Estym8 Pricing Strategy

Product context: Estym8 is built from scratch as an AI-first construction preconstruction platform—not a legacy takeoff stack with AI bolted on. AI runs across the product: bid-package ingestion and classification, multi-model takeoff and vision, plan intelligence, cross-file synthesis, Estee, estimate-to-submittal draft review, and optimization recommendations. Canonical framing: AI-first positioning.

A framework for thinking through pricing structure, charge models, and price points.

One-pager for sales/calls: Pricing one-pager — one-liner, tier table, talking points. Upsell (access → paid): Upsell strategy — when and how to convert users who need an active plan (and reinforce value after the first successful takeoff).

---

1. Value Proposition

Full solution description: See Vision & value for the complete value prop, solution pillars (takeoff, plan intelligence/RFI, optimization, Estee), and product reflection.

What we deliver (summary):

• Bid-ready takeoff from plan PDFs (electrical, mechanical, plumbing): device counts, materials, raceway/wire
• Plan intelligence: conflicts, code checks, draft RFIs, concerns, optimization recommendations
• Editable counts, discipline breakdown, export; Estee for questions, “Should I file an RFI?”, and the same run-grounded Q&A as before—plus, on an open estimate, preview + confirm before Estee applies supported summary edits (e.g. material overage on totals).

Time saved: Estimators spend 1–3 days (24–72 hours) on manual takeoffs. Estym8 delivers in 15–45 minutes.

---

2. Charge Models to Consider

---

3. Competitive Context

• Manual takeoff: In-house labor ($50–150/hr) or outsourced ($0.10–0.50 per sq ft)
• PlanSwift, Countfire, etc.: $50–200+/month or per-seat
• AI/automation tools: Emerging; pricing varies

---

3a. Market & Competitor Pricing (2024–2025)

Market size (BLS / industry)

• MEP estimators (US): ~25,000–35,000 (building equipment contractors + overlap in general construction).
• Avg salary: $82,000–$85,000/year ($39–42/hr loaded).
• Time per estimate: 24–72 hours (1–3 days) typical; 10–80 hours by project type; takeoff phase = 50–70% of total.
• Implication: 20–30% time savings across the segment = billions in productivity; Estym8’s 15–45 min delivery is a 20–50× time reduction vs. manual.

Competitor pricing (benchmarks)

Value anchor (why we can charge)

• Time: 1–3 days → 15–45 minutes. Saving even 1 day (8 hrs) × $40/hr = $320 value per estimate.
• Labor alternative: $200–500 per plan set (internal) or $150–400 outsourced. Estym8 should sit below that with faster turnaround.
• vs. $30/page: 50-page set = $1,500 at $30/page + 3-day wait. We offer speed (same day) and simplicity (per estimate, not per page); subscription smooths cost and avoids “not all pages equal” debates.

---

3b. Why Subscription (Not Per-Page) for Estym8

• Per-page is messy: “Not all pages are created equal” (Jonas)—complexity and scope vary; per-page invites disputes and doesn’t map cleanly to one “estimate” deliverable.
• Per-estimate / per-project aligns with how estimators think but is harder to meter automatically (what counts as one project?).
• Subscription with N takeoffs/month is simple, predictable, and matches PlanSwift/Countfire; we meter “one takeoff = one plan set processed” and cap by tier. Easy to explain and sell.
• Optional: Later add overage (e.g. $X per takeoff over tier limit) for heavy users without raising base price.

---

4. Suggested Structure (MVP)

Tier 1: Entry (sign up) — current gate

• Accounts register with no automatic takeoff unlock; checkTrialStatus requires verified email plus plan or approval (see lib/trial/trial-manager.ts: paidAccessAt, approved tester, demo, admin, etc.).
• Dashboard/onboarding still apply; running jobs requires access.
• Goal: Clear get access path (pricing, contact, tester approval) → paid subscription when checkout exists.

Tier 2: Paid (per-seat subscription)

Stripe checkout slugs: starter = Solo, pro = Studio (2-seat minimum list price).

Why these bands:

• Solo $179: Premium vs Countfire/PlanSwift point tools; still far below Trimble-class enterprise and outsourced takeoff per project.
• Studio $149/seat: Scales with shop size; pooled runs match bid-week folder volume.
• Annual discount: 15% (ANNUAL_DISCOUNT_PERCENT in lib/pricing/public-plans.ts); Solo ~$1,826/yr vs $2,148 at monthly list.
• Overage: $35 per completed run above allowance.

(Entry: sign up → verify → active plan or approval to run takeoffs.)

---

5. Price Point Guidelines

Anchoring:

• Labor: ~$200–500 per plan set (in-house) or $150–400 outsourced; 24–72 hrs at $40–80/hr.
• Software: PlanSwift ~$167/mo; Countfire $99/mo; AI tools $50–200/mo.
• AI per-page: ~$30/page, 3-day lead (market reference from team).
• Value delivered: Hours saved × hourly rate (e.g. 8 hrs × $40 = $320 per estimate).

Recommended ranges (May 2026 launch):

• Solo: $179/month – 1 seat, 8–10 completed runs, Estee, export.
• Studio: $149/seat/month (min 2 seats) – 12 runs/seat pooled.
• Firm: $129/seat/month (6–15 seats) – contact for onboarding.
• Enterprise: Custom – SLA, dedicated support, 16+ seats.
• Overage: $35 per completed run above allowance.

---

5b. Sales & Positioning (value in time and $)

One-liner: “Estym8 turns 1–3 days of takeoff into 15–45 minutes. One saved day pays for the month.”

Talking points:

• vs. manual: “Typical MEP estimate is 24–72 hours. We deliver in 15–45 minutes—same bid-ready counts, editable, with Estee for run Q&A; on an open estimate, supported summary tweaks preview first and save only after you confirm.”
• vs. $30/page AI: “We’re subscription so you don’t worry about page count or ‘not all pages equal.’ One plan set = one takeoff. And we’re same-day, not 3-day lead.”
• vs. outsourced: “Outsourced runs $150–400 per project and 1–6 days. We’re a fraction of that per estimate and you keep control in-house.”
• ROI: “At $82k salary, an estimator’s hour is ~$40. Saving 8 hours on one estimate = $320 value. Solo is $179/mo—one estimate pays for it.”

Internal cost (for Paul/Jonas): Token/page or per-takeoff cost is useful for margin and overage; “not all pages equal” is handled by not pricing per page—we price per takeoff (one plan set) and use subscription tiers so we don’t have to expose page-level complexity to the customer.

---

6. What to Charge For

---

6a. Unit Economics — Model / Token Cost per Takeoff

Detailed cost math (with real measured run data and per-provider rates, including Anthropic prompt-cache breakdown) lives in Model cost analysis (internal COGS notes) lives in the repository at docs/MODEL_COST_ANALYSIS.md — not published on this site.

One takeoff in our metering = one folder run (one bid package), typically 1–4 trades. Numbers below come from the instrumented LLM event log (test-results/llm-usage/events.jsonl) — every Anthropic call is priced from the actual token + cache breakdown, not estimated.

Observed cost per folder (DEN47 measured 2026-05-01, all Claude Sonnet 4.5)

DEN47 actual full-folder spend: $2.88 (15 files, 18.4 min, 14 Anthropic calls split across document harvest, supporting-file insight, project intelligence, and 2 multi-prompt MEP takeoffs).

These numbers replace earlier estimates from MODEL_COST_ANALYSIS.md v1 (Apr 17 Highline-only A/B) which were ~3–10× too high because the cost model didn't apply prompt-cache reads at 0.10×.

Margin at the current tiers

Implication: Solo and Studio are comfortable at typical medium-folder load. Heavy mega-folder months should use $35/run overage or Firm tier — steer via upload hints rather than surprise bills.

Pipeline recommendation (tied to pricing)

• Default pipeline: Pure Claude (MODEL_PROVIDER=claude, the default when MODEL_PROVIDER is empty). Claude is the only mode that has captured complete schedules across the diverse plan sets we've tested (Highline, DEN47, Villa Matka). Real cost is ~$1.50 per folder for small jobs, ~$3.50 for typical, ~$7 for large — price the product around this. With Anthropic prompt caching active on the system + last-document blocks (5-min TTL), warm-cache jobs recover ~30-40% of input cost; cold-cache jobs (first job in a session) pay the full 1.25× write surcharge — see MODEL_COST_ANALYSIS.md §6 finding 1.
• Middle tier: Hybrid (MODEL_PROVIDER=hybrid). Grok runs Steps 1-3 for quantity and Claude runs Steps 4-5 for forensic + narrative. Estimated COGS ~$0.40/takeoff at lower forensic quality. Reserved as a margin-pressure relief valve; revisit if Pro tier scales beyond the breakeven shapes above.
• Cheap / demo mode: Pure Grok (MODEL_PROVIDER=grok). Cost floor so low it can anchor a "Quick takeoff" feature without financial exposure; quality ceiling is visibly lower (fewer concerns, thinner narrative, weaker fixture recognition on dense schedules), reinforcing the upsell to paid Claude pipeline.

Overage pricing (recommended)

Above-tier usage priced for ~50–65% margin on the median folder shape:

• Medium folder overage: $7–10 per additional folder (~$3.50 COGS)
• Large folder overage: $15–20 per additional folder (~$7 COGS)
• Mega folder overage: $25–35 per additional folder (~$15 COGS)

Or simpler: flat overage $35/completed run (see OVERAGE_RUN_USD in lib/pricing/public-plans.ts).

Cost levers already live (May 2026)

• Anthropic prompt caching (5-min TTL) on system + last-document blocks in every multi-prompt step. Working as designed within a single takeoff job; underused across jobs (Finding 1 in Cost Analysis).
• Provider-aware USD pricing in lib/grok/token-usage.ts estimateUsdFromUsage — every Claude / Grok / Gemini call shows accurate per-call cost in test-results/llm-usage/events.jsonl.
• Cache breakdown (cacheCreationInputTokens, cacheReadInputTokens) separated from promptTokens in usage records so the cost helper applies the right multipliers (1.25× vs 0.10×) instead of silently treating cache traffic as full-price input.

Cost levers to investigate next

• Anthropic 1-hour cache TTL (cache_control: { type: 'ephemeral', ttl: '1h' }). Largest single win — estimated 15–25% folder cost reduction by cross-job cache reads inside a folder pipeline.
• Bundle dispatcher tightening — stop sending non-plan PDFs (RFP, geotech, supplier bid tabs) into the multi-prompt MEP path.
• Cross-bundle legend reuse — Step 1 (legend extraction) is identical for electrical / lighting / power bundles in the same plan set; reusing it saves ~$0.10 per duplicate-legend bundle.

---

7. Implementation Notes

• Stripe (or similar) for subscriptions
• Usage metering: Track takeoffs per user per billing period (one takeoff = one folder run; see §6a for cost shape per folder)
• Overage: $35 per completed run above allowance (see OVERAGE_RUN_USD)
• Annual discount: 15% — pinned to ANNUAL_DISCOUNT_PERCENT in lib/pricing/public-plans.ts

---

8. Next Steps

1. Validate with users: Survey 5–10 estimators on willingness-to-pay for folder-native takeoff vs manual days.
2. Internal cost model: Track COGS at Solo ($179) and Studio ($149/seat) run allowances.
3. Pilot: 2–3 early adopters; gather conversion and seat expansion data.
4. Launch: Solo + Studio self-serve; Firm/Enterprise via contact.
5. Revisit: Every 3–6 months; watch conversion, churn, and competitor moves.

---

9. Spot-check playbook (when “price points must be spot on”)

Use this when refreshing strategy—not a substitute for live Stripe and usage data.

Quantitative pulls

• Runs per subscriber: Distribution of completed takeoff runs per user per billing month (tie to entitlement enforcement). Heavy tails on Pro drive COGS faster than midpoint pricing assumes (§6a margin table).
• COGS sanity: Replay or sample test-results/llm-usage/events.jsonl (or production-equivalent aggregates) by folder shape—small vs medium vs large vs mega—after any pipeline or model-price change (MODEL_COST_ANALYSIS.md).
• Realized revenue: Stripe MRR / ARPA by price ID; reconcile published ranges (lib/pricing/public-plans.ts) with fixed Stripe prices (every Checkout SKU is one amount, not a range).

Positioning pulls

• WTP cadence: 8–12 short interviews per year, same anchor script: hours saved × loaded rate vs. subscription line item (§5b).
• Competitive rescan: Semiannual pass on subs in §3a table; confirm Solo still clears premium point-tool anchoring vs Countfire/PlanSwift and stays far below Trimble-class enterprise stacks.

Decisions that need explicit ownership

Canvas workbook: Structured summary for reviews lives in Cursor Canvas pricing-strategy-review.canvas.tsx (project canvases folder)—update when tiers or §6 cost assumptions change materially.