Credit Consumption Analysis

February 24, 2026 · 21 external organizations · Production data

Executive Summary

$10M
Target ARR
$833K/month · 24-month horizon
27.8M
Target Credits / Month
~710 orgs × $1.2K avg ARPU
40 : 60
Target AI : Query Ratio
Where the flywheel spins
$31K
Current ARR
$2,550/month · 13 active orgs
85K
Current Credits / Month
80% from 3 pipeline orgs
$196
Current Avg ARPU
Target: $1,211
29%
Current AI %
Target: 40%
Programmatic usage drives credit consumption. Credit consumption drives revenue. 86% of external credits come from programmatic access (API, CLI, automated pipelines) — not from users clicking in the UI. A user accessing the platform programmatically generates 3–60x more credits than a manual UI user. The product implication is clear: invest in features that enable and expand programmatic access — better APIs, CLI tools, MCP integrations, scheduled queries, agent-driven workflows — because that is what directly multiplies revenue per user and per org.

1. What Drives Credit Usage

Thesis: programmatic usage → credit consumption → revenue. Credits are consumed in two categories: AI tokens (chat completions, agent reasoning) and query execution (SQL seconds). Across 21 external orgs, 86% of all credits come from programmatic access (API, CLI, automated pipelines) — not from users clicking around in the UI. If we want to grow revenue, we need to invest in product that drives programmatic usage.

Programmatic vs. Manual: How We Classified

SignalClassificationWhat It Means
session_id = NULLProgrammatic — Direct API Query or AI call made outside any chat session. Pure API/SDK usage.
meta.source = “cli” | “api”Programmatic — CLI/MCP Chat session initiated from CLI, Cursor, or MCP tool. Has a session, but not human-in-the-UI.
meta.source = “web” | “agents-page”Manual — Web UI User in the browser, chatting with agents or running queries manually.
Burst queries (<2s gaps)Programmatic — Pipeline Rapid-fire queries at machine speed. Marinade: 90% burst rate. CMC: 98%.
Validation: Jim Myers described himself as “largely programmatic.” The initial session-presence heuristic classified him as 50/50 — wrong, because CLI-initiated sessions have a session_id. After incorporating meta.source, his classification moved to 97.7% programmatic (233 of 298 sessions are CLI-originated, plus all sessionless query events). This confirmed the heuristic works.

Platform-Wide Findings

86%
External Credits Are Programmatic
API calls, CLI, automated pipelines
98%
Query Credits Are Programmatic
Nearly all SQL execution is API-driven
91%
AI Credits Are Session-Based
Most AI usage happens inside chat sessions

Three Consumption Patterns

PatternOrgs% of CreditsProgrammatic %What They Do
API Pipeline380%~100% Automated SQL pipelines via API. Machine-speed burst queries. Zero AI. High volume, low stickiness.
AI Chat613%~15% Exploring data through agents in the UI. Human-paced. High AI %, low query volume. Undermonetized.
Mixed311%~60% Both AI exploration and query execution. Closest to the target pattern. Highest stickiness potential.
The thesis confirmed: programmatic access drives dramatically higher credit consumption. The top 3 pipeline orgs burn more credits in days than AI chat explorers burn in weeks. But volume alone isn’t health — pipeline orgs have zero AI usage, making them commodity consumers (any SQL warehouse will do). The highest-value pattern is mixed: both AI and query, both programmatic and manual.

The Revenue Multiplier by Access Pattern

Credit consumption follows a clear hierarchy. Each step up the programmatic ladder multiplies revenue per user:

Access PatternAvg Credits/DayRev MultiplierWhy
Manual UI only50–1001x Browsing, occasional chat. Human-speed ceiling caps consumption.
Manual chat + API queries200–3503–5x Chat for exploration, then direct API for extraction. Eric Stone’s pattern.
Agent-driven (CLI/Cursor/MCP)300–5005–10x One agent session triggers 3–10 downstream queries that iterate and self-correct.
Automated pipeline (API)1,800–6,00020–60x Machines run 24/7 at burst speed. No human bottleneck.
This is the investment thesis for the product roadmap. Every user who moves from “manual UI” to “agent-driven” generates 5–10x more credits. Every org that adds automated pipelines generates 20–60x. Features that enable programmatic access — better APIs, CLI tools, MCP integrations, scheduled queries, agent-driven workflows — directly multiply revenue per user. Features that only improve the manual UI experience have a 1x ceiling.

External Org Consumption (21 orgs)

Organization Credits AI % Qry % Cred/Day Est Rev/Mo Pattern
Marinade Finance64,025 3%97% 1,829$1,646 Pipeline
CMC Capital SRL20,817 0%100% 5,204$4,684 Pipeline
Stockholm SSE18,066 0%100% 6,022$5,420 Pipeline
CoW Swap6,922 36%64% 865$779 Mixed
Laminated Labs6,24291%9%223$201AI Chat
Near Foundation5,87931%69%210$189Mixed
Avalanche3,59181%19%257$231AI Chat
Monad2,68471%29%112$101AI Chat
Aleo2,63186%14%329$296AI Chat
Somnia1,30274%26%93$84AI Chat
Karatage1,22834%66%123$111Mixed
Flow82416%84%59$53Light
Attestant62888%12%45$41AI Chat
+ 8 orgs under 500 credits Circle, Solana, Ink, Polygon, Sui, Morpho, Kraken, Movement
Volume and health are not the same thing. The three highest-volume orgs (Marinade, CMC, Stockholm) consume 80% of all credits but have 0–3% AI usage. They’re running raw SQL pipelines — high revenue today, zero moat tomorrow. Meanwhile, orgs with balanced AI:Query ratios (CoW Swap at 36:64, Karatage at 34:66, Near at 31:69) are closer to the target pattern but consume 10–60x less. The gap between volume and health is the core product challenge.

2. The Model — What a Healthy Org Looks Like

The ideal customer uses AI agents to explore data and runs queries to extract it. Query-only users are high-volume but commoditized (any SQL warehouse will do). AI-only users are low-volume and substitutable (ChatGPT + a SQL tool). When both layers are engaged, AI drives queries and queries feed AI — a flywheel that’s hard to leave.

Validated Per-User Economics

Two internal power users with completely different workflows independently converged on similar consumption patterns. This gives us the per-user building block — the question is how many users per org we can drive.

Jim Myers — “Agent-First”Eric Stone — “Explorer”
AI : Query ratio52 : 4844 : 56
Credits / day~500326
Monthly projection~15,000 ($450)~9,800 ($294)
WorkflowCLI agents trigger cascading queriesChat exploration, then direct API queries
Channel78% CLI/Cursor, 49% no-session queries11 chat sessions, 105 direct API queries
InputValueBasis
Credits per user per day300–500Eric = 326, Jim = 500. Conservative: 350.
AI : Query ratio40 : 60Eric = 44:56, Jim = 52:48. Midpoint.
Active days per month22Weekdays
Revenue per user per month$198–$3306,600–11,000 credits × $0.03

What “Healthy” Means at the Org Level

A single user at $198–$330/month isn’t a business. Healthy means an org worth building for — one that justifies the $10M ARR ambition. That requires multiple users per org, both AI and query usage, and ideally a mix of programmatic and agent-driven access patterns.

Org StageUsersCred/U/DayCredits/MoRev/MoRev/YearHealth
Early — single user13006,600 $198$2,376Not yet healthy. Need to expand.
Growing — small team335023,100 $693$8,316Getting there. Team workflows forming.
Healthy — data team535038,500 $1,155$13,860Healthy. Pipelines + agents + analysts.
Strong — power team1045099,000 $2,970$35,640Strong. Multiple workflows, org-wide.
Enterprise — scaled20500220,000 $6,600$79,200Enterprise. Deep integration across teams.
A healthy org is $1,155–$2,970/month ($14K–$36K/year). That’s 5–10 active users doing 350–450 credits/day with a 40:60 AI:Query ratio. At the $10M target, we need an average ARPU of $1,211/month — which means most orgs need to be in the “healthy” band or above, not sitting at the single-user floor.

Today’s reality: avg ARPU is $196/month. Most orgs have 1–2 users. The path to healthy is driving seat expansion within existing orgs and ensuring new orgs onboard teams, not just individuals.

Why 40:60 AI:Query?

Both power users independently converged on ~40:60. Jim Myers (agent-first, CLI/Cursor) landed at 52:48. Eric Stone (explorer, web chat + API) landed at 44:56. The midpoint is 40:60. This ratio reflects the natural economics: queries are heavier per event (~7 credits vs ~1.5 for AI), and a single agent session triggers 3–10 downstream queries. The query tail is always longer.

The current platform average is 29:71 — skewed by three pipeline orgs running raw SQL with zero AI. That’s commodity behavior. Pushing toward 40% AI means users are in the agent layer, which is the differentiated product.

Margins are identical (80%) on AI and query credits. The 40:60 target is about stickiness and moat, not profitability per unit. More AI usage compounds revenue — one agent session at ~$0.75 of AI credits triggers ~$1.20 of query credits.

3. Working Backwards from $10M ARR

$10M ARR = $833K/month = 27.8M credits/month. The org count to $10M depends on how large orgs grow — not just how many you sign up. The distribution follows a power law: most orgs are small, but the tail is fat. Using the org stages from Section 2, here’s the mix we’re betting on.

The Path to $10M — Balanced Approach

Product drives adoption and expansion. Sales closes the largest accounts. This is the most realistic path given where we are today — 13 orgs, no sales team, strong product but early traction.

Org CohortCountAvg ARPURev/Mo% of RevWho These Are
Small (early / single user)300$250$75K9%New sign-ups, exploring. The top of the funnel.
Mid (growing teams)150$1,500$225K27%3–5 users, daily workflows. Expanding.
Large (healthy+)60$4,000$240K29%10+ users, pipelines + agents. The target org.
Enterprise (sales-assisted)50$6,000$300K36%20+ users, deep integration. Sales closes these.
Total560$1,500$840K100%
ARR$10.1MGP: $8.1M at 80% margin
Why this mix: 64% of revenue comes from orgs we can grow through product alone (small + mid + large). 36% comes from enterprise accounts that need sales. We don’t need a massive sales team to start — we need a product that drives expansion, and sales for the top 50 accounts.

The Expansion Curve — How Orgs Grow

The question isn’t just “how many new orgs?” — it’s “how fast do existing orgs expand?” If orgs grow from $250 → $1,200 → $3,500/month over 12–18 months as they add users and workflows, net revenue retention exceeds 150%. That expansion compounds — 80 orgs at $3,500/month is $280K/month alone.

StageUsersCredits/MoRev/MoWhat Changes
Month 1 — Sign up14,400$132First user explores with AI, runs a few queries
Month 3 — Add teammate213,200$396Second analyst joins, daily workflows forming
Month 6 — Data team538,500$1,155Team pipelines + agents running daily
Month 12 — Power team1099,000$2,970Org-wide: pipelines + agents + analysts
Month 18 — Scaled20220,000$6,600Full adoption, 20+ active users

Today’s evidence: Marinade is already at $1,646/month (54,870 credits). CMC Capital hit $4,680/month projected rate in just 4 days. Stockholm SSE hit $5,420/month projected. The spend levels are real — the question is whether we can get these orgs into the agent layer (currently 0–3% AI) to make the revenue sticky and expand it.

Two-Year Roadmap to $10M

OrgsAvg ARPURev/MoARRAI %What Has to Be True
Today
Feb ’26
13$196 $2,550$31K29% Starting point. 3 pipeline orgs carry 80% of credits. 9 orgs dormant.
Month 2
Apr ’26
18$228 $4,100$49K33% Activate 3–4 dormant orgs. Push AI adoption on Marinade/CMC/Stockholm. 1–2 net new orgs. ARPU stays low as new orgs ramp.
Month 4
Jun ’26
24$317 $7,600$91K37% Pipeline orgs starting to use agents. Early orgs deepening — first org hits $1K/mo. Adding 3/month net new.
Month 6
Aug ’26
30$416 $12.5K$150K40% Validation checkpoint. Dormant orgs activated. Pipeline orgs at 15%+ AI. 3–5 orgs above $500/mo. Expansion signal visible.
Month 9
Nov ’26
55$680 $37.4K$449K41% Adding ~8 orgs/month. Early cohorts expanding — 5–8 orgs above $1K/mo. ARPU inflecting as expansion kicks in.
Month 12
Feb ’27
100$1,040 $104K$1.25M42% Scale checkpoint. Adding ~12 orgs/month. 15+ orgs above $1K/mo. First $6K+ accounts. NRR > 130%.
Month 15
May ’27
200$1,125 $225K$2.7M43% Adding ~30 orgs/month. Expansion engine running. Multiple orgs at $3K–6K. Dedicated sales for enterprise tier.
Month 18
Aug ’27
350$1,190 $417K$5M44% Adding ~40 orgs/month. 30+ orgs above $3K/mo. Enterprise pipeline building. NRR > 150%.
Month 21
Nov ’27
530$1,200 $636K$7.6M44% Adding ~50 orgs/month. Portfolio maturing — older cohorts at $3.5K+ pulling avg ARPU up.
Month 24
Feb ’28
710$1,211 $860K$10.3M45% Target. ~80 orgs at $3.5K+, ~200 at $1.2K, ~400 at $250. ~30 enterprise at $6K+. Adding ~60 orgs/month.
What makes this aggressive: going from 13 orgs to 710 in 24 months requires adding ~30 orgs/month on average, accelerating to 60/month by the end. That’s a real go-to-market engine — not just product improvements. It also requires that early orgs actually expand: ARPU needs to grow from $196 to $1,211, which means orgs are adding users and deepening usage over time. If expansion stalls, you need more orgs to compensate — and the math gets much harder.

The first 6 months are the proof point. If we can’t get from 13 to 30 orgs and move AI% from 29% to 40% by August ’26, the 24-month target needs to be revisited. That’s the gate.

4. The Playbook — What to Build

The thesis says programmatic usage drives revenue. The playbook follows: prioritize product investments that move users up the programmatic ladder, ranked by expected credit impact.

1. Agent-to-query pipeline
5–10x

Build: Make it seamless for an agent session to trigger, iterate, and schedule follow-up queries. Today an agent session generates ~$0.75 in AI credits and ~$1.20 in query credits. If that query tail extends (scheduled refreshes, automated pipelines spawned from chat), each session becomes a recurring revenue stream.
Impact: Moves AI Chat orgs (6 orgs, 13% of credits, 79% AI) toward balanced consumption. Laminated, Avalanche, Monad, Aleo are all exploring with agents but never cross into query execution. This is the bridge.
Ship: “Save as scheduled query” from chat, agent-initiated query pipelines, one-click automation from results.

2. API & CLI developer experience
20–60x

Build: First-class API/SDK/CLI for running queries and invoking agents programmatically. The three pipeline orgs (Marinade, CMC, Stockholm) already hit $1.6K–$5.4K/month through raw API access alone — but with zero AI because there’s no good way to invoke agents from a pipeline.
Impact: +$1,957/mo immediately if pipeline orgs reach 40% AI at current query volume. Also unlocks the highest-spend tier for every new org — a team with API access spends 20–60x more than a UI-only user.
Ship: Agent API endpoints, Python/JS SDKs, CLI agent invocation, MCP server improvements, webhook triggers.

3. Team onboarding & seat expansion
ARPU × n

Build: Make it trivial to add teammates to an org and get them producing value on day one. Today most orgs have 1–2 users. Healthy is 5–10. Every additional user at $198–$330/month is linear ARPU growth.
Impact: Moving avg org from 1.5 users to 5 users triples ARPU with no acquisition cost. 9 dormant orgs (43% of base) never activated — likely an onboarding problem, not a product problem.
Ship: Team invite flow, shared dashboards/queries, org-level billing visibility, “first query in 5 minutes” onboarding.

4. Recurring credit consumption
NRR

Build: Features that generate credits on a recurring basis without human action. Scheduled queries, automated alerts, embedded dashboards that refresh, agents that run on a cron.
Impact: Transforms one-time exploration into ongoing baseline consumption. This is what makes NRR > 120% — orgs consume credits even when nobody is at their desk.
Ship: Scheduled query execution, alert thresholds on query results, embeddable dashboards, agent scheduling.

5. Key Risks — What We Don’t Know

RiskWhy It MattersHow to Validate
Pipeline orgs churn 80% of credits come from 3 orgs running raw SQL. They have zero switching cost — any SQL warehouse works. If one churns, revenue drops 25%+ overnight. Get pipeline orgs into the agent layer. AI usage = switching cost. Track AI % per pipeline org monthly.
Per-user economics don’t replicate 300–500 credits/day is validated on 2 internal power users. External users may consume less, use AI differently, or not adopt agents at all. Instrument per-user daily credit consumption for external orgs. Need 10+ external data points in 3 months.
Orgs don’t expand The model assumes orgs grow from 1 to 5–10 users. If most orgs stay at 1–2 users, avg ARPU stays at $200–400 and $10M requires 2,000–4,000 orgs. Track users-per-org and ARPU monthly. If no expansion signal by month 6, reassess the target.
Burst usage isn’t sustained CMC Capital and Stockholm SSE consumed credits in 3–4 day bursts. Monthly projections assume sustained usage. Actual may be 50–70% lower. Wait for 60-day observation window. Don’t plan around burst projections until confirmed.
Acquisition engine doesn’t exist yet Going from 13 to 710 orgs in 24 months requires adding ~30 orgs/month average. Today there’s no scalable acquisition channel. Identify top-of-funnel channel by month 4. Test: content, partnerships, PLG viral loops, outbound. Need 5+ orgs/month by month 6.

6. Scoreboard

MetricToday6-Month TargetWhy It Matters
AI % of total credits 29% 40% Moat. When this rises, users are in the agent layer.
Avg ARPU / org / month $196 $500+ Expansion. Are orgs growing spend via seats and usage?
Orgs > $1K/month 3 10 Power law. High-ARPU orgs drive the majority of revenue.
Orgs with both AI & Query >10% 7 20 Flywheel health. Balanced consumption = sticky.
Net Revenue Retention (monthly cohort) >120% The metric. If existing orgs expand, you win. Track monthly.
Pipeline orgs using agents 0 / 3 3 / 3 Expansion. Biggest revenue lever, no new sales needed.
First-query activation (30-day) 57% 85% Funnel. 43% of orgs never activated.

Appendix: Methodology & Limitations

Data source: Production OrgUsageEvent table via internal-cli -e prod admin usage-events. 21 external organizations, Flipside internal org (538,644 credits) excluded.

AI:Query classification: Events have category field: ai_usage (chat completions) or query_execution (SQL execution). Credits are calculated by the usage event service based on token counts and query seconds respectively.

Programmatic vs manual heuristic:
session_id = NULL → Direct API (no chat session involved)
ChatSession.meta.source = "cli" | "api" → CLI/MCP-driven
ChatSession.meta.source = "web" | "agents-page" → Web UI

Prototypical org derivation: Based on two validated users: Jim Myers (agent-first, 500 credits/day, 52:48 AI:Query, 298 sessions over observation period, 78% CLI-originated) and Eric Stone (explorer, 326 credits/day, 44:56 AI:Query, 11 sessions + 105 direct API queries, 5.8% burst rate). Conservative blend: 300 credits/user/day, 2.5 users/org, 22 active days/month.

Projection assumptions: New orgs ramp to prototypical over 2–3 months (25%/60%/100% for base case). Existing org growth: 10–20% MoM as users adopt AI + query workflows, capped at 1.5x prototypical per org. Churn: 1–5% monthly depending on scenario.

Limitations:
• Session source metadata only available for internal users (Jim, Eric). External org session sources inferred from session_id presence/absence.
• CMC Capital and Stockholm SSE consumed credits in 3–4 day bursts; monthly projections assume sustained usage.
• Eric’s observation window is 5 days — his monthly projection (9,800) is directional, not confirmed.
• The 2.5 users/org assumption is not validated against external org data — some orgs may have 1 user, others 10.