How recommendations are made

Stage A — Hard filters (eligibility)

Hard filters run in a fixed priority order and can eliminate a product outright. They cover: data integrity (we flag stale or conflicting evidence), market eligibility, mandatory accounting integration (for example, ongoing QuickBooks Desktop sync), mandatory workflow features, team capacity and seat structure, hard budget limits, and conflicts with the reasons you are leaving your current software. A product is only eliminated on verified incompatibility — never on missing information.

Three-valued logic

Every capability claim is one of: Supported, Not supported, or Unknown / conflicting. Unknown is never silently treated as yes or no. If a must-have cannot be verified, the product is marked “needs verification” and you get the exact question to ask the vendor.

Stage B — Soft-fit scoring

Products that pass the hard filters are scored 0–100 using fixed weights: accounting integration quality (25%), workflow coverage (25%, where your “important” items count about three times as much as “nice-to-haves” — must-haves are gates, not points), team economics (15%), company-size fit (10%), implementation fit (10%), growth headroom (10%), and secondary signals (5%). When evidence for a category is missing, the category is excluded and the remaining weights are re-normalized — unknowns never add or subtract points.

• Strong fit: 80–100
• Good fit: 65–79
• Conditional fit: workable, with named compromises
• Weak fit / Eliminated / Needs verification: shown with reasons

If the top two products land within a few points of each other, we say it’s a tie and show the genuine differences instead of manufacturing a false winner.

Pricing for your team, not the ad

Cost estimates use the minimum plan that actually satisfies your requirements, priced at your estimated team size, including per-seat charges. Quote-only products are never eliminated for missing pricing — they are flagged so you can request a quote for your exact configuration.

Evidence and freshness

Every decision-critical fact carries a source link, a source type, a verification date, and a confidence level. Facts have review intervals; stale or conflicting facts lower the displayed evidence confidence — which is shown separately from the fit score, because “how well it fits” and “how sure we are” are different questions.

The affiliate firewall

Some vendor links may pay us a commission. Affiliate data lives in a separate database that the recommendation engine never reads. Rankings and eliminations are computed first; links are attached afterward. A product with no affiliate program is recommended exactly as readily as one with a program.