Direct answer: a credible agency case study identifies the creator and consent status, measurement period, baseline, outcome definition, services changed, costs, data source, and important limitations. It does not establish what another creator will earn. If publishable evidence is unavailable, the responsible choice is to explain the evaluation method rather than label an invented scenario “real results.”
1. Apply the case-study evidence standard
Start with a written research question. “Did revenue grow?” is incomplete. Ask whether a defined net or gross measure changed over a defined time window after a documented set of operational changes, compared with a stated baseline, while accounting for costs and known events. Record who selected the example and why. A success chosen because it is unusually strong has selection limits that must be disclosed.
| Field | Minimum disclosure | Weak substitute |
|---|---|---|
| Identity and consent | Verified internally; explicit publication permission | Anonymous screenshot with no provenance |
| Denominator | Gross fan payments, creator receipts, or net income defined | “Revenue” without basis |
| Time window | Exact baseline and comparison dates | Best month compared with unknown month |
| Intervention | Services, timing, creator work, and external events | “Joined agency” as the only explanation |
| Verified artefacts | Statements, ledger, invoices, campaign logs, consent record | Cropped dashboard image |
| Selection limits | Why this case was chosen and what is not represented | Implication that it is typical |
2. Define the denominator and time window
Choose one primary measure and define every deduction. Gross fan payments, platform-adjusted creator receipts, agency billings, and creator net income answer different questions. If the headline uses gross payments, show platform deductions, refunds, agency fees, production, promotion, staff, and tax limits separately. Do not compare a pre-agency net figure with a post-agency gross figure.
Use dates long enough to include onboarding, renewals, refunds, and normal variation. Note promotions, collaborations, viral posts, account restrictions, seasonality, price changes, content volume, and time worked. Avoid annualising a brief spike. If the account had no stable baseline, say so and use a descriptive case rather than a causal comparison.
3. Verify artefacts without exposing sensitive data
A reviewer should trace totals from source statements to the published table. Keep original exports, a calculation sheet, data dictionary, version history, invoice evidence, and approval. Redact subscriber identity, legal names, banking details, addresses, and unnecessary account information. Redaction must not remove the dates, units, denominator, or context required to assess the claim.
Screenshots are secondary artefacts because they can omit filters, dates, and reversals. A testimonial can explain experience but cannot substitute for financial records. If the creator withdraws publication consent, follow the agreement and applicable law. A case study should have a named review date and be removed when its evidence or permission can no longer be maintained.
4. Separate observation from attribution
“The measure increased after the engagement began” is an observation. “The agency caused the increase” is an attribution claim requiring stronger support. Document every material change: content frequency, pricing, platform feature, promotion channel, creator availability, staff coverage, audience mix, and external event. Where several changed together, do not assign the result to one service without evidence.
Use careful language: “during the reviewed period,” “the records show,” “the case cannot isolate,” and “not representative of all accounts.” Avoid “proof,” “typical,” “guaranteed,” and “what you can expect” unless the evidence supports those exact claims. The ACCC says objective business claims should be accurate, based on reasonable grounds, and provable.
5. Include costs, workload, and adverse outcomes
A useful case study reports the creator’s production obligations, operating hours, agency fee, direct campaign costs, contractors, tools, refunds, and other material inputs. It should record complaints, boundary changes, access incidents, or platform restrictions when relevant. Higher gross revenue with substantially higher cost or workload is a different result from higher net income with stable workload.
Publish negative or neutral learning where consent permits. A discontinued experiment can show a safer decision process. Document the stop rule, evidence, and remediation. Excluding every non-success turns a case series into promotion rather than a balanced evidence library.
6. Blank review scorecard
- Provenance: can an authorised reviewer trace every figure to source records?
- Comparability: are baseline and result periods measured on the same basis?
- Completeness: are fees, costs, refunds, labour, and constraints visible?
- Attribution: are other explanations documented and causal language limited?
- Consent and privacy: is publication authorised and data minimised?
- Representativeness: are selection method and limits clear?
- Maintenance: is there a review owner, date, correction path, and removal trigger?
Score evidence quality, not the size of the outcome. A modest result with traceable records can be more useful than an extraordinary screenshot. Compare the economics using the net-income worksheet and the decision risks using the agency-versus-solo framework.
7. Questions to ask an agency
Ask how many eligible engagements existed in the review period, how the example was selected, whether the denominator is gross or net, which dates are included, who performed each service, what the creator contributed, which costs are excluded, whether refunds are final, and whether an independent person reconciled the records. Ask to see evidence under an appropriate confidentiality process rather than demanding another creator’s private dashboard.
Then inspect the contract. Case-study access does not correct unclear fees, account ownership, data access, authority, or exit. Use the Australia-focused contract checklist for those issues.
8. Use this case-study analysis template
Copy the following structure before reading the headline result. First, write the decision the case study is supposed to inform: for example, whether a defined messaging service reduced unanswered conversations without increasing complaints during a stated period. Second, identify the eligible population. Was the example selected from every client onboarded that quarter, only clients who completed onboarding, or a hand-picked success? Third, list the source records and the person who reconciled them. Finally, write the limits before the conclusion. This order reduces the temptation to design the method around an attractive screenshot.
The analysis record should contain a case identifier, consent status, publication scope, baseline dates, comparison dates, account currency, denominator, deductions, creator workload, agency workload, interventions, outside events, adverse outcomes, missing data, reviewer, and review date. Add a correction field so a refund, chargeback, late invoice, or withdrawn permission can be reflected later. Keep the public case concise, but retain the complete controlled record behind it.
Blank interpretation block
During [comparison dates], [defined measure] was [recorded change] relative to [baseline dates]. Records reviewed were [sources]. The engagement changed [documented services], while [other material factors] also changed. Included costs were [items]; excluded or unavailable costs were [items]. This selected case [selection reason] and is not evidence of a typical or guaranteed result. The review cannot isolate [known attribution limits].
9. Run denominator and time-window checks before calculating change
Reconcile the numerator and denominator line by line. If a page says an account “doubled,” ask what doubled: subscriber count, gross fan payments, platform-adjusted receipts, agency-attributed sales, or creator net income. Check whether the starting value was unusually low, whether the ending value includes a temporary promotion, and whether both periods use the same currency and refund treatment. A percentage change without the underlying values can exaggerate a small base, so retain both amounts even if the public version uses a percentage.
Use comparable windows. Match the number of days, day-of-week mix, billing cycle, and treatment of incomplete transactions where practical. A launch week should not be compared with a quiet partial week and described as a stable operating improvement. For recurring subscriptions, include enough time to observe renewals and cancellations. For promotion, record spend, reach, tracked visits, subscription events, and the limits of the attribution method. If the available period is short, label it an early observation and schedule a later review instead of annualising it.
Keep a calendar of confounders beside the financial table. Content releases, price changes, collaborations, press coverage, account restrictions, staffing changes, holidays, creator illness, and platform incidents may affect the same measure. The presence of a confounder does not make the case useless; it changes the strength of the conclusion. A transparent descriptive case can still teach process, even when it cannot support a causal result claim.
10. Detect selection and survivorship bias
Ask for the denominator behind the case library: how many engagements started, how many had usable baseline data, how many completed the review window, and how many permitted publication? A gallery containing only the strongest finishers hides withdrawals, incomplete records, neutral outcomes, and poor outcomes. That is survivorship bias. An agency does not need to expose private client details, but it should be able to describe the selection rule and why the featured case is or is not representative.
Check who benefits from publication. If the creator received a discount, referral payment, free service, or another incentive, record the connection clearly. The US Federal Trade Commission’s current endorsement guidance says endorsements must be truthful and not misleading, and material connections that could affect how an endorsement is evaluated should be disclosed. Retrieved 29 July 2026, that guidance also warns that an exceptional testimonial can imply a generally expected outcome unless the presentation supplies appropriate context. Australian readers should also verify current ACCC guidance on false or misleading claims.
Do not “fix” selection bias with a vague results-vary disclaimer. Instead, publish the selection method, eligible-case count where permission and evidence allow, measurement definition, costs, and material differences between the featured account and the intended reader. If those fields are unavailable, present the item as a creator story or operational example, not as proof of agency performance.
11. Compare providers without turning cases into a ranking
Use the same evidence standard for every provider. The agency comparison scorecard helps normalize scope and written evidence; the Australian agency comparison adds local contract and privacy questions; and the new-creator agency guide tests whether a service fits an early operating stage. A provider with a dramatic case is not automatically a better fit than one offering narrower, well-documented support.
End with a decision memo, not a winner badge. State which evidence was sufficient, which questions remain open, which risks the creator accepts, and what would cause the decision to be reconsidered. If the creator chooses a pilot, define baseline records, scope, permissions, stop conditions, costs, and the first review date before work begins. That creates the next case record prospectively and is more reliable than reconstructing a success story afterward.
Case-study substantiation and prediction limits
Source register, Retrieved 26 July 2026: the ACCC’s false or misleading claims guidance says business claims should be accurate and provable; the US FTC’s Advertising Substantiation Policy Statement explains prior support for objective claims; and the OAIC’s personal information security guide informs data minimisation and access control. This page contains no published SirenCY result claim and cannot predict performance.