Price and package a product

To: CEO From: Head of Product Re: Alternative to the across-the-board 30% increase

Summary

Margin fell from 41% to 24% because pages are free inside the seat price and our cost is per page. Six enterprise accounts use 71% of all pages and pay 44% of revenue. Their combined gross margin is −22% (−$4.2k on $19.0k last month), and four of the six lose money. Startups run at 87% margin and mid-market at 41%.

A flat 30% doesn't fix this, for three reasons: - Contracts can't be repriced in January. All six renew in 2027, and I haven't seen any reopener clauses. - Enterprise stays near zero margin. Even at +30%, enterprise would reach only about 6% margin, and it would erode again as pages grow. Northwind's pages tripled in a year. - The price test doesn't support it (below).

I recommend pricing on pages, which is what drives our cost.

Proposed model and effect by segment

Figures are monthly, at last-30-day volumes and today's $0.0012 cost per page, with no churn.

Why these rates

• Overage at $0.005: about 76% margin on the extra pages. It is below the cheapest competitor's per-page rate ($0.008).
• Enterprise at $0.002: 40% margin today. If cost keeps falling 10% a quarter, it reaches about $0.0008 after four quarters, which puts this rate near 60% margin. That gives room to concede on rate later without losing money.
• Competitor comparison: at Northwind's 4.8M pages, DocuLens list price is about $43k a month. We would charge $9.6k. Arcadia is evaluating a competitor, but at its 3.3M pages DocuLens lists near $31k against our $6.6k.
• Enterprise seats: I'm dropping seat charges. Seats don't drive cost and mostly add negotiating friction.

Gross margin comparison

The CFO plan assumes zero churn. Even on that assumption, its best case only gets back to last year's 41%.

Getting there

Self-serve (mid-market and startup): - Meter and display pooled page usage in the dashboard now. - Send mid-market a 60-day notice with a personal usage preview. Launch the new plan around April 1. - Charge overage at half rate for the first quarter, which softens the three biggest increases. - Startups see only a new allowance line, with no change in price.

Enterprise: Start before the renewal dates, and have legal check the contracts first. - Northwind (Feb 2027): Decline "the same rate". At 3x page growth again, its cost would be about $17k a month against $3.2k of revenue. Offer a two-year deal instead, with a price lock, a committed volume and a modest discount above 6M pages. That gives them the certainty they want. - Halden (Jun 2027): Same structure as Northwind, with an early-renewal offer. - Arcadia (Mar 2027): It has the smallest increase of the six accounts (+47%) and a competitor in play. Lead with value, and keep the rate flat to the competitor's list. - Copperfield (Jan 2027): First to renew, so the first live test of the model. Start talks in Q2 2026. - Meridian and Brightwater: Standard offer at renewal.

Northwind and Halden together lose about $4.7k a month, so losing them would improve gross margin dollars. We can hold the line. For the two losers the loss continues until renewal unless they take an early-renewal offer, so I'd offer a glide path (a year-one rate between $0.0015 and $0.002) in exchange for committing early. That keeps even the worst case above cost.

The price test

I wouldn't use it as cover, because it doesn't show what the readout says.

1. The split is broken. A 50/50 split of 2,110 visitors should give about 1,055 per arm. We got 1,240 and 870, which is a 59/41 split. That is far too lopsided to be chance (p well below 0.001). Something in assignment or tracking failed, so the arms aren't comparable.
2. "Not significant" is not "no effect". The difference is −0.17 points, with a 95% interval of about −2.0 to +1.7 points. $129 pays off only if conversion stays above about 3.8% (a drop of less than 23%), and the data can't rule out falling below that.
3. It's the wrong test. It ran two weeks in December. It covers new sign-ups only and measures starting to pay, not retention or usage. It says nothing about existing customers, enterprise, or per-page pricing. Under my plan startups don't change price at all.

To test $129 properly, we'd need working randomisation and revenue per visitor plus 90-day retention as the measures. Detecting a one-point conversion change needs about 7,500 visitors per arm. I'd run that only if we want a startup price rise later.

Risks and assumptions

• Figures assume static volumes, no churn, and today's unit cost. Enterprise page growth raises both revenue and cost under the new model, and the per-page rate protects margin.
• Mid-market bill shock is the main self-serve risk. The three accounts with 2–4x increases are the ones most worth a personal call. They are also the three loss-makers (P, Q and V).
• A 2–3x enterprise increase may cost us one or two accounts. I think that is acceptable, but it is a judgment call.
• Full enterprise run-rate arrives only after the last renewal in November 2027. Savings this year come mostly from self-serve and any early renewals.
• Not yet checked: contract terms (fair-use or reopener clauses) and whether billing can meter pooled pages by April.

Decision requested: approve the model for planning, and let me begin the Copperfield and Northwind conversations this quarter.

MEMORANDUM

TO: Chief Executive Officer FROM: Head of Product, Parsewell DATE: October 24, 2024 SUBJECT: Alternative Pricing Model vs. Tom’s 30% Across-the-Board Increase

---

1. Assessment of the Growth Team’s Price Test

Tom’s reliance on the December price test to justify a blanket 30% price hike is dangerously flawed. The test suffered from two fatal errors: 1. Sample Selection & Volume Discrepancy: Arm A ($99) attracted 1,240 visitors, while Arm B ($129) attracted only 870 visitors—a 30% drop in top-of-funnel capture. This points to a broken traffic allocation or immediate audience suppression at the higher price point. 2. Confounding Variables: Running a pricing test during December (8th–21st) captures holiday-shortened enterprise/developer activity, creating noise that masks true price elasticity.

While a 30% seat price increase showed a statistically insignificant conversion dip over a two-week window, applying that logic to usage or enterprise contracts ignores reality. Parsewell is an infrastructure API. Developers do not buy seats for fun; they buy seats to build systems that process millions of pages. Tying our revenue strictly to developer seats while our core cost driver—and customer value metric—is document volume is a structural mismatch.

---

2. The Current State: Margin Erosion & Value Leakage

An analysis of our usage export reveals why Tom is panicking about gross margins (currently sitting at roughly 24% overall): * The Enterprise Subsidy: Our 6 Enterprise accounts processed 19.16 million pages last month (79% of total company volume) but generated only $18,600 in revenue. Because they enjoy "unlimited pages" tied to discounted seat licenses, their heavy usage swamps our inference costs ($0.0012/page). Consequently, 4 out of 6 Enterprise accounts have negative gross margins, costing us thousands monthly (e.g., Northwind Logistics loses $2,584/month). * The Self-Serve Engine: Mid-market and Startup tiers are healthy, generating positive gross margins because their volume scales predictably with seat counts.

A flat 30% seat hike fixes nothing for Enterprise: adding 30% to Northwind’s $3,200/month bill reduces their negative margin, but they still bleed cash because they process 4.8 million pages for a flat fee.

---

3. The Alternative Pricing Model: Value-Based Usage Tiers

To fix gross margins structurally, we must decouple our monetization from developer seats and align it with the metric that drives our costs and customer value: processed pages.

We propose launching Parsewell V2 on January 1: 1. Self-Serve (Startups & Mid-market): Retain the base seat fee ($99/seat/month) to cover dashboard access, but introduce a generous included page allowance (50,000 pages/seat/month) with a clear, predictable overage rate of $0.005 per page. 2. Enterprise Contracts: Move away from flat seat-license bundles. Transition to a Platform Fee + Usage Model modeled after industry standards (competing with DocuLens/ExtractIQ): * Base Platform Fee: $1,500/month (includes dashboard, SSO, priority support, and 250k pages). * Overage / Metered Rate: $0.008 per page for all volume exceeding the base allowance.

#### Financial Impact Summary (Based on Last 30 Days Data) * Current Monthly Revenue: ~$41,000 | Current Cost to Serve: ~$31,000 | Current Gross Margin: ~24% * New Model Monthly Revenue: ~$81,500 | New Cost to Serve: ~$31,000 | New Gross Margin: ~62%

---

4. Segment-by-Segment Impact & Table

---

5. Execution Strategy & Enterprise Renewals

Transitioning self-serve is straightforward via a Stripe billing update with a 30-day grandfathering notice. Enterprise renewals require a tactical playbook to prevent churn, especially with accounts like Arcadia Health evaluating competitors.

• Copperfield Freight (Renews Jan 2027): Their contract is up immediately in January. We will introduce the new usage-based framework during December QBR. Position the $0.008/page rate against competitor DocuLens ($0.012/page) to show they are still getting a volume discount.
• Northwind Logistics (Renews Feb 2027): They want a 2-year renewal at the same rate despite a 3x volume surge. Do not accept. Counter-offer: Honor the flat seat rate for 12 months only if they agree to a volume cap of 2M pages/month, with overages billed at $0.008/page thereafter. Otherwise, transition them directly to the new platform + usage model.
• Arcadia Health (Renews Mar 2027 - Evaluating Competitor): Because they are price-sensitive and shopping around, offer a transitional "Volume Partnership Discount": a lower platform fee ($1,000) for Q1 while locking them into a 2-year tiered page commitment.

Conclusion

Tom wants a blunt instrument (30% across the board) to solve a surgical problem. A flat 30% price hike leaves money on the table for high-volume users while alienating self-serve customers. Our value-based usage model repairs our gross margins to over 60%, prices our heavy enterprise users fairly based on the compute power they consume, and keeps us competitively positioned against ExtractIQ and DocuLens.

Recommendation to the CEO

Reject the blanket January increase. Replace unlimited-page seat pricing with a platform fee plus metered pages, and migrate enterprise accounts at contractual renewal. Validate the proposed rates before broad rollout; the December test does not establish willingness to pay for this model.

The problem is the pricing metric, not simply the price level. Dashboard seats do not track extraction value or cost. Six enterprise accounts generate 19.36 million pages—71% of all usage—but lose $4,232 a month at today’s costs. Meanwhile, small startups subsidize heavy users.

Across all 80 accounts, current monthly revenue is $43,057, cost to serve is $32,903, and gross margin is 23.6%, consistent with Tom’s reported 24%.

Proposed model and customer impact

Include unlimited dashboard users. Charge for successfully processed pages, with clear definitions for multipage documents, retries and failures. The calculations below assume exported pages are billable; validate that assumption before launch.

These are static, fully migrated scenarios, not forecasts: they assume unchanged usage, no churn and today’s exported costs. Monthly gross profit rises from $10,154 to $94,073. A further 10% cost reduction would raise proposed gross margin to approximately 76.7%, but we should not depend on it.

The startup revenue reduction is deliberate: remove the collaboration tax and preserve an accessible entry point. Heavy users should fund their own consumption.

The rates are plausible hypotheses, not proven willingness to pay. Our $0.006 self-serve rate is below the published competitor page rates; enterprise at $0.004 is half DocuLens’s published high-volume rate before platform fees. Those comparisons require fresh verification and equivalent accuracy, features and service levels.

Why not Tom’s approach?

At unchanged usage, a universal 30% increase produces $55,974 monthly revenue and 41.2% gross margin. It restores the historical aggregate margin but leaves the structural problem intact:

• Enterprise gross margin remains only 5.9%.
• Northwind and Halden remain loss-making.
• Growing page volume can erase the improvement without adding seats.
• Low-volume customers pay more despite being highly profitable already.
• Existing enterprise contracts cannot simply be repriced in January without contractual authority or agreement.

Our alternative is economically stronger, but its large mid-market and enterprise increases create meaningful retention risk. We must test that risk rather than treat modeled revenue as secured.

Migration and enterprise renewals

January: launch metering and shadow invoices, not a forced universal increase. Instrument billable pages, audit costs and billing, and show customers their projected bills. Provide usage forecasts, threshold alerts and optional spend limits; do not silently interrupt production workflows.

For new self-serve customers, pilot the model after repairing experimentation. For existing self-serve customers, provide at least 90 days’ notice, subject to applicable terms. Offer a six-month, explicitly expiring migration credit for large increases. Publish both discounted and eventual bills so credits do not disguise the destination price. Separate price effects from usage growth in reporting.

For enterprise, honor existing agreements until renewal unless customers voluntarily amend. Proposed first-renewal pricing is $500 + $0.003/page for 12 months, stepping to $0.004 in year two. Even the transitional rate produces approximately 62% enterprise gross margin at current usage and cost.

Renewal sequence, with monthly-equivalent first-year prices:

• Copperfield, January 2027: $7,640; begin negotiations immediately.
• Northwind, February: $14,960. Reject two years of unlimited usage “at the same rate.” Offer two-year rate certainty only against a defined page commitment, with paid excess usage. Its threefold volume growth makes this essential.
• Arcadia, March: $10,430. Start an executive-led retention discussion now; benchmark extraction quality and switching economics before conceding price.
• Halden, June: $12,380.
• Meridian, September: $6,950.
• Brightwater, November: $8,720.

Commitments should reflect validated usage and seasonality, with transparent true-ups and no unlimited overage loophole. Do not promise that renewal discounts are permanent. Escalate concessions below our margin floor to Finance and Product jointly.

Revenue will ramp through these renewals and subsequent step-ups, not jump to the table’s total in January. Build the forecast account by account, including churn, credits and volume scenarios.

One data issue needs resolution first: enterprise revenue per listed seat exceeds $99, despite the pricing document describing discounted seat rates. Use observed revenue for this analysis, but reconcile actual contract units and entitlements before quoting renewals.

What the price test actually says

The intended 50/50 allocation produced 1,240 versus 870 visitors. That is severe sample-ratio mismatch: approximately χ² = 64.9, p < 10⁻¹⁴ under equal allocation. Investigate assignment, exposure logging, eligibility, channel mix and exclusions before trusting the causal result.

Observed conversion was 5.00% versus 4.83%. Ignoring the allocation problem, the difference is approximately −0.17 percentage points, with a rough 95% interval of −2.04 to +1.70 points. “Not statistically significant” does not establish equivalence or rule out commercially material harm.

It also tested only new self-serve sign-ups, over two weeks—not existing customers, retention, enterprise renewals or usage-based pricing.

Repair the experiment, predefine a commercially meaningful conversion-loss threshold, and measure revenue per eligible visitor, paid activation, retention, usage and contribution margin. Use evidence to choose pricing—not the test as cover for a predetermined increase.