How we select vendors and service providers

The funnel, not a directory

Afterstack maps an AI-native substitute for software a team already pays for, then a Migration Agent for the cutover. Services partners escalate only when complexity is high. Catalog, search, and categories feed matching. CRM, Support, Sales engagement, Recruiting, Marketing, and Knowledge are the live scored quizzes. Specialists never enter a software rank.

How the Migration Agent works

After you pick a from → to pair (and optional context: company size, stack notes, constraints), an LLM agent writes a detailed migration plan — narrative plus structured phases and executable steps — then coaches the cutover in chat, one step at a time. Catalog attributes ground the model; they are not vendor docs, and the agent must not invent compliance certifications. Affiliates never affect advice. Without an OpenAI or Vercel AI Gateway key, the agent shows a setup state instead of a fake questionnaire. Specialists are ranked from editorial stack tags (product slugs, categories, partner focus). Paid profiles never change the list.

How we select software

A product is listed when a team can actually switch to it: generally available, not a waitlist-only slide, not a feature buried inside someone else’s SKU unless buyers shop it that way (Fin / Intercom Fin is listed because teams shop Fin). We include:

• Incumbents people already pay for, so you can start from what you want to leave.
• AI-native or modern alternatives that are plausible substitutes in that category — public product surface, pricing page, and documented attributes.
• Honest gaps. If it is an agent layer and not a full desk, the briefing says so.

We do not sell a listing, a guaranteed #1, exclusive categories, or a featured module on quiz results. Editorial judgment can be wrong; corrections go through git, not a payout.

How we select service providers

Services firms are migration specialists for complex switches, listed separately from software. We add a firm when it is a generally available consultancy or implementation partner with a public site and a plausible stack (CRM cutovers, support transformations, knowledge/data, and the like). They are never passed into scoreCategory. Paying for a profile cannot put them above Attio, Sierra, or anyone else on a software shortlist.

How categories go live

A category ships when two things are true: operators already pay for an incumbent in that market, and there is a real AI-native crop worth switching to. CRM was the first scored path because lock-in and AI-native challengers collide there. Support, sales engagement, recruiting, marketing, and knowledge followed on the same rule. We do not open empty “coming soon” shelves to look bigger.

When a quiz ships

A catalog can be live without a quiz. Thin lists (ten to twenty names) make a scored shortlist noise. We only ship a quiz after a category clears a curated vendor floor — incumbents people leave, modern mid-market, and an AI-native challenger crop — not coupon junk or duplicate SKUs.

• CRM & sales: 40 vendors. Quiz is live.
• Customer support: 30 vendors. Quiz is live.
• Sales engagement: 25 vendors. Quiz is live.
• Recruiting: 25 vendors. Quiz is live.
• Marketing: 25 vendors. Quiz is live.
• Knowledge: 25 vendors. Quiz is live.

Hitting the floor is a prerequisite, not an automatic launch. Research inputs include public editorial leader lists and known AI-native challengers; we curate into Afterstack JSON. We do not buy a rank, and we do not invent review scores. Field-level sourcing is on the data provenance page.

Monetization firewall

Affiliate programs, Broaden Your Reach, and paid placement never affect:

• Whether a product is included in the catalog or in quiz results.
• The order of a scored shortlist. The published number is 0–100 against a perfect match for those answers.
• Migration Agent complexity or which specialists are suggested (those use editorial stack tags and partner focus, not payouts).

A commercial relationship is a live referral link on /disclosure, which is the same config /go/[slug] uses. A stored partnership flag is not that list. The scorer serializes only fit fields and throws if an affiliate key or migrationDifficulty appears in that payload.

How scored quizzes work

CRM, Support, Sales engagement, Recruiting, Marketing, and Knowledge are scored today. Hard filters run first (free plan, self-host, email or voice when you require them, AI as a dealbreaker, the tool you are replacing). A product owned by an incumbent vendor, or under an announced agreement to be acquired by one, is also filtered out of replace rankings for incumbents in that same category. Neutral category standing does not apply that filter. Survivors are scored as a share of a perfect match on those answers, from 0 to 100, using published 1–10 attributes, matching features, and the replacement bonus. 100 means every rewarded signal is maxed. We show five leads and three also-consider products. Ties break alphabetically. An empty result is a valid result. Specialists never enter a software rank. Buyer-facing version: how Afterstack scoring works.

Company pages

Company records (HQ, attributed ownership, curated execs, press) exist so buyers can see parent risk — including M&A. They never enter scoreCategory. Missing funding is omitted, not guessed. Category standing on the same page is the product’s catalog order, not a rank of the company. Data provenance.

Questions on product pages

Product pages accept a first-party question. Submissions are logged like claims and reviews, emailed when notify mail is configured, and stay unpublished until editorial review. They never change a quiz rank. This is not a third-party discussion network.

Reviews on product pages

First-party reviews submitted on Leave a review are stored unpublished (email never published). On Vercel they persist in private Blob when BLOB_READ_WRITE_TOKEN is set, with content/product-reviews.json as the published git fallback. Operators with REVIEW_MODERATION_TOKEN or ADMIN_SECRET publish or reject from /admin/reviews (query ?token= or Authorization: Bearer). The reviews hub, product pages, and per-product indexes show only approved rows, or an empty state that says No published reviews yet. Reviews never enter scoreCategory. Catalog integration and compliance chips are existing feature flags (SSO, SOC 2, self-host, email/API connections) — Afterstack does not invent GDPR or HIPAA marks.

Product screenshots

Product and company pages render a Screenshots gallery only from vendor-submitted https URLs or Afterstack-hosted files that have a display license. Public submissions on Broaden Your Reach include a catalog product, caption, alt text, and a permission checkbox. They are stored unpublished (email is never written into the media JSON). On Vercel they persist in private Blob when BLOB_READ_WRITE_TOKEN is set, with content/product-media.json as the published git fallback. Operators publish or reject from /admin/media with the same token as review moderation. We do not scrape third-party screenshots or reviews. Media never enters scoreCategory.

What we are not

We are not a lab bench. Attribute scores are editorial judgments from public product surfaces, pricing pages, and hands-on use where we have it. Migration plans are general guidance generated from those attributes — not the vendor’s implementation guide, and not a data mover. List prices go stale. Always confirm on the vendor site.