Price the outcomes, cap the usage, and earn trust with predictability. For early‑stage AI SaaS, package by workflow and autonomy level, meter “actions” (not tokens), and include hard caps with auto‑fallback to avoid bill shock. Offer a free or low‑friction entry, prove lift with decision logs and holdouts, then expand via outcome‑linked add‑ons. Track cost per successful action to ensure margins improve as adoption grows.
Step‑by‑step pricing blueprint
- Define the unit of value
- Pick 1–3 “actions” that map to real work: e.g., ticket resolved, invoice matched, meeting booked, job scheduled, record updated within policy.
- Define “successful action” (sticks without reversal) for premium/outcome components.
- Document exclusions and guardrails (caps, approvals, change windows).
- Package the product
- Platform + workflow modules: core governance/connectors plus job‑to‑be‑done add‑ons (support automation, AP exceptions, onboarding copilot, pricing guardrails).
- Autonomy tiers:
- Tier 0: Suggest
- Tier 1: One‑click apply (preview + undo)
- Tier 2: Unattended for low‑risk, reversible steps (opt‑in)
- Deployment options as add‑ons later (VPC/private inference, BYO‑key) when moving up‑market.
- Choose meters customers understand
- Primary: actions executed (schema‑validated tool‑calls).
- Premium KPI: successful actions (no rollback/reversal).
- Secondary (technical): partner API calls or GPU‑seconds for heavy inference.
- Avoid: raw tokens/messages or vague “AI units”.
- Make pricing predictable
- Pooled quotas and hard caps per plan; auto‑fallback to suggest‑only when caps hit.
- Budget controls: alerts at 60/80/100%, in‑product usage dashboard (actions, router mix, cache hit), pause/resume.
- Seasonal “burst packs” for temporary capacity without plan changes.
- Tie price to proof
- Bake in decision SLOs (e.g., p95 ≤ 2 s, JSON/action validity ≥ 99%) with credits for sustained breaches.
- Keep holdouts/ghost offers to attribute lift; use weekly value recaps: actions, success rate, reversals, incremental outcomes, spend vs cap.
Recommended plan templates (copy‑ready)
- Starter (PLG/self‑serve)
- $99–$299/month base + 5–10 seats
- Includes 3k–5k actions/month, standard connectors, p95 ≤ 2 s
- Hard cap with auto‑fallback to suggest‑only; add‑on action bundles
- Ideal for proving value quickly without sales friction
- Growth (sales‑assisted SMB/MM)
- $1k–$3k/month base + pooled 25k–50k actions + 25–50 seats
- 1–2 premium connectors, Tier 1 autonomy (one‑click) enabled
- Optional outcome bonus (e.g., $Y per verified save/upgrade beyond baseline)
- SLO credits; seasonal burst packs
- Enterprise (once ready)
- Annual commit with reserved capacity; VPC/private inference/BYO‑key add‑ons
- Custom approval matrices; audit exports; Tier 2 autonomy for approved low‑risk flows
- Price protection and multi‑region residency options
Note: Start lower on actions, not on base fee—anchor value in outcomes and predictability; scale with bundles.
Add‑ons that monetize without chaos
- Data/connectors: premium third‑party connectors, real‑time streams, historical backfills
- Governance and compliance: audit exports, model/prompt registry access, fairness dashboards
- Performance: dedicated throughput, higher p95/p99 SLOs
- Services: pilot setup, golden evals, integration development
Free tier and trial design
- Free tier (optional)
- Strict monthly action cap (e.g., 300–500), watermarking/logging, standard connectors only
- Clear upgrade paths when near cap; safe sandbox data or limited scopes
- Trials
- 14–30 days with guided setup; carry configuration into paid
- Pre‑agreed success metric (e.g., 200 successful actions or 15% lift vs holdout)
- Weekly recap emails; one‑click purchase with prorated credits
Pricing guardrails and policies
- Fair‑use: rate limits, quiet hours, variant caps to prevent abuse
- Autonomy gating: unlock Tier 1 only after acceptance/reversal KPIs pass; Tier 2 by exception
- Change management: simulate impact and show rollback plans before executing costly actions
- Refunds/credits: tied to SLO breaches or mis‑metered actions, not “model errors” broadly
Instrumentation to keep margins healthy
- Dashboards: cost per successful action, router mix (small vs large), cache hit, JSON/action validity, reversal rate, p95/p99 latency
- Budgets: per‑plan/tenant alerts; separate interactive vs batch lanes; pre‑warm during launches
- Cost controls: small‑first routing, caching embeddings/snippets, cap variants, batch heavy tasks
Common pitfalls (and fixes)
- Bill shock from token‑based pricing
- Translate to actions; include pooled quotas and hard caps with auto‑fallback
- Selling “AI” instead of workflow outcomes
- Package by job and autonomy; show decision logs and incremental lift
- Over‑automation too early
- Start suggest → one‑click; require approvals for sensitive steps; track reversal rate as a KPI
- Complex menus
- Keep 2–3 plans + clear add‑ons; document meters and caps in plain language
- Free trials that don’t convert
- Connect to real systems with safe scopes; define success upfront; ship weekly proof
30‑day quick start
- Week 1: Define action and successful action; set SLOs and caps; draft Starter/Growth plans
- Week 2: Instrument dashboards (CPSA, router mix, cache hit, reversals); add budget UI and alerts
- Week 3: Launch PLG plan + guided trial; run two pilots with holdouts and weekly value recaps
- Week 4: Tune quotas and prices from real usage; introduce action bundles and an outcome bonus option
Bottom line: For startups, the winning pricing play is simple, predictable, and proof‑backed. Package by workflow and autonomy, meter actions with hard caps, add outcome‑linked bonuses where attribution is clean, and expose budgets and SLOs. Keep cost per successful action trending down with routing and caching, and expansions will follow from demonstrated, auditable value.