, full cost, and a reversible decision.
Define the first-phase decision
Record the problem, service line, audience, period, proposed commitment, and decision options: retain, run a project, change scope, choose another provider, or hold. State the outcome the first phase can actually influence. Keep a promised long-term result separate from the evidence the first phase can collect.
Measure hypothesis quality
For each hypothesis, capture problem statement, supporting evidence, alternative explanation, expected observation, owner, and review date. Score confidence as observed, sampled, estimated, or unknown. A hypothesis that cannot name an observation is a tactic preference, not a measurement-ready proposition.
Ask whether the hypothesis includes the offer, audience, handoff, delivery capacity, and data quality. A marketing channel can be measured correctly and still be the wrong remedy for an unclear service or slow sales response.
Measure access and source readiness
Create an access register for analytics, search, advertising, website, CRM, sales, support, finance, and stakeholder input. Record permission owner, refresh date, export path, privacy boundary, and exception. The SBA business-management guide is a reminder that management controls depend on assigned responsibility; access without ownership is not readiness.
Measure the share of required fields that are present, defined, current, and reviewable. Keep missing, not applicable, and not measured separate. Do not let an agency infer a clean baseline from a partial export.
Record the access test itself: account, permission, export, timestamp, reviewer, and fields that could not be verified. An agency may be competent while a source system is unavailable; the measurement decision should preserve that distinction and price the repair explicitly.
Add a confidence note to each baseline field and keep an explicit exception path for data that arrives late or cannot be exported.
Measure the baseline and change log
Record the starting values, period, cohort, source, calculation, and limitation for each measure. Add a change log for definitions, filters, attribution windows, tracking, and site or offer changes. A baseline that can be silently rewritten cannot support an engagement decision.
Include qualitative evidence: sales objections, customer questions, support cases, delivery constraints, and approval delays. A first phase should explain what the numbers cannot show as well as what they show.
Measure deliverables and acceptance
Give every deliverable an ID, owner, due date, source, acceptance condition, dependency, revision limit, and status. Examples include an evidence map, baseline, brief, experiment, repaired event, landing-page review, CRM handoff, or decision record.
Do not use “strategy” or “optimisation” as a complete measure. Ask what file, change, test, or decision will exist and who can accept it. Record whether the client can use the output after the retainer ends.
Measure the commercial path
Trace one path from source to meaningful action, lead acceptance, opportunity, and outcome. If events are needed, use the GA4 event guidance for technical structure and add local definitions, consent, retention, and deduplication.
For CRM stages, record lead source, eligibility, owner, acceptance, response, opportunity, disposition, and lag. The Salesforce lead implementation guide illustrates why lifecycle and ownership fields need explicit meaning. Do not call a tracked event a qualified lead or a forecast opportunity booked revenue.
Measure ownership and operating load
Track client and agency hours, expert availability, approval time, blocked dependencies, implementation effort, and response queue. Record who can approve claims, spend, creative, data access, and release. A phase can have excellent recommendations and still fail because no one can execute them.
Measure handoff quality: complete brief, clear next action, owner, timestamp, and exception. Keep a state for “waiting on client” and “waiting on agency” so a delay is not misread as lack of demand.
Measure economics and risk
Capture retainer, setup, media, production, software, travel, subcontractor, and client-side costs. Separate one-time from recurring. Record minimum term, notice, work ownership, data access after exit, and dependencies on a named person or vendor.
Use a risk ledger with likelihood description, impact, owner, mitigation, and stop trigger. Do not assign numerical probabilities without evidence. A qualitative risk with a clear stop path is more useful than a precise but invented score.
Use the evaluation ledger
| Layer | Core measures | Required qualifier | | — | — | — | | hypothesis | problem, evidence, alternative, test | confidence and owner | | access | system, field, permission, refresh | privacy and exception | | baseline | value, cohort, period, source | formula and limitation | | delivery | artifact, date, acceptance, dependency | usable after exit? | | commercial path | event, lead, opportunity, outcome | stage and lag | | ownership | approver, responder, escalation | blocked-state reason | | economics | full cost and term | recurring vs one-time | | risk | trigger, impact, mitigation | reversible next step |
Set the retain-or-hold gate
Retain when the first phase has a testable hypothesis, reliable-enough access, signed baseline, accepted deliverables, traceable commercial path, available owners, declared economics, and an exit condition. Choose a project or diagnostic when only part of the chain is ready. Hold when the agency’s promised outcome is outside its control, measurement definitions are vague, access is ownerless, or the cost cannot be reversed.
Keep the ledger local and non-indexable until provider overlap, contract, source, technical, and editorial review are complete. Measurement is ready when it improves the buying decision, not when the agency has filled every dashboard tile.
How did this article land?
Choose one reaction. You can change it anytime.
