Skip to content
PALM SOUQ

For B2B AI SaaS

Protect your 80% SaaS gross margins in an AI-native world.

Flat-rate plans assume the next user costs almost nothing. With LLM features, your heaviest tenants can cost several times what they pay. PALM SOUQ meters every tenant and enforces the allowance your pricing assumes.

Defaults from the worked example: 500 customers x $99 = $49,500 MRR, 25 power users at $400 raw spend.
tenant-policy / power-users500 tenants
$99/mo
25
$400/mo
$35/mo
Per-tenant hard cap
Margin per power user, uncapped
-$301
Margin per power user, now
$64
Recovered per month
$9,125

25 x ($400 - $35) = $9,125

The unit economics crisis

How one busy tenant turns a $50 plan into a $150 monthly loss.

A support-assistant conversation that sends 30k tokens of context and returns 1.5k tokens costs $0.072 on GPT-4.1 at list price. That is invisible for a casual user.

A tenant that runs 2,800 of those conversations a month, about 93 a day, costs $201.60. On a $50 plan, that tenant loses you -$151.60 every month, and nothing in a flat-rate invoice tells you which tenant it is.

Multiply that by a handful of power users and the plan that looked like 80% gross margin in the spreadsheet runs well under 50% in the bank account.

tenant acct_7Q2 / Octobermargin negative
Cost per conversation
30k x $2.00/M + 1.5k x $8.00/M = $0.072
Conversations this month
2,800
LLM COGS
2,800 x $0.072 = $201.60
Plan revenue
$50.00
Tenant margin
-$151.60 / month

Hybrid pricing

Keep the subscription. Meter the tokens behind it.

The plans that hold margin combine a base subscription with an included token allowance and a clear rule for what happens at the limit.

ModelAt the limitBest for
Allowance + hard capBlock or upgrade promptSelf-serve plans with a clear upgrade path
Allowance + fallbackSmaller model, same featureProducts where availability matters more than peak quality
Allowance + metered overageUsage billed at cost plus markupTeam and enterprise plans with procurement-friendly invoices
Prepaid creditsTop-up or pauseBursty workloads and agencies

Fallback is cheaper than it sounds: GPT-4.1 mini lists at $0.40/M input and $1.60/M output against $2.00 and $8.00 for GPT-4.1, one fifth of the price for the same call. See model fallbacks and billing sync.

policies/team.tg
tier "team" {
  price_usd        = 99
  allowance_usd    = 35          # token spend included in the plan
  soft_cap         = 0.80        # in-app banner + webhook
  fallback_at      = 0.90        # gpt-4.1 -> gpt-4.1-mini
  fallback_chain   = ["gpt-4.1", "gpt-4.1-mini"]
  at_100           = "overage"   # or "block" | "notify_upgrade"
  overage          = { sync = "stripe", markup = 0.30 }
}
  1. allowance_usd ties the plan to a margin floor: $99 x (1 - 0.65) = $34.65.
  2. fallback_at keeps the feature working near the limit.
  3. overage turns the hard cap into revenue on the next invoice.

Margin defense

A B2B AI document platform went from 44% to 78% gross margin.

Extraction moved to a small model, document templates were cached, and the heaviest accounts moved to metered overage. The arithmetic, month before versus Day 90:

Revenue

$120,000

$140,000

includes $20,000 of overage

LLM COGS

$58,000

$21,600

small-model extraction, cached templates

Other COGS

$9,200

$9,200

unchanged

Gross margin

44.0%

78.0%

(revenue - COGS) / revenue

($140,000 - $21,600 - $9,200) / $140,000 = 78.0%

“Connecting PALM SOUQ to Stripe Billing let us launch usage-based token overages in an afternoon.”
Chief Product Officer, Seed-stage AI writing tool
Read four customer stories, with the numbers

Questions, answered

Will caps make my product feel broken?

Not if the cap does something useful. Most teams use a soft cap for a warning, a fallback to a smaller model near the limit, and an overage or upgrade prompt at 100% instead of a hard error.

How do I pick the allowance per plan?

Start from the gross margin you want. For a $99 plan and a 65% LLM margin floor, the allowance is $99 x (1 - 0.65) = $34.65, so about $35. PALM SOUQ shows the last 30 days of spend per tenant so you can see how many tenants that allowance would affect.

Does PALM SOUQ charge my customers?

No. PALM SOUQ sends metered usage to your billing platform (Stripe, Metronome, Orb or Lago) and your existing invoices carry the overage line.

What if a tenant uses several products or features?

Add X-PalmSouq-Feature to each call. Budgets can apply to the tenant as a whole or to a feature within it.

Find your underwater tenants this week.

Route traffic through PALM SOUQ, pass X-PalmSouq-Tenant, and the first margin table is ready within an hour of traffic.

14-day free production trial · 100k requests included · no card required