Score B · 6.58/10Free to read$100K Hackathon · November 16, 2026 → December 16, 2026Nov 16, 2026

Idle Hours GPU Waste Autopsy

Upload your cloud/GPU billing export and get a dollar-quantified PDF autopsy of exactly where your GPU spend is being wasted, ranked by recoverable dollars per month.

One product, one founder, thirty days. Everything below is built to be executed inside the contest window — the official rules still decide what counts.

Decide in 20 seconds

The facts that decide whether this is worth thirty days of yours.

Promising6.58/10
MVP build (solo)
5 days
Launch price
$299
Sales to $1,000 revenue
4 customers
First paying customer
Direct X replies and DMs to founders publicly posting about a surprising GPU, AWS, Lambda Labs or RunPod bill.
Who pays

Technical founder or ML lead at a 2-15 person AI startup spending roughly $2k-$20k/month on GPU compute, triggered by a monthly invoice that came in higher than expected.

Biggest execution risk

Monetization potential 5.1/10

Why this could work in 30 days: one founder can ship the MVP in 5 days; the plan targets a first sale on day 11; 4 sales at $299 clears $1,000 of collected revenue; strongest signal is speed to first revenue at 8.6/10.

Score and projections come from the generated report and are not guarantees. The $100K Hackathon is run by Ship or Die — read the official rules before you build.

Your first 24 hours

10 concrete actions. Tick them off as you go — nothing is saved or sent.

0/10
Entry price
$299
Solo build
5 days
30d base revenue
$4,992
30d hackathon profit
$4,455
Section 1 of 8

Product opportunity and positioning

Target customer

Technical founder or ML lead at a 2-15 person AI startup spending roughly $2k-$20k/month on GPU compute, triggered by a monthly invoice that came in higher than expected.

Why now

Cloud FinOps dashboards require credentials, setup and interpretation; a single-file upload with a forwardable document is deliverable the same day with zero trust cycle.

Problem

The buyer cannot tell which line items on a GPU bill are waste versus legitimate training cost, so stopped-node disks, 24/7 dev boxes and spot-viable on-demand jobs get paid for month after month.

Solution

A CSV-in, PDF-out forensic pipeline: the buyer uploads an AWS Cost & Usage Report, GCP billing export or RunPod/Lambda invoice; the product normalizes it into a GPU-hour ledger, runs ten waste heuristics and returns a 6-page autopsy plus a remediation CSV with each finding priced and its arithmetic printed.

Differentiation · Dashboards show charts the buyer must interpret and consultants cost five figures over two weeks; Idle Hours returns a finished, arithmetic-transparent document for $299 with no cloud access.

1. Product opportunity and positioning

Idle Hours GPU Waste Autopsy sells a document, not a dashboard. The buyer is a technical founder or ML lead at a 2-15 person AI startup spending $2k-$20k/month on GPU compute across AWS, GCP, Lambda Labs or RunPod. The trigger is concrete: an invoice arrived higher than expected and nobody can explain which lines are waste. Stopped-node disks, 24/7 dev boxes, on-demand pricing on spot-viable jobs and oversized instance families get paid for month after month because reconciling a billing export against real workload assumptions takes half a day nobody has.

The wedge works because it asks for nothing. No cloud credentials, no IAM role, no agent, no Kubernetes operator, no code change, no procurement. One file upload produces a six-page PDF within 24 hours with every finding priced in dollars per month, the arithmetic printed on the page and a confidence label. The output is forwardable to a co-founder or a CFO, which makes it a deliverable rather than a subscription commitment.

Competitive position: generic cloud cost explorers and FinOps dashboards show charts the buyer must interpret; consultants cost five figures and take two weeks. Idle Hours sits between them — dashboard capability, consultant-shaped output, $299, same day.

JEV scored this concept 6.58/10 (band B, confidence 0.68). Strengths: speed to first revenue (8.62 normalized) and reachable customers (7.99) — the buyer is literally posting about the pain on X. Weaknesses: monetization potential (5.14) and market demand certainty (5.80). The plan compensates for both by refusing to build a subscription, by keeping the price at a single $299 decision, and by treating the free one-page Waste Score as the real sales asset. The scope does not broaden in the first 30 days: no monitoring, no auto-remediation, no credentials.

Section 2 of 8

Smallest sellable MVP

A single page with a file input, an email field and a Stripe button, backed by a normalizer, ten heuristics, a dollar model and a PDF renderer — sellable the moment a stranger can upload a billing export and get a priced six-page autopsy within 24 hours without speaking to the founder.

Must-have features

  • Upload endpoint accepting AWS Cost & Usage Report CSV, GCP billing export CSV and RunPod/Lambda invoice CSV or PDF
  • Provider detector plus adapter that normalizes each file into one GPU-hour ledger (instance, region, hours, unit price, tags)
  • Ten waste heuristics implemented as pure functions over the ledger
  • Dollar-quantification model ranking findings by recoverable $/month with hours x unit price x confidence printed on the page
  • Six-page PDF autopsy generator with a per-finding narrative paragraph
  • Remediation CSV punch list ordered by recoverable dollars, pasteable into Jira or Linear
  • Free one-page Waste Score emailed within 60 seconds of upload, top 3 findings and redacted totals only
  • Stripe one-time checkout that unlocks the full autopsy via signed download link

Explicitly out of scope

  • Continuous monitoring dashboard or subscription
  • Cloud API credentials, read-only roles or write-access remediation
  • Agent, sidecar or Kubernetes operator in the customer cluster
  • Team accounts, SSO, audit logs
  • Support SLA or live chat — async email only
  • Enterprise security review or procurement paperwork
  • Historical trend tracking beyond a single upload

Demo flow (what the buyer sees)

  1. Buyer uploads a billing export and enters an email.
  2. Worker detects the provider, stores the file in R2 and writes a job row in D1.
  3. Adapter converts the file into the normalized GPU-hour ledger.
  4. Ten heuristics fire; the dollar model prices and ranks each finding.
  5. Free one-page Waste Score is emailed within 60 seconds.
  6. Buyer clicks Stripe checkout for $299.
  7. Webhook generates the six-page PDF plus remediation CSV and emails a signed R2 link.
  8. Buyer forwards the PDF to a co-founder and opens the CSV as tickets.
Build plan

Technical implementation plan

Stack

Cloudflare Pages + Workers (TypeScript)Cloudflare R2 for uploaded bills and generated PDFsCloudflare D1 for job and order statepdf-lib for the PDF rendererStripe Checkout one-time paymentsResend for transactional emailOne hosted LLM API for per-finding narrative paragraphsPapaparse for CSV ingestion

Architecture

A single Worker handles upload, job creation and delivery. Uploads land in R2 keyed by job id; a D1 row tracks state through uploaded, normalized, scored, paid and delivered. A header-signature detector picks an adapter, each adapter maps raw columns into one GPU-hour ledger schema. Ten pure-function heuristics run over the ledger and emit findings; the dollar model multiplies hours by unit price by confidence and sorts descending. pdf-lib renders the report from a findings object, and the same object serializes to the remediation CSV. Stripe webhook flips the order to paid, triggers rendering, writes the PDF to R2 and emails a time-limited signed URL. The LLM is called once per finding with the already-computed number as input and is instructed to narrate only, never calculate.

Build steps

  1. Deploy Pages + Worker skeleton and a placeholder pricing page.
  2. Create R2 bucket and D1 schema for jobs, orders and findings.
  3. Ship the upload endpoint with email capture and Resend confirmation.
  4. Write the provider detector using header signatures and column fingerprints.
  5. Implement the AWS CUR adapter, then GCP, then the RunPod/Lambda invoice adapter.
  6. Implement the ten heuristics as pure functions with unit tests on synthetic ledgers.
  7. Add the dollar model, confidence labels and descending rank.
  8. Render the six-page PDF and the remediation CSV from one findings object.
  9. Wire Stripe Checkout, the webhook, signed R2 download and delivery email.
  10. Add failure alerting to founder email and a manual-parse fallback path.
  11. Generate three synthetic bills with known waste and tune thresholds against them.

AI components

  • One hosted LLM call per finding to write a two-sentence plain-English explanation of an already-computed dollar figure
  • No LLM is used for arithmetic, ranking or parsing
Hardest technical risk: Billing file format variance — column names, currency handling and PDF invoices change between providers and plans. Mitigate with a header-signature detector, per-format column mapping profiles and a manual-parse fallback email instead of a silently failed job.
Section 3 of 8

Pricing and monetization

Autopsy Report

Entry offer
$299 one-time

For Seed-stage AI team with a GPU bill they cannot explain

  • 6-page dollar-quantified GPU Waste Autopsy PDF
  • Remediation CSV punch list ranked by recoverable $/month
  • Shown arithmetic and confidence label per finding

Autopsy + 30-Day Recheck

$499 one-time

For Teams that want proof the fixes reduced spend

  • Everything in Autopsy Report
  • Second run on next month's billing export
  • Before/after savings statement

Model: Two one-time report SKUs at $299 and $499, with a free one-page Waste Score as the lead magnet; no subscription during the 30-day window.

First revenue offer: Send me your last GPU or cloud billing export and I will send back a free one-page waste score within 24 hours. No cloud access, no install. If it finds more than $300/month, the full 6-page priced autopsy is $299.

Expected ACV: $315

Upsell path: The $499 recheck is the only in-window upsell: offered in every delivery email and pushed to any team billing across two or more providers. Post-window, repeat autopsies per quarter rather than a subscription.

Target customers
35
Revenue model
$10,465
Costs
$420
Profit model
$10,045
Section 4 of 8

First-customer acquisition

Primary channel

Direct X replies and DMs to founders publicly posting about a surprising GPU, AWS, Lambda Labs or RunPod bill.

Where to find them

  • X search for 'GPU bill', 'our AWS bill doubled', 'Lambda invoice', 'RunPod spend'
  • X replies under posts about GPU cost overspend at seed-stage AI startups
  • Indie AI founder Discords and Slacks the founder already belongs to
  • Public build-in-public threads from 2-15 person AI teams
  • Reply threads under GPU provider status and preemption announcements

Exact outreach motions

  • Daily: build a 20-40 name list from X search and log it in a tracker.
  • DM each with the free one-page Waste Score offer and a 24-hour turnaround promise.
  • Deliver the free score, then send the $299 line referencing their own top finding.
  • Follow up on day 3 with the single ROI sentence computed from their free score.
  • Publish one anonymized autopsy publicly with the recoverable-dollar number in the first line.
  • Post the same anonymized autopsy in two founder communities and answer every comment with the upload offer.
  • Ask every buyer for one intro to a founder with a larger GPU bill.

Message template

Saw your post about the GPU bill. I run a forensic pass on billing exports — upload your CSV or invoice, I send back a free 1-page waste score in 24h with the top 3 line items priced in $/month. No cloud access, no install. Want it?
Week-one goal: MVP live by day 5 and 25 DMs sent by day 9, producing at least 5 free-score requests and 1 paid $299 order by day 11.
Section 5 of 8

30-day execution roadmap

MVP deadline: day 5First sale target: day 1130 days planned
Week 1 · Build & validate · 7 days
Day 1Foundation

Ship the shell

  • Create Cloudflare Pages + Worker repo and deploy a placeholder page.
  • Register the domain and point it at Pages.
  • Create Stripe products at $299 and $499.

Deliverable: Live placeholder page with pricing and a working Stripe test checkout.

Day 2Ingestion

File in, job stored

  • Build the upload endpoint writing files to R2 with a D1 job row.
  • Email-gate the upload with a Resend confirmation.

Deliverable: Upload to stored-job to confirmation-email loop works end to end.

Day 3Normalizer

Three provider adapters

  • Write the AWS Cost & Usage Report CSV adapter.
  • Write the GCP billing export adapter.
  • Write the RunPod/Lambda invoice CSV and PDF adapter.

Deliverable: Three sample files normalize into one GPU-hour ledger schema.

Day 4Heuristics

Findings with dollars

  • Implement the ten waste heuristics as pure functions over the ledger.
  • Add dollar quantification and confidence labels.
  • Rank findings by recoverable $/month.

Deliverable: Findings JSON with priced, ranked, arithmetic-bearing entries.

Day 5ReportMVP shipped

MVP freeze

  • Render the six-page autopsy with pdf-lib.
  • Generate the remediation CSV punch list.
  • Render the free one-page Waste Score.

Deliverable: CSV in, PDF plus CSV out. MVP frozen on day 5.

Day 6Payments

Unattended delivery

  • Wire the Stripe webhook to generate and email a signed R2 download link.
  • Add failure alerting to the founder email.

Deliverable: An unattended paid order delivers the full autopsy.

Day 7Dogfood

Threshold tuning

  • Build three synthetic bills with known waste.
  • Tune thresholds so each yields three findings over $100/month.
  • Publish two anonymized sample autopsies.

Deliverable: Two public sample PDFs and tuned heuristics.

Week 2 · Launch & first sale · 7 days
Day 8Outreach prep

Build the list

  • Collect 40 X posters complaining about GPU, AWS, Lambda or RunPod bills.
  • Write three DM variants and open the tracker sheet.

Deliverable: 40-name outreach list with three DM templates.

Day 9First sends

Volume one

  • Send 25 free Waste Score DMs.
  • Log every reply and non-reply in the tracker.

Deliverable: 25 DMs sent, reply tracker live.

Day 10Delivery

Keep the 24h promise

  • Process every incoming file within 24 hours, by hand where needed.
  • Fix the parser breaks the first real files expose.

Deliverable: At least three free Waste Scores delivered.

Day 11CloseFirst sale target

First paid order

  • Send the $299 autopsy offer to every free-score recipient.
  • Follow up with a second DM quoting their own top finding.

Deliverable: First paid order recorded in Stripe.

Day 12Proof

Public artifact

  • Publish one anonymized autopsy on X with the recoverable-dollar number in line one.
  • Pin it and link the free score offer.

Deliverable: Public anonymized autopsy thread live.

Day 13Communities

Borrowed audiences

  • Post the anonymized autopsy in two indie AI Discords or Slacks.
  • Answer every comment with the upload offer.

Deliverable: Two community posts and replies logged.

Day 14Week 2 review

Measure the funnel

  • Compute DM to upload and upload to paid conversion rates.
  • Rewrite the DM around whichever angle converted.

Deliverable: Conversion table and a revised DM template.

Week 3 · Repeat & systematize · 7 days
Day 15Hardening

Reduce manual minutes

  • Add format detection and a clear unsupported-file fallback email.
  • Cap manual parsing at 30 minutes per file.

Deliverable: Parser coverage notes and a working fallback path.

Day 16Recheck tier

Second SKU

  • Ship the $499 Autopsy + 30-Day Recheck checkout.
  • Draft the before/after savings statement template.

Deliverable: $499 tier purchasable and fulfillable.

Day 17Wave two

Fresh list

  • Send 40 DMs to a new X list.
  • Re-open every day-9 non-reply with one ROI sentence.

Deliverable: 40 new DMs and 25 follow-ups sent.

Day 18Content

Worked arithmetic

  • Publish a teardown of five common GPU bill waste patterns with worked arithmetic.

Deliverable: One long-form X post plus a landing page section.

Day 19Referrals

Borrowed trust

  • Ask each buyer for one intro to a founder with a bigger GPU bill.
  • Offer a $50 credit for a converting intro.

Deliverable: Referral asks sent to all buyers.

Day 20Speed

Turnaround

  • Automate the Waste Score email to fire in under 60 seconds.
  • Cut manual turnaround to two hours.

Deliverable: Sub-60s free score, sub-2h paid delivery.

Day 21Week 3 review

Double down or cut

  • Review revenue against the base-case projection.
  • Decide which channel activity to double.

Deliverable: Written go, double or cut decision.

Week 4 · Scale what works · 7 days
Day 22Wave three

Multi-provider teams

  • Send 50 DMs including teams posting across multiple providers.
  • Target founders mentioning migrations between providers.

Deliverable: 50 DMs sent.

Day 23Landing conversion

Sample above the fold

  • Put a sample PDF above the fold with a recoverable-dollar headline.
  • Add a three-step how-it-works with stated turnaround.

Deliverable: Updated landing page.

Day 24Follow-up sweep

Warm non-buyers

  • Email every free-score recipient who did not buy.
  • Include a one-line ROI computed from their own file.

Deliverable: Full follow-up sweep sent.

Day 25Upsell push

$499 conversion

  • Offer the $499 recheck to every team with two or more providers.
  • Add a one-click upgrade link to the delivery email.

Deliverable: Upgrade path live in every delivery email.

Day 26Second proof

Bigger number

  • Publish a second anonymized autopsy with a larger recoverable figure.
  • Cross-post to the day-13 communities.

Deliverable: Second public artifact live.

Day 27Distribution

Owned audiences

  • Pitch three indie AI newsletters a free teardown for their readers.
  • Post in two founder Slacks the founder is already in.

Deliverable: Three pitches sent and two posts live.

Day 28Week 4 review

Reconcile

  • Reconcile Stripe receipts against delivered reports.
  • Issue any refunds inside the window.

Deliverable: Reconciled revenue ledger.

Days 29-30 · Consolidate · 2 days
Day 29Final push

Deadline pressure

  • DM every warm lead with the hard cutoff of 16 Dec 13:00 UTC.
  • Close outstanding free scores into paid orders.

Deliverable: All warm leads contacted with an explicit deadline.

Day 30Wrap

Deliver and freeze

  • Deliver every outstanding report before 13:00 UTC.
  • Snapshot verified revenue, refunds and processing fees for the submission.

Deliverable: Zero open orders and frozen submission numbers.

Section 6 of 8

Revenue and profit scenarios

How the numbers are defined

Hackathon profit — Collected revenue − refunds − chargebacks − payment processing fees − paid marketing spend.

Business profit — Hackathon profit − operating costs (hosting, domains, tooling, AI/API delivery costs).

The contest ranks entrants on hackathon profit only. Hosting, tooling and AI/API delivery costs are not deducted by the organisers, but each sale still has to cover them — so we show both. Definitions verified on 2026-11-15 against the official rules.

Stated assumptions

  • Base case: a cold X DM to a founder already complaining about GPU spend converts 2-5% to a free upload, and 20-30% of free uploads convert to a paid report.
  • Conservative: 60 DMs per week over three waves yields 6 paid reports; Base: 15-16; Upside: 34 driven by two public autopsies and one newsletter pickup.
  • ARPU is blended across the $299 and $499 tiers — Conservative assumes every sale is $299, Base assumes roughly one in nine upgrades, Upside assumes roughly one in three upgrades.
  • Payment processing assumed at Stripe blended 4.5% (2.9% + $0.30 per transaction).
  • Refunds: Conservative 0, Base one $299 refund, Upside one $499 refund. Chargebacks assumed 0 in all cases because arithmetic is printed on every finding.
  • Paid marketing is $0 in Conservative and Base (organic outreach only); Upside includes $250 for one sponsored slot in an indie AI newsletter.
  • Operating costs cover Cloudflare hosting, domain, Resend and LLM tokens for report narration, roughly $2 per delivered report plus fixed tooling.
  • No subscription revenue is modelled; every dollar is a one-time report sale inside the 30-day window.
  • All three scenarios are projections built from stated assumptions, not verified revenue, customers or traction.

Conservative

6 customers
$1,794
collected revenue · avg payment $299
Refunds
$0
Chargebacks
$0
Payment fees
$81
Paid marketing
$0
Hackathon profit
$1,713
Operating costs
$25
Business profit
$1,688

Base case

16 customers
$4,992
collected revenue · avg payment $312
Refunds
$312
Chargebacks
$0
Payment fees
$225
Paid marketing
$0
Hackathon profit
$4,455
Operating costs
$60
Business profit
$4,395

Optimistic

34 customers
$11,390
collected revenue · avg payment $335
Refunds
$499
Chargebacks
$0
Payment fees
$513
Paid marketing
$250
Hackathon profit
$10,128
Operating costs
$120
Business profit
$10,008

Model your own numbers

Runs in your browser · nothing is saved
Collected revenue
$4,784
Payment fees (3.9%)
$187
Paid marketing
$0
Hackathon profit
$4,597
Business profit
$4,537

Hypothetical arithmetic, not a forecast. The fee rate is a flat 3.9% assumption, refunds and chargebacks are left at zero here, and none of this is tracked data.

Scenario models derived from stated assumptions. These are projections, not verified revenue, existing customers or real traction.

Section 7 of 8

Main risks and validation experiments

Risk 1Real billing exports break the parser and consume founder hours that should go to outreach.

Mitigation

Header-signature detection, per-format column mapping profiles, and a manual-parse fallback that still honours the 24-hour promise.

Validation experiment

Hand-parse the files from the first five real uploads and log minutes spent and failure reason per file.

Kill criteria

If fewer than 2 of the first 10 files parse without manual intervention after day 15, collapse to a single 'any CSV or invoice, parsed by hand' mode and stop building adapters.

Risk 2Buyers expect the tool to fix the waste, not just report it, and ask for refunds.

Mitigation

State on the landing page and in the delivery email that the deliverable is a report and a ticket-ready remediation CSV, not remediation.

Validation experiment

Ask each buyer on delivery what they expected, and track refund and chargeback rate across the first 10 sales.

Kill criteria

If refunds plus chargebacks exceed 15% of collected revenue, change the base tier to include a 30-minute walkthrough or drop the offer.

Risk 3Free cloud cost explorers and existing FinOps tools make $299 feel unnecessary.

Mitigation

Lead with a specific recoverable-dollar figure computed from the buyer's own file before asking for money, and never position against a dashboard.

Validation experiment

Pre-sell the paid autopsy to 20 free-score recipients on day 10 before the tier is fully automated, and count written commitments.

Kill criteria

If 0 of 20 free-score recipients convert after 60 delivered free scores, reprice to $99 or stop and re-wedge.

Risk 4One-time pricing produces no compounding revenue, so results track outreach volume only.

Mitigation

Track DM volume against daily sales and push the $499 recheck rather than adding new products.

Validation experiment

Plot daily DMs sent against daily paid orders across waves two and three and measure per-DM conversion.

Kill criteria

If per-DM conversion stays under 0.5% after 150 DMs, the channel is the constraint — switch primary channel or make $499 the default offer.

Risk 5Savings estimates that turn out wrong trigger disputes and damage the public-artifact strategy.

Mitigation

Every number prints its arithmetic and a confidence label; no estimate is ever stated without hours, unit price and confidence.

Validation experiment

Re-check the top three findings against the next invoice for the first three buying teams.

Kill criteria

If two of three re-checks show estimates overstated by more than 50%, widen the confidence penalty and reissue the affected reports.

Section 8 of 8

Hour-by-hour: the first 24 hours

Hours 1-2

  • Create the Cloudflare Pages and Worker project, deploy a placeholder pricing page.
  • Register the domain and point DNS at Pages.

Output: A public URL with the $299 and $499 tiers listed.

Hours 3-4

  • Create both Stripe products and prices.
  • Run a test checkout to confirm the payment link works.

Output: Two working Stripe payment links.

Hours 5-7

  • Build the upload endpoint writing to R2 with a D1 job row.
  • Send the confirmation email through Resend.

Output: A live upload form that stores a file and emails the buyer.

Hours 8-10

  • Write the AWS Cost & Usage Report adapter and the ledger schema.
  • Normalize one synthetic bill end to end.

Output: A printed normalized GPU-hour ledger from one real-format file.

Hours 11-12

  • Build the first 20-name X outreach list of founders posting about GPU bills.
  • Draft the free Waste Score DM and the tracker sheet.

Output: A 20-name list, one DM template and a tracker ready for day 9.

Complete report

Full 30-day launch report

Read the full generated report

Idle Hours GPU Waste Autopsy — 30-Day Launch Plan

Solo founder. Window: 16 Nov 2026 13:00 UTC to 16 Dec 2026 13:00 UTC. Ranking is verified profit, so every conversation must end in a Stripe payment inside the window. Total budget is time plus roughly $50 of tooling.

1. Product opportunity and positioning

Idle Hours sells a document, not a dashboard. The buyer is a technical founder or ML lead at a 2-15 person AI startup spending $2k-$20k/month on GPU compute across AWS, GCP, Lambda Labs or RunPod. The trigger is concrete: an invoice arrived higher than expected and nobody can explain which lines are waste. Stopped-node disks, 24/7 dev boxes, on-demand pricing on spot-viable jobs and oversized instance families get paid for month after month because reconciling a billing export against real workload assumptions takes half a day nobody has.

The wedge works because it asks for nothing. No cloud credentials, no IAM role, no agent, no Kubernetes operator, no code change, no procurement. One file upload produces a six-page PDF within 24 hours with every finding priced in dollars per month, the arithmetic printed on the page and a confidence label. The output is forwardable to a co-founder or a CFO, which makes it a deliverable rather than a subscription commitment.

Generic cost explorers show charts the buyer must interpret; consultants cost five figures over two weeks. Idle Hours sits between them — dashboard capability, consultant-shaped output, $299, same day.

JEV scored this concept 6.58/10 (band B, confidence 0.68). Strengths: speed to first revenue (8.62 normalized) and reachable customers (7.99) — the buyer is posting about the pain on X. Weaknesses: monetization potential (5.14) and market demand certainty (5.80). The plan compensates for both by refusing to build a subscription, by keeping the price a single $299 decision, and by treating the free one-page Waste Score as the real sales asset. Scope does not broaden in 30 days.

2. Smallest sellable MVP

One page: file input, email field, Stripe button. Behind it a provider-agnostic normalizer, ten heuristics, a dollar model and a PDF renderer. It is sellable the moment a stranger can upload a billing export and receive a priced six-page autopsy within 24 hours without speaking to the founder. Ships day 5 with three adapters (AWS CUR CSV, GCP billing export CSV, RunPod/Lambda invoice). If a file will not parse, the founder parses it by hand — that fallback keeps the promise while the parser catches up. Out of scope: dashboards, credentials, agents, team accounts, subscriptions, support SLAs, trend history.

3. Technical implementation plan

Cloudflare Pages plus Workers in TypeScript, R2 for uploads and PDFs, D1 for job and order state, pdf-lib for the renderer, Stripe Checkout one-time, Resend for delivery, one hosted LLM call per finding. Flow: upload lands in R2, a Worker detects the provider by header signature, the adapter writes a normalized GPU-hour ledger (instance, region, hours, unit price, tags), ten pure-function heuristics run over it, the dollar model multiplies hours by unit price by confidence and sorts descending, and pdf-lib renders the report plus the remediation CSV from the same findings object. The Stripe webhook flips the order to paid, renders, stores the PDF and emails a signed URL. The LLM never calculates — it narrates a number the rules engine already produced. Hardest risk is file-format variance; mitigate with a header-signature detector, per-format column profiles and a manual fallback email rather than a failed job.

4. Pricing and monetization

Two one-time SKUs: $299 Autopsy Report and $499 Autopsy + 30-Day Recheck. No subscription inside the window — a recurring plan implies monitoring, dunning and expectations that are explicitly not built. The free one-page Waste Score is the top of funnel and must arrive in under 60 seconds. Processing assumed at Stripe blended 4.5% (2.9% + $0.30). Delivery cost per report stays under $2 in LLM tokens, storage and email. The only in-window upsell is the recheck, pushed to any team billing across two or more providers.

5. First-customer acquisition

One channel: direct X replies and DMs to founders publicly posting about a GPU, AWS, Lambda Labs or RunPod bill. Daily routine: build a 20-40 name list, DM the free Waste Score offer with a 24-hour promise, deliver the score, then send the $299 line quoting their own top finding. Follow up on day 3 with the ROI sentence computed from their file. Publish one anonymized autopsy per week with the recoverable-dollar figure in the first line and cross-post it into founder Discords and Slacks the founder already belongs to. No ads, no SEO, no Product Hunt launch inside the window — none of those compound in 30 days. Week one goal: MVP live day 5, 25 DMs sent by day 9, five free-score requests and one paid order by day 11.

6. 30-day execution roadmap

Week 1 (days 1-7) is build only. Day 1 placeholder page and Stripe products, day 3 three adapters, day 4 ten priced heuristics, day 5 MVP frozen with PDF output, day 6 unattended paid delivery via webhook, day 7 two anonymized sample autopsies published.

Week 2 (days 8-14) opens the funnel. Day 8 list of 40, day 9 twenty-five DMs, day 10 free scores delivered and parser breaks fixed, day 11 first paid order, day 12 first public autopsy, day 13 community posts, day 14 conversion review and DM rewrite.

Week 3 (days 15-21) hardens and widens. Parser fallback, $499 tier live, second DM wave of 40, a five-pattern waste teardown post, referral asks, turnaround under two hours, written go/double/cut decision.

Week 4 (days 22-28) is volume and proof. Fifty DMs, landing page rewritten around a sample PDF, follow-up sweep on every warm non-buyer, second public autopsy, three newsletter pitches, revenue reconciliation and refunds.

Days 29-30 close the window: every warm lead gets the hard 16 Dec 13:00 UTC deadline, all outstanding reports are delivered, and submission numbers are frozen from Stripe.

7. Revenue and profit scenarios

Projections only, derived from the stated assumptions.

ScenarioBuyersARPUGrossRefundsFeesAd spendHackathon profitOperatingBusiness profit
Conservative6$299$1,794$0$81$0$1,713$25$1,688
Base16$312$4,992$312$225$0$4,455$60$4,395
Upside34$335$11,390$499$513$250$10,128$120$10,008

The gap between scenarios is driven almost entirely by outreach volume and by whether a public autopsy gets picked up. Conservative assumes every sale is the $299 tier and 60 DMs per week convert at the low end. Upside requires two public artifacts landing with an audience and one newsletter pickup. Hackathon profit excludes hosting and API delivery cost; business profit deducts it. These are hypotheses, not verified figures.

8. Main risks and validation experiments

Parser breakage: hand-parse the first five real uploads, log minutes per file, and if fewer than 2 of 10 files parse cleanly after day 15 collapse to a manual-first mode. Expectation mismatch: state plainly that the deliverable is a report plus a ticket-ready CSV; if refunds plus chargebacks exceed 15% of revenue, add a walkthrough to the base tier. Willingness to pay: pre-sell the paid autopsy to 20 free-score recipients on day 10; zero conversions after 60 free scores means reprice to $99 or stop. Channel ceiling: plot daily DMs against daily orders across waves two and three; below 0.5% per DM, the channel is the constraint. Estimate accuracy: re-check the top three findings against the next invoice for the first three buyers, and widen the confidence penalty if two of three are overstated by more than half.

9. First 24-hour action plan

Hours 1-2: deploy the Pages and Worker skeleton, register the domain, publish a pricing page. Hours 3-4: create both Stripe products and verify a test checkout. Hours 5-7: ship the upload endpoint into R2 with a D1 job row and a Resend confirmation email. Hours 8-10: write the AWS CUR adapter and the ledger schema, then normalize one synthetic bill end to end. Hours 11-12: build a 20-name X outreach list, draft the free Waste Score DM and open the tracker. By the end of day one the founder has a live URL, a working payment link, a stored upload and a named prospect list.

This brief was generated from public data about established startups, then narrowed and scored for a solo 30-day launch. Revenue and profit figures are scenario models, not verified results, guarantees or evidence of existing customers. Validate demand yourself before you spend a month on it — and check the contest rules for what counts toward your score.

Related ideas

All ideas →