Puzzle

Start with the official Puzzle product page and choose one bounded workflow that matches the product's core job: AI-native accounting platform for startups with automated bookkeeping, reconciliations, financial statements, and close workflows. Use representative material rather than a generic demo so you can judge whether the AI understands the terminology, context, and edge cases that matter in your environment.

Connect only the minimum knowledge sources or business systems needed for the test. Where the product can take actions, begin with restricted permissions and keep a human approval step around consequential changes. For vertical products, configure the relevant templates, policies, brand guidance, accounting rules, recruiting criteria, clinical workflow, or other domain context before judging output quality.

Run several examples, including difficult or incomplete inputs. Review accuracy, citations or source grounding where available, consistency, escalation behavior, and the amount of correction required. If the product is an agent, inspect what it did as well as what it said; if it produces documents or records, confirm that the structured output lands in the correct downstream system.

Before wider rollout, test permissions, privacy, audit logs, latency, failure recovery, and cost at realistic volume. The catalog currently classifies the pricing model as subscription, but plans can change, so confirm current commercial terms and any domain-specific compliance requirements directly with the provider.