One buyer scenario, one evidence standard, one complete cost boundary
Define representative work and mandatory constraints, require comparable evidence, normalize complete cost, and record uncertainty rather than hiding it in a score.
1. Frame the decision
Name the business outcome, users and roles, data, volume, exceptions, integrations, controls, location, deadline, term, budget boundary, and final approver. A comparison without this scenario is a general product description, not a purchase recommendation.
2. Build mandatory gates
Identify requirements whose failure prevents purchase: a critical workflow, security control, legal term, data location, accessibility requirement, export, integration, availability level, or implementation deadline. Gates are decided before weighted preferences and cannot be averaged away.
3. Use an evidence ladder
PR97 distinguishes a vendor web statement, written support answer, scripted demonstration, buyer-controlled trial, configuration capture, export sample, security report, quote, and contract commitment. Stronger evidence is required as business consequence rises.
4. Normalize commercial inputs
Use the same user population, account types, period, currency, tax treatment, commitment, growth, support, implementation scope, and renewal assumptions. Separate licenses, consumption, services, migration, internal labor, dependent tools, and exit. Unknown is a range with an owner, not zero.
5. Score observed fit
Evaluators score the same scripted scenarios independently before consensus. The record cites what happened, which plan and configuration were used, who observed it, what support was required, and which evidence remains open. Roadmap statements are not scored as delivered capability.
6. Review implementation and operation
Compare configuration, identity, permissions, migration, validation, integrations, testing, change management, training, acceptance, support transition, administration, and recovery. A product can fit a demo and still fail the buyer's implementation capacity.
7. Review contract, renewal, and exit
Verify term, notice, renewal, quantity reduction, uplift, service levels, support, security commitments, data rights, suspension, termination, export, transition, deletion, retention, and remedies. Test a representative export where feasible before commitment.
8. State confidence and limitation
PR97 public comparisons identify their source date and do not imply access to a buyer's tenant or contract. A conclusion is conditional on the stated scenario. The buyer must recheck facts that change by region, plan, quote, or date.
Evidence labels used in a comparison
PR97 distinguishes an observed result from a statement found in official documentation, a vendor answer, and an unresolved assumption. The distinction matters because a feature may exist but require a different edition, configuration, service, integration, country, or contract. Buyers should retain the source, date, product version, role, and scenario beside any conclusion that affects approval.
Why there is no universal score
A weighted score can be useful only after mandatory gates are separated from preferences. Security, privacy, legal, data, critical workflow, accessibility, and budget failures cannot be cancelled out by a large number of optional features. PR97 therefore emphasizes scenario results, complete cost, migration evidence, and explicit decision conditions instead of a single rating.
How uncertainty is handled
Unknown information remains visible as a question, range, assumption, or condition. It is assigned to an owner and deadline. A decision can proceed with a documented residual risk, but the record should state who accepted it and what would cause the choice to be revisited.