IDEA FORGE RESEARCH DESK · REALITY CHECK

Derek Johnson

Conf. 51%
Verdict BORDERLINE
RUBRIC FLIP
This idea was also evaluated under the lifestyle rubric.
See both verdicts side-by-side · Security and ops burden don't shrink at lifestyle scale.
Conf.51%BORDERLINE
Evaluated against: Venture-scale — $50k+/mo, fundable
The savings-share model is genuinely clever, but the trust premise it requires — "you can't verify what we did, just send us money" — is exactly the wedge OpenRouter and the model providers themselves are closing in real time.
0%
DO NOT PURSUE
CONCERNS
BORDERLINE
PURSUE W/ CAVEATS
PURSUE W/ CONFIDENCE
BORDERLINE
0%25%50%75%100%

Smart pricing model in a category where the value gap is actively shrinking.

The 20%-of-savings model is the strongest part of this idea. It mirrors Cloudability and Vantage in cloud-cost optimization (real businesses, $100M+ ARR each), and it solves the trust problem at signup ("if no savings, no charge") that makes most B2B AI tools hard to convert. But three structural forces are closing the savings gap before Scruuge can capture it. (1) OpenRouter's own routing has been improving aggressively; "smart router" is now their default. (2) Anthropic and OpenAI have cut prices 60-80% over 2024-2025, with Haiku 4.5 at $1/$5 per Mtok and gpt-4o-mini at $0.15/$0.60. The savings ceiling shrinks every quarter. (3) Cursor, Windsurf, and Claude Code already do behind-the-scenes routing for their users — the indie-dev primary persona may not have a routing problem at all if they're using these tools. The "drop-in OpenAI-compatible proxy" approach has a security objection wall ("you're proxying our keys?") that takes 6+ months to navigate at SMB scale. ZDR helps but doesn't fully resolve it. The path to survival exists — partner-distribution into Cursor or Windsurf, OR vertical specialization on one workload type — but the generic-cost-optimizer position is being squeezed from above (the model providers cutting prices) and below (OpenRouter's own routing).
WHAT WOULD CHANGE THIS VERDICT
  1. 01Land a distribution partnership with Cursor, Windsurf, Replit, or Claude Code where Scruuge powers their default routing — partner pays you, end-users get savings invisibly [+24 → 75]
  2. 02Pivot to vertical-specific optimizer: RAG-only, code-completion-only, or one specific workload where the routing problem has unique structure (e.g., RAG retrieval + answer generation use different optimal models). Smaller TAM but defensible niche [+18 → 69]
  3. 03Win a Y Combinator or a16z RFS slot specifically for AI infrastructure — capital + distribution shortcut compounds the model. Without external accelerant, organic growth in this category is too slow vs. price-cut velocity [+15 → 66]
IF NOT THIS — THREE ADJACENT BETS

Same domain, same research, same vendor pain. Three nearby ideas with their own confidence estimates derived from the analysis above.

RAG-Cost Optimizer

64%

Narrow to retrieval-augmented generation only.

RAG has unique cost structure (embedding model + retrieval + answer generation, each with different optimal models). Vertical depth lets you build heuristics no horizontal proxy can match. Target: AI startups doing customer support, enterprise search, internal knowledge bases. Smaller market but higher conversion.

Validate: 8-week sprint to RAG-specific MVP + 10 design partners: ~$8K

Risk: RAG market itself may consolidate around 2-3 winners (Vectara, Pinecone) who build optimization in-house.

AI Spend Observability (no routing)

58%

Pure analytics: show users where they are overspending, but do not proxy.

Removes the security objection entirely. Sells to engineering managers and CTOs who want visibility before they're ready to delegate routing. SaaS with a clear $99/mo per team price point. Scruuge as Datadog-for-AI-spend.

Validate: 6-week MVP read-only dashboard + 5 paid pilots: ~$10K

Risk: Datadog itself is moving into AI cost monitoring. Window may be 12-18 months.

Cursor/Windsurf Integration Partnership

67%

Sell the routing engine into one IDE.

Skip end-user acquisition entirely. Cursor has 200K+ paid users; Windsurf is growing fast. They have an interest in keeping costs down for their users. Revenue-share or licensing deal removes the security/trust objection.

Validate: Direct outreach + 3-month conversation + technical integration POC: ~$15K and 4 months

Risk: Partner can build internally for less than they pay you; relationship is concentrated single-customer risk.

REFINE THE VERDICT — ROUND 1 OF 2
2 rounds remaining

Add context the analysis missed, change a constraint, or disagree with a specific conclusion. The verdict will re-evaluate, and you will see what moved — and what did not.

0 / 2000

Included in your $29. Two rounds max — use them wisely.

RATE THIS REPORT
SIXTY SECOND TAKE

A clever revenue model (20% of verified savings) sitting in a category where the savings ceiling is shrinking faster than the customer-acquisition velocity. The OpenAI-compatible proxy approach is sound engineering but raises security objections that take 6+ months to overcome at SMB scale. Anthropic and OpenAI cutting prices, OpenRouter improving its routing, and Cursor/Windsurf already doing internal optimization compress the addressable savings. The strongest survival path is either (a) partnering INTO the IDE that owns the indie dev or (b) specializing on one workload type (RAG) where horizontal proxies cannot match vertical depth.

Was this useful?
FIVE COMPETITORS
OpenRouterDIRECT
5% margin built into provider rates; no separate fee
https://openrouter.ai/

Gap: OpenRouter is the rails Scruuge proxies through. Their improving "smart router" closes the gap Scruuge fills. They could add savings-share in 90 days if motivated.

Cloudability / Vantage / Vega CloudTANGENTIAL
$0-2K/mo per workload + savings share
https://www.vantage.sh/

Gap: Cloud-cost-optimization template Scruuge is borrowing from. They will move into AI-cost in 2026; the question is whether Scruuge has a wedge before they do.

Anthropic Direct / OpenAI Direct (price cuts)DIRECT
Variable; Haiku 4.5 at $1/$5 per Mtok
https://www.anthropic.com/pricing

Gap: Not a competitor in the traditional sense — but every quarter they cut prices, the savings Scruuge can capture shrinks. The 60-80% price cuts of 2024-2025 already eroded most of the obvious wedge.

HeliconeADJACENT
Free + $50-500/mo paid tiers
https://www.helicone.ai/

Gap: Observability-without-routing. Solves the security objection but lacks routing optimization. The clean adjacent.

PortkeyAIDIRECT
$0-99/mo + paid tiers
https://portkey.ai/

Gap: AI gateway with routing, observability, fallbacks. Most direct competitor. YC-backed. Aggressive content marketing. Has 2-year head start on routing IP.

Was this useful?
THREE NUMBERS
Anthropic / OpenAI median price cut, last 18 months
~70% (Sonnet 4.6 vs. Sonnet 3.5 launch price)

The savings ceiling is moving faster than customer acquisition. Scruuge needs to capture value during a window that is actively closing.

https://www.anthropic.com/news/claude-3-5-sonnet
Indie developer median monthly LLM spend (developer surveys)
$78/mo

20% of average savings = $7-15/mo per indie developer. Customer acquisition cost must be under $50 LTV-payback. Tight unit economics.

https://www.jetbrains.com/lp/devecosystem-2024/
Cloudability ARR at acquisition by Apptio (2019)
~$50M ARR

Proof point that the savings-share-on-cloud-spend model can build a real business — but Cloudability had 8 years to grow before the cloud category matured.

https://www.crunchbase.com/organization/cloudability
Was this useful?
FIVE HARD QUESTIONS
  1. 01

    When OpenAI cuts prices another 50% in 2026 (probable), the savings you can capture halves. What's the strategy when the gap closes from below?

  2. 02

    How do you handle the security review at SMB customers? "You're proxying our API keys" is a 6-month conversation with each customer's security team. What's the unit-economics model that absorbs that sales cycle?

  3. 03

    Cursor, Windsurf, and Claude Code already optimize routing for their users. If your primary persona uses one of these, what gap are you actually filling?

  4. 04

    What's the moat in 24 months? Routing IP is hard to defend; OpenRouter is iterating; the providers themselves are improving in-network routing. The savings-share pricing is copyable in 90 days. What durable advantage does Scruuge build over its first 12 months?

Was this useful?
ACTION PLAN

You have a clever pricing model in a market where the savings opportunity is shrinking. The next 60 days are about either landing distribution or finding a vertical wedge — generic horizontal routing is not a survivable position alone.

This week
  1. Pick: distribution partner OR vertical wedge

    Either (a) target Cursor, Windsurf, Replit, Claude Code as a partner-of-record, or (b) commit to RAG-only optimization. Both are defensible; horizontal routing is not.

    4 hours strategic decision
  2. Build the savings-attribution proof artifact

    Pick 5 sample real-world workloads (RAG QA, code generation, summarization, customer support classification, agentic loops). Show on the website: "Workload X, baseline $Y, optimized $Z, verified savings W%." Make the trust proof immediate.

    12-16 hours
  3. Cold-test the partnership angle

    Email Cursor, Windsurf, Replit founders directly. Pitch: "We'll route your traffic for free if you let us share the savings with the user." See if anyone bites. Even a "no" reshapes the strategy.

    2 hours outreach + 2 weeks waiting
This month
  1. Sign 5 design partners on whichever path you chose

    For RAG: 5 AI startups doing retrieval-augmented apps. For partnership: 5 IDE companies. Both: free pilots, signed feedback agreements, NDA flexibility.

    $1K + 4 weeks of meetings
  2. Prove the savings on a real workload, publicly

    Pick one design partner. Run their workload through Scruuge for 30 days. Publish: workload type, baseline spend, optimized spend, verified savings %. The case study is the marketing engine.

    $2-5K LLM costs, no engineering
  3. Decide: scale or pivot

    If 3+ design partners say "I would pay 20% of these savings forever" → scale. If < 2 → pivot to observability-only or to a different vertical wedge. Hard kill criterion.

    0 (decision)
Before you spend a dollar
  1. Write down the 90-day kill criterion

    No paying customer at $200+ MRR within 90 days = kill or pivot. The savings-share model only works when customers see real, attributable value within the first month.

    15 min
POSITIONING CHART
APPROACHSPECIALIZATIONObservability only (no proxy)Active routing (proxy required)Vertical / workload-specificHorizontal / genericOpenRouterPortkeyHeliconeVantage / CloudabilityAnthropic / OpenAI directScruuge (as pitched)

The pitched position is in the most crowded zone — horizontal routing where OpenRouter already exists. The defensible moves are up (vertical specialization) or left (observability-without-proxy).

SAMPLE REPORT · v1.3.2 · ILLUSTRATIVE — NOT A LIVE RUN · 7 SOURCES SHOWN
HONESTY-AUDIT PROMPT SHA256:7f867a14952f878d… · SAMPLE REPORT