Skip to content

Launch the Solo economic debugger

Bottom line

Sell one self-serve result: backtest an AI cost cut before production and get an evidence-backed ship, hold, or collect-more-evidence decision. The SDK and self-hosted UI remain open source. Zeroth Cloud charges $39/month for hosted history, recurring comparisons, quotas, and post-change evidence after a 14-day trial.

Do not lead with observability, generic workflow management, FinOps, provider routing, or governance. Those are mechanisms or expansion paths. The first buyer must understand the transaction without a call:

5–25 labeled cases + incumbent + candidate
  → bounded economic backtest
  → cost and outcome comparison
  → approve candidate | keep incumbent | collect evidence

Initial user

Target a solo developer or AI engineer who:

  • owns a production or near-production AI feature;
  • can label success for a small representative case set;
  • is considering a cheaper model, prompt, or implementation change; and
  • cares about avoiding either a quality regression or an unjustified AI bill.

This is narrower than the eventual enterprise buyer. It is intentionally self-serve, has a short path to first value, and can validate recurring behavior before Zeroth builds sales, Team collaboration, or enterprise procurement.

Offer and activation

  • Trial: 14 days, one hosted backtest, 100 provider calls.
  • Solo: $39/month, three backtests, 300 provider calls, retained history, and five daily schedules.
  • First value: one retained decision from at least five labeled cases.
  • Recurring value: a second backtest, scheduled decision, or post-change verification during the first paid period.
  • Support: asynchronous documentation, email, and public issue triage. Do not add a calendar link or require onboarding calls.

Paddle must show the same amount, cadence, trial, payment-method requirement, renewal, and cancellation terms. Do not discount, add annual billing, or add a second SKU until the initial funnel produces interpretable evidence.

One-channel launch

Use the public repository and installable open-source debugger as the try-before-signup artifact. Its only hosted call to action is the Solo trial. Run one acquisition channel at a time for 14 days so source attribution does not require cross-site tracking:

  1. Publish the exact SDK and hosted candidate from public main.
  2. Confirm that the README and package metadata link to the public HTTPS origin.
  3. Choose one earned developer channel where the owner can participate legitimately. Show HN is acceptable only with an eligible established account and a runnable public artifact; otherwise use one relevant developer community where self-promotion is permitted.
  4. Publish one technical demonstration using synthetic or explicitly authorized data. Show the incumbent, candidate, outcome constraint, cost change, verdict, and abstention behavior.
  5. End with one action: Run your first economic backtest.
  6. Answer asynchronously. Do not ask for calls, invoice files, prompts, credentials, or customer payloads.

Suggested launch sentence:

Zeroth tells you whether a cheaper model or workflow change should ship. Give it 5–25 labeled cases; it compares cost and accepted outcomes, refuses false certainty, and retains the decision for the next production change.

Do not claim proven savings from synthetic data, imply causal waste from a failed run, or advertise enterprise governance that is not purchasable.

Measure without surveillance

Run the aggregate report against the hosted economic database:

uv run python release/cloud_funnel_report.py \
  --window-days 30 \
  --output .evidence/cloud-funnel.json

It emits no tenant, user, email, WorkOS, Paddle, or payload identifiers. The signup cohort is measured through server-owned organization bindings; first and repeat value through retained backtests; checkout completion through an external Paddle customer projection; and paid, canceled, or past-due state through signed billing events.

The report intentionally does not count page views or package downloads. Anonymous traffic is not activation, and one-channel windows provide enough source attribution for the first experiment without cookies or an analytics vendor.

Predetermined decisions

These are operating thresholds, not market benchmarks:

  • At least 10 signups but no first-value backtest: fix SDK activation and examples; do not change price or add features.
  • At least 10 first-value users but no checkout-completed subscription: test the offer and price presentation; do not build Team.
  • Checkout-completed trials but no paid activation after their 14-day windows: the recurring value is unproven; inspect whether users run a second backtest or schedule before adding acquisition spend.
  • One paid user who reaches recurring value and remains active through the first renewal: continue Solo and test a second acquisition channel.
  • Paid users repeatedly request shared policy, approval, or audit ownership: then evaluate Team governance. A generic request for "more seats" is not enough to bypass enforced collaboration limits.

Do not buy ads before signup-to-first-value works in an earned channel. Do not interpret a tiny conversion percentage as stable; retain the raw aggregate counts and time-to-first-value alongside every rate.

Launch sequence

  1. Pass the vendor-readiness report in sandbox.
  2. Deploy the candidate and pass the read-only cloud preflight.
  3. Record signup → first backtest → checkout → signed trial/active webhook → portal → cancellation → HTTP 402 in sandbox.
  4. Publish owner/counsel-approved policy and support pages.
  5. Repeat vendor readiness, preflight, and the commercial journey in production.
  6. Open checkout, run one 14-day channel, and review the aggregate funnel.

The hosted-commerce runbook owns configuration and acceptance details. The cloud evidence record owns the go/no-go evidence. Neither a green local test nor a vendor dashboard alone authorizes production payment.

Adversarial review

The strongest objection is that a $39 backtest allowance may be too occasional to retain users. Scheduled decisions and post-change verification are the retention hypothesis, not established demand. The first renewal and repeat-use events must decide that question.

The simpler alternative is a one-time paid report, but it conflicts with the selected subscription model and would teach little about recurring workflow economics. The safer subscription path is therefore a low fixed price, a real trial, strict cost limits, and no Team investment until recurring Solo behavior is observed.