# Launch Checklist for a Tiny A...

Source: https://notesgarden.com/publish/Yob1VSyGDbPd

## 🧩 Offer definition

- [ ] Define one repeatable workflow your product automates or speeds up
- [ ] Choose a buyer who already pays in time or money for this workflow today
- [ ] Describe the before state in operational terms, not AI hype language
- [ ] Describe the after state as a measurable output such as fewer minutes, fewer clicks, or higher throughput
- [ ] Write a promise that survives model drift and prompt changes

Promise the business outcome or job-to-be-done, not that a specific model will always behave the same way.

- [ ] Decide what the user must bring such as source documents, CRM data, or internal knowledge base pages
- [ ] List what your tool will not do so prospects do not assume a general AI assistant
- [ ] Choose whether you are selling to a solo operator, a small team, or a manager with direct reports
- [ ] Pick one activation event that proves the user got value
- [ ] Choose one clear reason a team would keep paying next month

## **⚙️** Model and workflow reliability

- [ ] Test the full workflow on messy real inputs rather than polished demo data only
- [ ] Build a fallback for empty, partial, or low-confidence outputs
- [ ] Log prompt version, model version, and workflow version for every production run
- [ ] Define which steps must be deterministic and which can stay generative
- [ ] Keep system prompts and output schemas in version control
- [ ] Add validation for structured outputs before they reach a user-facing screen or downstream action
- [ ] Set timeouts and retries that fail gracefully instead of freezing the workflow
- [ ] Decide what happens when the model is wrong but confident

The product needs an operational answer here, such as human review, source citations, or confidence gating.

- [ ] Measure average latency for the end-to-end task, not just the raw model response
- [ ] Run a regression set whenever you change prompts, tools, chunking, or model choice

## 🚀 Onboarding and first-run experience

- [ ] Make the first-run path short enough that a founder can complete it without booking a demo
- [ ] Ask only for inputs that are essential to reaching the first value moment
- [ ] Provide sample data if the real integration takes time to prepare
- [ ] Show the expected output format before the user runs the workflow
- [ ] Explain what good input looks like for the workflow
- [ ] Include one visible note about what the AI may still need human review for
- [ ] Save partial onboarding progress if setup includes integrations or document uploads
- [ ] Instrument every drop-off point in onboarding

Early churn often looks like pricing churn when it is really setup friction.

- [ ] Write the empty-state copy so a new user knows exactly what to do next
- [ ] Give the user a shareable output, export, or internal handoff artifact from the first successful run

## 💰 Pricing and packaging

- [ ] Pick a pricing unit that matches how the buyer thinks about value
- [ ] Decide whether your margin is safer with seat-based, usage-based, or workflow-batch pricing
- [ ] Estimate gross margin using real model costs, retries, storage, and support time
- [ ] Create a plan boundary that naturally nudges expansion without hiding core value
- [ ] Add a fair use threshold if a small number of heavy users could destroy margins
- [ ] Write pricing copy that explains outcomes instead of token counts where possible
- [ ] Choose whether free trial, free credits, or a paid pilot best matches your sales motion
- [ ] Prepare one exception policy for early design partners

Without a defined rule, every discount request turns into a custom negotiation that eats founder time.

- [ ] Decide what happens when a user hits limits in the middle of a workflow
- [ ] Set a manual review alert for accounts whose usage pattern looks economically unhealthy

## 🔒 Trust, privacy, and buyer objections

- [ ] Know whether customers will upload sensitive data, customer records, or internal documents
- [ ] Write a plain-language explanation of where data goes during processing
- [ ] State retention windows for inputs, outputs, logs, and deleted content
- [ ] Decide who inside a customer account can see generated outputs by default
- [ ] Offer one honest answer to the question of model accuracy and hallucinations
- [ ] Prepare a short security questionnaire response for prospects who ask basic vendor questions
- [ ] Explain how a customer can export and remove their data
- [ ] Document any human-in-the-loop steps that protect quality or compliance

This often reduces fear more effectively than pretending the workflow is magically autonomous.

- [ ] Make sure terms, privacy copy, and marketing promises do not contradict each other
- [ ] Decide how you will communicate incidents, degradations, or model outages to paying teams

## 📣 Go-to-marke and first customers

- [ ] Write one outbound message that describes the painful workflow without sounding like mass AI spam
- [ ] Create a landing page that shows the input, transformation, and final output clearly
- [ ] Include a screenshot or short video of the real workflow instead of a floating abstract AI graphic
- [ ] Prepare two founder-led demo scripts for two different buyer personas if needed
- [ ] Track which prospect objection appears before anyone even starts a trial
- [ ] Have a manual concierge version ready if the product still needs setup help for early accounts
- [ ] Create a simple case-study template for your first successful team
- [ ] Collect the phrases customers use to describe the workflow pain in their own words
- [ ] Decide which communities or channels are credible for this niche instead of posting everywhere
- [ ] Make sure the first 25 customers all have a clear path to sending a referral or internal share