productivityatlas.

Research / Chapter 10

The price beyond the subscription

Seats, credits, review time and the cost of switching.

Research edition · 14 September 2026

The report’s observations, prices and forecasts are a dated snapshot. Examples of savings are estimates unless explicitly identified as study results. This is not a fresh verification of every claim.

Business-model comparison

Model Customer benefit Vendor economics Buyer risk
Freemium Low-friction acquisition and trial Paid conversion and usage learning Limits can make the free tier unsuitable for real work
Monthly / annual subscription Predictable budget and recurring revenue Better planning and retention; annual cash upfront Annual commitment before proving value
Per-seat Simple procurement and expansion Aligns revenue with users, not necessarily usage Paying for inactive or light users
Usage-based Aligns price with activity or value Captures heavy use and protects gross margin Variable bills, retries, and poor cost predictability
Credit-based Bundles different models and features Lets vendor meter expensive inference Credits obscure actual work cost and may expire
Per-resolution / outcome Ties payment to customer-service result High monetization when automation works Ambiguous “resolution,” escalation gaming, and unpredictable volume cost
Enterprise license Governance, support, procurement, and customization Larger contracts and lower churn Negotiated prices obscure comparison and implementation cost
API / token pricing Flexible building block Revenue scales with developer usage Engineering, monitoring, retries, and model migration become buyer’s burden
Add-ons / marketplace Modular expansion Monetizes specialized features and distribution Stack sprawl, security review, and integration fees
Advertising / data monetization Low or no user price Monetizes attention or aggregated data Incentives may conflict with privacy and trust
White-label Faster product launch for the buyer Vendor captures infrastructure and model margin Dependency, differentiation limits, and data ownership terms

Inference cost is falling rapidly for a fixed level of benchmark performance. Stanford’s 2025 AI Index reported that the cost of querying a model at roughly GPT-3.5-level MMLU performance fell from $20 per million tokens in late 2022 to $0.07 by October 2024, a more than 280-fold reduction.[43] That does not make every AI feature cheap: long context, premium reasoning, video generation, agent retries, storage, retrieval, transcription, support, compliance, and distribution all add cost. The product pages above show vendors increasingly combining subscriptions with credits, limits, or outcome pricing.

Illustrative monthly budgets

These are planning ranges, not quotes. They exclude tax, implementation, training, procurement, custom development, and overages.

Scenario Example stack assumption Illustrative monthly budget
Individual One general assistant ($0–$20), one task/knowledge tool ($0–$10), optional Readwise or creative tool ($10–$30) $0–$60
Freelancer General assistant, Todoist or Motion, meeting notes, Canva/Descript as needed $50–$130
Small business, 10 people One suite AI plan for all staff, 2–4 specialist seats, light automation, shared storage $250–$700
Startup, 25 people One governed workspace, coding seats for developers, meeting capture for customer-facing staff, automation budget $800–$2,500
Remote team, 50 people Slack/Teams plus work-management AI, selected meeting seats, coding/creative seats, metered automation $1,800–$5,000
Enterprise department, 100 people Enterprise suite or search platform, 20–60 specialist/agent seats, security and implementation allocation $6,000–$25,000+

Budget formula

Monthly cost = fixed platform fees
             + (assigned seats × seat price)
             + expected AI credits / outcomes / tokens
             + storage, recording, and connector fees
             + implementation and review labor
             + contingency for overages and migration

For a pilot, budget human review as a real cost. A tool that costs $20/month but adds 10 hours of checking is not a $20 productivity solution.