Retention, Unit Economics & the Expansion Flywheel
Analyst note
Blended GRR is 78% (Feb 2026). ICP GRR is 93%. The gap is explained by a churn decomposition: 67% of historical churn was onboarding failure (no CS team, customers never reached first value), 33% was non-ICP and involuntary SMB loss. Onboarding was fixed in March 2025 (activation 42% → 85%). CAM Assist 2.0 shipped December 2025. ICP qualification now filters pre-sale. The page models blended GRR converging to ~90% by mid-2027 as improved cohorts wash through the 12-month lagging metric. Top 100 users average 11.6x ROI and save $4.7M combined annually. CAC is $5,347, LTV:CAC is 3.9x on land alone, 4.9x blended with expansion. Four proven acquisition channels (inbound 4.6 ROAS, sales-led, resellers, PLG). Long-term maturity assumes ~90% GRR, not 95%+, because structural SMB churn is honestly modelled as a permanent floor.
Method: cohort retention analysis (Q1 2025 – Q1 2026), churn root-cause decomposition, per-factory usage telemetry, multi-channel CAC attribution.
Headline Metrics
The Retention Story
Historical churn broke down into two categories. The dominant driver was entirely fixable. The remainder is structural and honestly modeled.
Historical churn breakdown
What we fixed
The Lagging Metric Story
Blended GRR is a 12-month lagging metric. It includes every historical cohort: pre-CS, pre-product-fix, non-ICP. That number is rising, but it takes a full year for improved cohorts to fully wash through. The leading indicator is ICP GRR, which shows what happens when we sell to the right customers with the right product.
GRR trajectory
| Period | Blended GRR | Why |
|---|---|---|
| Dec 2025 (Today) | 78% | Dragged down by pre-fix cohorts. 12-month lagging metric. |
| Jun 2026 | ~84% | Onboarded cohorts hitting 12-month anniversary. Much more significant uptake of CAM Assist 2.0. |
| Dec 2026 | ~88% | High ICP percentage in new cohorts washing through. |
| Jun 2027+ | ~90% | Long-term maturity. Structural SMB churn is the only remaining drag. |
Product-Market Fit Signal
CAM Assist is delivering substantial value today, and the power user data quantifies exactly how much.
Power users prove the ROI is real
A single power user saves up to $391K per year. The top 100 save 46,500+ hours annually. That is 22 full-time engineers' worth of capacity returned to the shop floor.
CAM Assist 2.0 delivers proven ROI today across 3-axis and 3+2 machining. The agentic engine currently in development extends this to a significantly wider range of parts, a higher percentage of completion on each part on the path to full autonomy, and opens upstream DFM and quoting as new product surfaces. Today's retention and ROI numbers represent a strong foundation, not the ceiling.
What the Agentic Layer Unlocks
CAM Assist 2.0 delivers strong ROI for mid-size shops with high-mix, low-batch work across 3-axis and 3+2 machining. These customers are retained at 93% GRR. However, for smaller shops, while individual part ROI is real, their volume of supported part types means CAM usage is less frequent, and the absence of DFM and quoting support means their highest-frequency use cases are not yet served.
The agentic layer addresses this through two mechanisms:
Together, these two mechanisms expand both the depth of value per factory and the breadth of factories that engage daily. This is the foundation of the expansion model.
Unit Economics: The Cost-Effective Land Grab
The land metrics demonstrate efficient customer acquisition across multiple channels, providing a strong base for scaled expansion.
With proven acquisition efficiency across four channels and a 1:1 all-time S&M-to-ARR ratio, we have demonstrated that scaling spend translates directly into scaling ARR. As expansion credits drive per-factory revenue well beyond the initial land ACV, blended LTV:CAC compounds from 4.9x today toward 15-18x at scale.
Multi-channel acquisition, de-risked
Four acquisition channels have been proven. We are not dependent on any single one. If any channel saturates, the others pick up.
The Expansion Flywheel
The strategy is a three-phase land-and-expand play. Acquire factories cheaply with CAM Assist, expand them with agentic products, then lock them in by dominating the workflow.
Product Expansion Stack
Expansion Model: Bottom-Up Proof
Every number in the expansion model traces back to observable data. This section walks through each assumption, shows per-factory unit economics, and cross-checks the aggregate against real usage.
Key Expansion Assumptions
| Assumption | Value | Reasonableness Check |
|---|---|---|
| 1 credit = 15 min saved | $80/hr fully loaded machinist cost | $80/hr reflects the fully loaded cost including overhead, benefits, and charge-out rate shops use in quotes. 15 minutes is the lowest unit of saving (a simple part quote or DFM check). Complex CAM can save significantly more. |
| Full credit price | $5.03 | Value-to-price ratio: 4× at full price ($20 value / $5 price). Customer ROI starts higher during growth (6.4×) as discounted credits deliver outsized value, and settles at 4× at mature pricing. |
| Launch discount | 50% off ($2.51/credit) | Initial discount drives adoption. Fades over 24 months to full price. Customers capture 8× ROI initially, settling to 4× at maturity. |
| Starting credits per user/month | 60 | 20 parts × 3 credits/part. 20 parts/month ≈ 1 part/working day. At this volume the tool becomes embedded in the daily workflow. |
| Mature credits per user/month | 130 | 130 credits/user/mo = 32.5 hrs saved per user/mo ≈ 8 hrs/week. A machinist works ~37.5 hrs/week (7.5 hrs/day). At maturity we automate ~21% of their time across CAM, DFM, and quoting tasks. |
| Mature users per account | 4 (industry avg) | Weighted average across the target market. 83.9% of CNC shops have <20 employees. In a typical 10-person shop, 3-5 people interact with CAM, quoting, and DFM (machinists, estimators, operators, owner). Larger shops skew higher. |
| Ramp to maturity | 16 months | Time for both user count and per-user usage to reach steady state. Consistent with typical SaaS expansion ramps (12-24 months). |
Per-factory credit ramp (at scale pricing, excluding launch discounts)
| Stage | Month | Users | Credits/User | Total Credits | Free | Billable | Expansion Rev/Mo | Customer Value/Mo | Credit ROI |
|---|---|---|---|---|---|---|---|---|---|
| Expanding | 6 | 2 | 80 | 160 | 100 | 60 | $226 | $1,447 | 6.4× |
| Scaling | 10 | 3 | 105 | 315 | 100 | 215 | $901 | $4,804 | 5.3× |
| Mature | 16 | 4 | 130 | 520 | 100 | 420 | $2,113 | $8,442 | 4.0× |
Time replaced per factory: At maturity, 4 users consuming 130 credits each at 15 minutes per credit equates to 130 hours of manual work replaced per factory per month, or approximately 8 hours per user per week. CAM programming, DFM review, and quoting collectively represent the majority of a machinist's non-machining time. The model automates ~21% of total working hours — a meaningful but achievable share of these high-frequency tasks.
Revenue per factory at maturity: Expansion credit revenue reaches ~$2,100/mo per factory (~$25K/yr) on top of the ~$6.5K base subscription. Total ARPU of ~$31.5K/yr, growing as the product surface area expands. This revenue trajectory is anchored in the credit ramp above, with each step traceable to user count and usage frequency.
Power user cross-reference
The model assumptions can be validated against actual power user data:
| Metric | Model Assumption (Mature) | Power User Actual (Top 100) | Reasonableness |
|---|---|---|---|
| Hours saved per user/year | 390 hrs | 465 hrs/year | Model is 84% of observed actuals. Conservative. |
| Dollar value per user/year | $31,200 | $46,600 | Model assumes average users, not power users. 67% of actual. Conservative. |
| ROI on seat cost | 4-6× on expansion credits | 11.6× on subscription | Power users massively over-index. Model reflects the average, not best case. |
The expansion model does not require exceptional users. It applies a significant haircut to current power user performance and projects that discounted level across the broader customer base. Power users are already achieving these numbers today on a product without the agentic engine. The model assumes the average factory reaches 67-84% of observed power user output as the product expands.
Why Removal Becomes Impossible
As factories adopt more of the platform, CloudNC moves from a productivity tool to operational infrastructure. At full adoption, we handle the workflow from part acceptance through CAM programming, quoting, DFM, tooling selection, and scheduling. Each additional module increases the value delivered and deepens integration into daily operations.
The competitive dynamic at this stage is pen and paper versus a spreadsheet. A factory running the full CloudNC stack operates at a fundamentally different speed and cost structure to one without it. The question is not whether they want to keep paying — it is whether they can afford to compete without it.
This translates directly into compounding revenue. Each new module drives expansion within the existing base. NRR crosses 100% during 2027 and reaches above 150% by 2028, with average revenue per customer compounding year over year as factories adopt more of the platform.
The expansion flywheel compounds on three axes simultaneously: more factories joining the platform, more products adopted per factory, and more credits consumed per product. Each axis reinforces the others. Factories that adopt the full stack become operationally dependent on it, and each year they derive more value and pay more revenue.
All amounts in USD. GBP figures from financial model converted at £1 = $1.34. Financial model v4 (Mar 2026). Power user analysis (annualized). Retention cohort data from internal CS tracking. ICP GRR from CS tracking. Blended GRR and unit economics from financial model KPI summary. Expansion timeline from product roadmap (board-approved). CAMBench scores from internal evaluation.