Engagement Hub — AWS Cost Model

Conservative estimates · ap-southeast-2 (Sydney) · Jun 2026 · Real production billing + 139 AI token samples (P90)

—

Infra cost / practice

—

AI cost / practice

—

Total cost / practice

—

Gross margin

Pricing Model

Monthly fees

Practice seat$150

Advisor seat$30

Assistant seat$15

Client overage (each)$5

Team per practice

Advisors2

Assistants2

Clients (100 incl. free)100

AI cost included in price

AI Job Volumes (jobs / month)

Unit Economics — 1 Practice

Practice count: 1

Scale Projection

Revenue and cost trajectory as practices grow. Fixed overhead ($10.01) amortises rapidly — margin improves at scale.

Revenue vs AWS cost by practice count

Gross margin % by practice count

Practices	Revenue/mo	Infra cost	AI cost	Total cost	Profit	Margin

Infrastructure Cost Structure

May 2026 actuals · ~1.5 active practices · ap-southeast-2

Cost breakdown by service (May 2026)

Service	May cost	Type	Per practice
Amplify (hosting)	$4.41	Fixed	—
Domain ($48/yr ÷ 12)	$4.00	Fixed (annual)	—
S3 (storage + requests)	$2.47	Variable	$1.65
Route 53 + ECR	$1.60	Fixed	—
SQS	$1.21	Variable	$0.81
AppSync	$0.26	Variable	$0.17
DynamoDB	$0.15	Variable	$0.10
Lambda	$0.00	Free tier	—
Cognito / CW / SNS / KMS	$0.00	Free tier	—

Variable cost per practice: $3.00/mo (conservative; actual $2.79).
Fixed overhead: $10.01/mo shared — amortises rapidly at scale.

AI Feature Costs

Adjust monthly job volumes to model usage scenarios. Token data from production logs (P90 conservative — 139 samples).

Total AI cost / practice / month —

Actual AWS Billing — Validation

Bedrock vs Infrastructure spend (Mar–May 2026)

Month	Bedrock	Infra (AU)	Total AU	Note
Mar 2026	$13.61	$5.59	$75.16	Incl. $48 domain + dev costs
Apr 2026	$36.31	$7.35	$48.04	—
May 2026	$29.20	$10.16	$43.30	—

March anomaly: $75.16 total includes a $48 one-time annual domain charge and ~$13.50 in US East dev/test costs. Normalised AU infra is $5.59. Infra trend Mar→May: $5.59 → $7.35 → $10.16 reflects S3 accumulation.

Free Tier Scale Thresholds

Lambda, Cognito, CloudWatch, SNS, KMS all billed $0.00 in May 2026. Current usage at ~1.5 practices.

Service	May usage	Free limit	Breaks free tier at
Lambda	23,907 GB-Sec (6%)	400,000 GB-Sec/mo	~25 practices
Cognito	898 MAU	50,000 MAU	~83 practices
CloudWatch Logs	0.57 GB-Mo	5 GB/month	~13 practices
SNS	74 notifications	1,000/month	~20 practices

Cost Optimisation Opportunities

Prompt caching — Compliance Coach ~40% AI saving

165K input tokens per call with zero caching. System prompt is ~70K constant tokens across all calls.

Scenario	Cost / call
Current (no cache)	—
With prompt caching	—
Saving per call	—

Prompt caching — AI Reports ~40% AI saving

134K input tokens per call with a large fixed template. Caching could save 40–50% on repeat calls within the TTL window.

Scenario	Cost / call
Current (no cache)	—
With prompt caching	—
Saving per call	—

Model tiering — extraction tasks ~80% saving on sub-task

Action item extraction runs on Sonnet at $3.30/M input. Claude Haiku 3.5 costs $0.80/M input — same quality for short extraction tasks.

Model	Cost / extraction
Sonnet 4.5 (current)	—
Haiku 3.5	—
Saving per call	—

S3 lifecycle policy Ongoing cost risk

45 GB stored in S3 as of May 2026 (~30 GB/practice) with no expiry rule. AI outputs (PDFs, transcripts, presentations) accumulate indefinitely — S3 nearly doubled Mar→May. Recommend: transition to S3-IA after 90 days, Glacier after 365 days. Estimated saving: 40–60% of storage costs at steady state.

Data Confidence

Cost item	Source	Confidence	Note
Infrastructure costs	May 2026 AWS invoice	High	Real billing, ap-southeast-2
Bedrock pricing	AWS console, Jun 2026	High	Confirmed Sonnet 4.5 & 4.6 rates
AI token usage	139 production log samples	High	P90 across 5 features
INFRA_VARIABLE ($3.00/mo)	May actuals ÷ 1.5 practices	Medium	7% buffer above actual $2.79
S3 storage model	45 GB observed, May 2026	Medium	Cumulative — no lifecycle policy yet
Scale projections	Linear extrapolation	Medium	Lambda free tier expires ~25 practices