How to Design an Intent Program Case Study That Separates Evidence From Claims

Who is this for?

This guide is for marketing and RevOps leaders, analysts, agency owners, customer marketers, executives, and legal or compliance reviewers who need a credible intent data case study. It covers evidence design for one intent program. It does not cover generic customer-story copy, an agency-wide case-study strategy, QBR design, an executive dashboard, or a standalone ROI proof hub.

Intent and identity signals are probabilistic. They can document observed research, engagement, or a probable account or person match under a stated method. They do not prove consent, individual identity, buying intent, causation, or a future result. A useful case study makes those boundaries visible.

Start with the decision the case study should inform

Choose one primary decision: expand a use case, revise the operating model, continue a pilot, approve a larger evaluation, or stop. A case study designed to answer “Should we expand account prioritization?” needs different evidence from one designed to answer “Can our agency reliably produce branded topic reports?”

Write the protocol before reviewing the most favorable accounts. Include:

• decision owner and intended audience;
• unit of analysis: person, account, buying group, campaign, opportunity, or client;
• eligible population and exclusions;
• intervention and operating workflow;
• business-as-usual baseline or counterfactual;
• primary and secondary outcomes;
• observation and outcome windows;
• assignment or comparison method;
• calculation and currency rules;
• minimum sample and materiality threshold;
• privacy, customer, legal, and publication approvals;
• stop conditions and known limitations.

Pre-specification reduces the temptation to select a window, segment, or metric after seeing the result. If the design changed, keep the original and amended versions and explain why.

Separate baseline, intervention, exposure, and outcome

The baseline describes what happened before or without the program: existing account selection, response, opportunity, and cost patterns. The intervention is the complete change – not merely “we added intent data.” It might include new signals, identity resolution, routing, account research, ads, outbound drafts, sales training, and reporting.

Exposure records which eligible units actually received the intervention. An account that entered an audience but received no impression is not exposed to the same degree as an account that received ads and a seller action. An account that generated a signal but was never accepted by sales is different again.

Outcome is the predefined later event: accepted account, qualified meeting, accepted opportunity, stage progression, win, gross profit, or a process result. Keep outcome records distinct from provider scores and influenced-pipeline labels.

The counterfactual asks what would have happened without the intervention. A randomized holdout is strongest when feasible. Other options include a phased rollout, comparable region, cutoff design, or a carefully matched cohort. A before-and-after comparison can be descriptive but is vulnerable to seasonality, staff changes, media shifts, and pipeline mix.

The Magenta Book’s evaluation guidance explains that impact evaluation needs a credible comparison to estimate what occurred because of an intervention, while process evidence helps explain how and why implementation worked. That distinction should anchor an intent data case study design.

Build an evidence table before drafting prose

Create one row per material statement with these fields:

1. Claim ID and proposed wording.
2. Claim type: fact, calculation, observation, attribution, causal finding, testimonial, projection, or opinion.
3. Unit, population, and denominator.
4. Source system and source record.
5. Data owner and extraction logic.
6. Method and comparison.
7. Time and geography scope.
8. Evidence grade and uncertainty.
9. Known limitations and alternative explanations.
10. Customer approval and material-connection disclosure.
11. Privacy, legal, compliance, and editorial review status.
12. Approved wording and channels.

This is the core intent data case study template. The published article may be short; the underlying evidence table should be detailed enough for another qualified reviewer to reproduce the calculation.

Use a transparent evidence ladder:

• Causal finding: supported by a credible experimental or quasi-experimental design and appropriate analysis.
• Comparative observation: treatment or exposed units differed from a stated comparison, with material limitations.
• Descriptive observation: a measured fact about the program without causal attribution.
• Calculation: reproducible arithmetic from named, quality-checked inputs.
• Testimonial or opinion: a person’s honest experience, not independent proof of performance.
• Projection: a future scenario based on labeled assumptions, not a result.

Never promote a descriptive observation into a causal headline. “Opportunity creation was higher among activated accounts” is not the same as “intent activation increased opportunities.” The latter requires a design capable of supporting the causal interpretation.

Use an evidence-grade framework consistently

Give each source a grade based on relevance and method, not on whether it favors the program:

• Grade A: prospective comparison with strong assignment, exposure, data quality, and analysis.
• Grade B: relevant quasi-experimental comparison or strong internal observational analysis with bounded assumptions.
• Grade C: quality-checked descriptive program data, reproducible calculations, or a closely comparable primary case.
• Grade D: vendor claim, customer recollection, generic benchmark, anecdote, or unverified estimate.

A Grade D source can still appear when clearly labeled – for example, a customer explaining why the project mattered. It should not substantiate an objective performance claim. The case study’s conclusion should reflect the weakest critical link, not the average grade across all rows.

The FTC’s advertising substantiation policy says advertisers and agencies need a reasonable basis for objective claims before disseminating them. The exact legal standard depends on context and jurisdiction, so legal counsel should review the final use. Operationally, this means the evidence packet must exist before the headline goes live.

Capture the full operating method

A credible case study describes the workflow with enough detail to judge transferability:

1. How intent topics and thresholds were chosen.
2. Which first-, second-, or third-party signals were used.
3. How accounts or people were resolved and confidence recorded.
4. Which fit, freshness, suppression, and consent or lawful-use rules applied.
5. How records were routed and who accepted or rejected them.
6. Which ads, research, content, or outreach actions were performed.
7. How quickly actions occurred after the signal.
8. Which concurrent programs could affect the outcome.
9. How control or comparison units were protected.
10. How results, corrections, complaints, and missing data were logged.

Include implementation failures. If only half the eligible accounts received the action, that is part of the result. Report both intent-to-treat outcomes based on assignment and an action-taken view when useful, clearly noting that action-taken analysis can be selected by rep behavior.

Choose analytics tools and templates for reproducibility

The right case study design for intent programs does not require a particular software brand. It requires controlled inputs and repeatable calculations:

• a warehouse or exported source files with immutable snapshots;
• a data dictionary defining signals, identities, exposures, outcomes, and exclusions;
• versioned SQL or formulas for every reported number;
• an experiment or assignment register;
• an exposure and contamination log;
• a calculation workbook with denominators and currency rules;
• a claim register linked to screenshots or source records;
• an approval log for the customer, brand, legal, privacy, and editorial teams;
• a change log showing corrections after review.

For a small program, a locked spreadsheet plus exported CRM and platform files can be sufficient. For a larger program, use version control, warehouse models, access controls, and reproducible notebooks. Tool sophistication does not compensate for changing definitions or cherry-picked accounts.

Compare experimental, causal, attribution, and observational approaches

Randomized assignment is preferred when eligible accounts can be assigned to threshold-guided activation and business as usual. Randomization helps balance both observed and unobserved differences, but contamination, noncompliance, and small samples can still weaken the conclusion.

A phased rollout or difference-in-differences design can estimate change when a comparable untreated group and parallel pre-trends exist. A cutoff may support regression discontinuity when a pre-existing score determines treatment and units near the cutoff are comparable. Matched cohorts can reduce visible differences but cannot eliminate unobserved selection.

Multi-touch attribution and platform conversion reports are descriptive allocation systems. They are valuable for understanding recorded journeys and operations, but they do not automatically estimate the counterfactual. Likewise, “influenced pipeline” can show that opportunities overlapped with an audience or touch; it is not necessarily incremental pipeline.

Use the strongest method feasible and state what it can answer. The Magenta Book analytical methods outlines experimental, quasi-experimental, and theory-based methods and their limits. A commercial case study can adopt the decision logic without copying public-sector complexity.

Budget for evidence, not only storytelling

Case study cost includes protocol design, data extraction, identity and outcome reconciliation, analyst time, quality assurance, customer interviews, copy and design, privacy and legal review, customer approval, redaction, hosting, and future maintenance. Add the opportunity cost of preserving a holdout or delaying the rollout for measurement.

If evidence is likely to be reused in sales, proposals, ads, partner content, and LLM-facing pages, invest in a claim register that records approved wording by channel. If the customer requires anonymity, budget for aggregation, masked screenshots, and re-identification risk review.

The most efficient path is often to build evidence collection into the program from the start. Reconstructing exposure, definitions, and permissions after a success has been noticed is expensive and prone to bias.

Define metrics and confidence rules

Report the complete funnel: eligible units, signaled units, resolved units, fit-qualified units, accepted units, activated units, exposed units, meetings, accepted opportunities, wins, gross profit, cost, corrections, suppressions, and complaints. Always show denominators.

Predefine deduplication, account and opportunity windows, multi-client rules, currency conversion, gross-margin assumptions, and treatment of missing data. Segment results only when the sample supports the cut; otherwise label segment observations as exploratory.

For comparative results, report absolute and relative differences, uncertainty intervals, sample sizes, and practical significance. A large percentage lift from a tiny baseline can be commercially trivial and statistically unstable. Do not hide zero or negative outcomes in an appendix.

Set publication rules before analysis:

• publish a causal claim only when the design and review support it;
• publish a comparative observation with the comparison and limitation close by;
• publish a descriptive metric with its population and denominator;
• publish a projection only as a scenario with its assumptions;
• remove or narrow any claim that cannot be reproduced;
• stop publication for unresolved permission, privacy, security, or material accuracy issues.

Know when the available evidence is insufficient

A case study is decision-useful when definitions are stable, source records are accessible, exposure is known, outcomes are recorded, the comparison is credible, and the operating context is described. It is insufficient for strong performance claims when the sample is tiny, only successful accounts were selected, the baseline changed, concurrent campaigns are uncontrolled, sales manually chose treatment, or the outcome window excludes delayed failures.

Do not solve insufficient evidence with confident copy. Publish a process case study instead: the problem, workflow, feasibility metrics, lessons, and next test. Saying “the pilot established that the team could resolve, review, and route accounts within the approved SLA” can be useful even when revenue impact remains unknown.

Combine intent, identity, activation, and outcome evidence carefully

Maintain separate fields for each evidence class. Intent describes the observed topic or behavioral signal and its source. Fit describes whether the account meets the chosen commercial criteria. Identity describes the probable match and confidence. Freshness describes actionability. Activation records the actual action. Outcome records what occurred later.

Never use identity confidence as intent intensity or treat an activation event as a positive outcome. Preserve corrections and unmatched records. Explain whether the signal is account-level or person-level and avoid language that attributes account research to a named individual without adequate evidence and permitted use.

NIST’s Privacy Framework provides a practical structure for identifying and managing risks from data processing. Apply purpose limitation, minimization, access controls, retention, deletion, suppression, and incident response to both the operating program and the published evidence packet.

Control testimonials, disclosures, and typicality

A customer quote must be honest, approved, and presented in context. Record who wrote or edited it, whether the customer received compensation or another benefit, and which objective claims it implies. A testimonial is not substantiation for a claim requiring objective evidence.

The FTC’s endorsement guidance explains that material connections should be disclosed and that exceptional results can imply typical performance. A vague “results may vary” statement does not automatically cure the net impression. Put relevant limitations and generally expected performance near the claim where applicable, and obtain legal review for the intended jurisdictions and channels.

Avoid composite customers unless plainly disclosed. Do not fabricate quotes, roles, screenshots, events, or timelines. If confidentiality requires alteration, disclose the type of alteration and ensure it does not change the conclusion.

Where BrandWell fits

BrandWell’s separate agency-reseller product is designed to help agencies turn intent and identity inputs into branded topic reports, a white-label sales-and-delivery engine, agency-controlled billing, and agent-ready workflow instructions. It is distinct from the legacy BrandWell SEO writer. The operating artifacts can support a case-study evidence packet: signal definitions, branded reports, routing instructions, activation records, and a claim register – provided the agency preserves the comparison and outcome data.

BrandWell agency plans range from $2,500 to $5,000 per month, depending on topic count, term, and available contractually scoped topic exclusivity. The current written quote and Order Form control. Any topic exclusivity is conditional on availability, scope, purchase, and written terms. Treat it as a preliminary planning range, not a public list price, and require a current written quote. Where available under written terms, topic protection is conditional on scope and availability rather than universal.

before use, require product, pricing, privacy, security, compliance, legal, and platform-policy review.

A $70 seven-day reseller pilot can generate branded topic reports and test data usability, review effort, white-label delivery, and client approval workflow. Confirm the current written pilot terms and operational readiness before making client-facing promises. A short pilot can become a process case study; it usually cannot support a broad revenue or ROI claim.

Make the evidence workflow agent-ready

Claude or ChatGPT can prepare the protocol, data dictionary, calculation checks, claim register, limitation language, and review packet. Moxby can optionally perform approved browser steps for evidence capture and workflow updates. Agents should receive only approved data and minimum necessary fields.

Use an instruction such as: “For each proposed statement, classify the claim, locate the supporting source record, reproduce the calculation, name the denominator and comparison, assign an evidence grade, list plausible alternative explanations, and recommend the narrowest accurate wording. Flag missing permissions and material connections. Do not invent sources, alter customer quotes, publish content, contact the customer, or approve legal claims.”

Require a named human to approve data access, identity use, customer outreach, quote edits, screenshots, disclosures, claims, and publication. Preserve the agent’s source citations and edits so reviewers can audit the path from record to copy.

Final case study checklist

• Freeze the decision, protocol, population, intervention, and outcomes.
• Preserve business as usual and a credible comparison where possible.
• Record assignment, exposure, noncompliance, and contamination.
• Separate intent, fit, identity, activation, and outcome evidence.
• Link every objective claim to a reproducible source and calculation.
• Show denominators, uncertainty, segments, and material limitations.
• Distinguish causal findings, comparative observations, descriptions, testimonials, and projections.
• Record customer permission, material connections, and channel-specific approvals.
• Review privacy, security, legal, compliance, and platform policy.
• Use human approval before any public claim or publication.

The strongest intent program case study is not the one with the largest number. It is the one in which a skeptical reader can see what was done, what changed, how the comparison worked, what remains uncertain, and exactly which claims the evidence can bear.