Pricing Packaging & Package Naming — working memo

Status: Internal strategy working doc backing Foldspace-PLG-Outcome-Pricing.pptx. Not a persona×angle marketing asset. Date: 2026-08-13 Companion: self-serve-activation-plan.md — the activation ladder (R0–R6) this packaging monetizes.

0. Three naming systems are live at once

We currently have three different answers to "what are our packages called," and none of them agree:

Before naming anything, this has to collapse to one system. Shipping a fourth name set on top of three is how pricing pages rot.

1. What a tier name has to do

A package name does exactly three jobs, in this order:

1. Let a buyer self-select in under two seconds. If a developer has to read the feature list to know which column is theirs, the name failed.
2. Predict the next tier. Good ladders are ordinal on one visible axis — the buyer should be able to guess what's above them and why they'd move.
3. Survive the pricing change. The unit price and the free cap will change (the deck says so). Names tied to a number or a feature break; names tied to a state the customer is in don't.

The corollary is the rule most pricing pages break: pick one axis and stay on it. The three axes that work are

• Scale — how much of it you do (Hobby → Pro → Scale → Enterprise)
• Audience — who you are (Developer → Team → Business → Enterprise)
• Stage/job — what you're doing with it (Build → Ship → Run)

Mixing axes is what produces Starter (audience) → Kinetic (nothing) → Scale (scale) → Enterprise (audience): a ladder with no readable direction.

2. Why Build · Operate · Automate · Control doesn't hold

The deck's proposed names are a real improvement on Kinetic — they're at least made of English. But four specific failures:

1. They describe what Foldspace does, not what the customer is. "Operate" is our verb for the meter turning on. To the buyer, all four tiers "operate."
2. "Operate" and "Automate" are synonyms at buying distance. The actual difference — one action vs. a multi-step task — is invisible in the name. A buyer cannot guess which one they need.
3. "Control" as a paid tier is a hostage sentence. "Buy control" says we are holding the customer's control of their own agent until they pay. Governance tiers are named for the assurance (Trust, Enterprise, Governed) or the buyer (Enterprise), never for the withheld thing. The deck's own title slide says "Buy control" out loud — that line should not survive.
4. Four bare verbs are unownable. Every agent vendor's pricing page in 2026 is made of these words. Nothing here is searchable, quotable, or ours.

3. The naming problem is hiding a structural problem

This is the more important finding, and it is worth its own slide.

The four packages are not four rungs of one ladder. Look at what each one actually is:

Presenting these as 01 → 02 → 03 → 04 tells the buyer they replace each other. They don't. A customer on heavy single-action volume with SSO and Trust Lab needs 02 + 04 and never touches 03. The numbering actively misinforms.

Two consequences:

• The contradiction. Slide 5 argues that usage gates convert 1.5–2× better than feature gates — and then tier 03 gates Task Agents, the Data-Query Agent, and voice behind a feature wall. That is the lower-converting design the deck is arguing against, two slides earlier. Task Agents should be a more expensive unit on the meter, not a tier you buy.
• The honest structure is three planes, not four steps:

Same revenue model, one fewer lie in the diagram. It also fixes the naming problem for free, because now each thing only has to be named for what it is.

4. ⚠️ The deck's evidence base does not survive checking

Read this before the naming section. It is the higher-priority finding.

Every statistic on slide 2 and several on slide 3 were verified against primary sources. Most of slide 2 cannot ship.

The core problem

Slide 2 attributes its four headline numbers to Bessemer's "The AI pricing playbook for founders." That document was retrieved and read in full. It is a two-page qualitative playbook. It contains zero adoption percentages, zero NRR figures, and zero 2026 projections. The 21%→15%, 27%→41%, 61%, and +38% figures are not in it.

They trace instead to a cluster of AI-generated SEO content farms that attribute them to Bessemer without linking to it. One of them (SaaS Mag, 2026-04-19) actually cites Chargebee for the 61% and FlexPrice — a billing vendor's marketing blog — for the 38%.

The attribution is laundered. If this ships to a board or an investor and one person clicks through, it is a credibility event, not a footnote correction.

Verdict table

What replaces it — real numbers, same argument

The strategic conclusion does not change. It just needs sources that hold.

• 75% of 230 B2B companies changed pricing or packaging in the past year (Growth Unhinged, 2026-05-13) — pairs with the confirmed 92%.
• Credit-card-required trials convert at 30%, more than 5× those that don't — and only 20% of products require one (ChartMogul × Growth Unhinged × ProductLed, n=200, Jan 2026). This is a far stronger, properly-sourced version of the "gate design matters" point than the n=11 claim it replaces.
• A dual CTA — freemium or a 14-day card-required trial — produced +26% (same study).
• AI gross margins run 50–60% vs 80–90% for classic SaaS (Bessemer, actually in the playbook) — this is the real structural number and it argues the packaging case better than a fake NRR delta.
• 14 days is the trial length for 62% of products.

Slide 3 corrections

Also missing: Zendesk completed its acquisition of Forethought on 2026-03-26, and the band is compressing — HubSpot halved its rate and switched from per-conversation to per-resolution in April; Zendesk added dual verification in May so it bills less. Salesforce at $2.00 win-or-lose is now the conspicuous outlier. That trend is a better slide than a static table.

One competitive finding that changes the packaging question

CopilotKit — the closest analog to Foldspace — deliberately does not price as its customer's COGS. It meters threads and charges $100 per developer seat (capped at 5). That makes it a dev-tools line item, not COGS. Its free tier gates durability (200 threads, 3-day retention) rather than volume — you can build and demo, but not run production.

If Foldspace prices per end-user outcome, we are choosing a materially different model from our nearest competitor, closer to Auth0/Stream. That is defensible, but the deck should argue it as a choice, not assume it.

5. What the market actually names its tiers

65 live pricing pages fetched and read (developer infra, AI/agent platforms, general SaaS). The findings are unusually clean.

The dominant shape

A generic ladder, one non-generic free rung, Enterprise on top.

• Enterprise appears verbatim in 44 of 58 ladders (~76%). Nobody gets creative with the governance tier. Not one company names its top tier after governance, trust, or control.
• All naming creativity lives in the free rung. 13 of 58 avoid the word "Free": Hobby ×5 (Vercel, Clerk, Railway, Render, Cursor), Developer ×5 (LangSmith, CopilotKit, Knock, Sentry, LaunchDarkly), Build ×3 (Stream, Vapi, Anthropic).
• Scale is the one verb-ish word that stuck — 8 occurrences, and in every single case it sits directly below Enterprise. Never at the top.

The two findings that decide this

1. Build as a free tier is real and precedented. Stream (Build = $0), Vapi (Build = PAYG entry), Anthropic (Build = rate-limit tier). Our instinct there was right — keep it.

2. Operate, Automate, and Control appear as tier names on ZERO of the 65 pages. No precedent in developer infra, AI, or general SaaS.

Why our ladder fails, in the practitioners' terms

Every source agrees the tier name's one job is self-selection — "when a user reads the name, they should immediately know if they belong in that bucket" (Mercury). Ours fails four documented anti-patterns:

1. It's non-monotonic. A buyer cannot tell that Automate is more than Operate, or Control more than Automate. Compare Build → Launch → Scale → Enterprise, where the order is self-evident.
2. It names what the product does, not who the buyer is. Every practitioner source says name the buyer's stage.
3. It names the billing mechanic. Free → per-outcome → higher-rate → subscription is a pricing architecture. Only WorkOS and Twilio Flex name tiers this way, and both are the least legible pages in the harvest.
4. Mixed metaphors. Build is a user action, Operate/Automate are product actions, Control is a property — three logics in four names. That's the Pusher failure mode (9 rungs, 4 naming systems).

Plus: Control reads as a de-escalation. After "Automate," it sounds like taking control back — a brake, not a bigger plan. It also collides with the crowded "AI control plane" category (ServiceNow AI Control Tower, Lyzr, TrueFoundry), so it reads as a product category rather than a tier.

On Kinetic

It's a FreshBooks Mighty Oak — FreshBooks shipped Seedling / Evergreen / Mighty Oak, confused customers, and replaced them with Lite / Plus / Premium. The meta-pattern across every documented rename (GitLab, Zapier, Airtable, Render, Slack, Mailchimp, PlanetScale) is companies moving from arbitrary/invented labels toward audience or capability labels. Nobody has ever moved the other direction. Retire Kinetic. Keep Scale.

6. Recommendation — the Four Levels are the price ladder

Superseded: an earlier draft of this section recommended Build · Launch · Scale · Enterprise. Owner rejected it, correctly: those names describe the customer's company stage, not an outcome. For a product priced on outcomes, a package called "Build" tells the buyer nothing about what they get. The generic developer-infra ladder is what companies use when they have no conceptual spine to name tiers after. Foldspace has one.

The source

The Four Levels of Agentic Product Adoption already defines a ranked ladder where each rung is a click-tax reduction — and the mapping onto the live product's capability list is near-exact:

The packages

Answer · Guide · Generate · Learn — each named for the highest level it unlocks. Monotonic, outcome-named, and already published under the company's own byline, so the pricing page teaches the framework and the framework sells the pricing page.

The rate card — the more important half

An outcome is not one price. Fin charges $0.99 whether the agent answered a question or did the work, and it is their most criticised design. The levels give us a defensible spread:

This replaces the earlier Action / Task split, which was a thinner version of the same idea with none of the company's own vocabulary behind it.

The unresolved tension — flag it, do not bury it

Naming packages by level makes the levels the gates. That collides with two things this deck argues elsewhere:

1. Free would stop at Level 1 — shipping a help-docs chatbot as the entry product, which is precisely what the Four Levels article calls dead.
2. It contradicts the activation spine. The aha is the agent acting (Level 3). Gating it behind the paywall is the lower-converting design, and the free-tier slide says so.

Proposed fix: the free tier is Answer plus a capped Generate allowance — a reverse trial into Level 1. Levels price the outcome; they do not withhold it. This is carried on the deck as an open decision rather than silently resolved.

Still true from the research

• Retire Kinetic permanently. Retire the activation plan's Free/Starter/Growth/Business. Three name systems must collapse to one.
• Pure outcome-priced companies (Sierra, Decagon, Ada) publish no tier names at all, so there is no convention being violated here — the territory is unclaimed.
• Worth stealing regardless of naming: split Level 4 into self-serve and sales-assisted so mid-market can buy governance without a call, and unbundle compliance as add-ons (a cleaner home for Trust Lab than a tier, since it is still early access).