Empire Teas CRM — Running Cost Annexe (for the Statement of Work)
Prepared by Altrium for Empire Teas (Pvt) Ltd Purpose: the re-derived monthly running cost of the estate actually being built — an estimate of the AWS bill Empire Teas receives directly (see the billing-model note below). The proposal’s §A.2.2 and §A.8 stand as issued and are not edited; this annexe restates them on the current basis. Basis date: 21 August 2026 · AWS and AI provider list pricing, before any committed-use discount
Revised 18 November 2026: one test (UAT) environment now serves both acceptance rounds. It is kept, with its data, for about 11 weeks, from Phase 1’s round to the end of Phase 2’s, and may be paused between the rounds. It costs up to about $100 to $155 in total (Section 4). The monthly figures do not change.
Revised 16 November 2026: planning a show in the trade-show capture application — researching and ranking every exhibitor on the show’s plan — has its own line in Section 6, about $7.50 for a 500-exhibitor show; Section 6’s web-search line now counts those searches, and Section 5 notes a month with a planned show.
Billing model (revised 29 September 2026): the AWS account is Empire Teas’ own, held by Empire Teas IT and paid by Empire Teas’ card, so AWS bills Empire Teas directly for AWS and Bedrock (AI) usage, on consumption. Altrium builds the environments inside that account and bills only the USD 360/month application support fee under SOW ET-001. The figures in this annexe are therefore Altrium’s estimate of Empire Teas’ monthly AWS bill, and they vary with use. Replaced: from 13 to 29 September this note described SOW ET-002, under which Altrium would have owned the accounts and billed a fixed USD 650/month; Empire Teas declined it (
aws-sow.md, marked superseded).
1. Why this annexe exists
The proposal was issued on 5 August 2026 and commercially accepted on 16 August. Three things have happened since that change the shape of what §A.2.2 priced — not the scope, and not the price of the build:
- Both applications are now served as static files from S3 and CloudFront, rather than the frontend running as a container task. An internal application with no public search requirement does not need a server to hand out its own pages.
- Client UAT now runs in its own temporary environment, created for Phase 1’s acceptance round, kept until Phase 2’s round ends, and then removed (Section 4). The proposal assumed the development environment would double up. Keeping them separate means an acceptance round never competes with the build, and the development environment is never reset mid-phase to make a round possible.
- The AWS region is settled: US East. §A.1 left this open and warned that a Singapore or European region would run 10% to 25% above the basis used. US East is at that basis, so the warned uplift does not apply. Users reach the applications through CloudFront’s global edge network, so the choice does not slow the interface down.
The proposal’s own header also records an arithmetic gap in §A.2.2’s development sub-total, to be re-derived here before any figure is quoted onward. This annexe does that.
Every change below moves the figure down. Nothing here is a change request, and nothing in it alters the one-time build price or the monthly support fee.
2. Basis of estimate
Unchanged from the proposal’s §A.1, restated so this annexe stands alone:
| Assumption | Value used |
|---|---|
| Named users / peak concurrent | 15 / 20 |
| Record volume | Low tens of thousands of accounts, contacts, deals, leads, activities |
| Permanent environments | Two — development and production, in separate AWS accounts |
| Acceptance environment | Temporary, created for Phase 1’s UAT round and destroyed after Phase 2’s |
| Production availability | Business-hours, single availability zone; recovery is a restore from backup, not automatic failover |
| AWS region | US East — the basis used, with no regional uplift |
| Pricing basis | AWS and AI provider list pricing, August 2026, before committed-use discounts |
These are engineering estimates, not a quotation. AWS bills on actual consumption, directly to Empire Teas, which owns the account; Bedrock (AI) usage is on the same AWS invoice.
3. The re-derived estate
Rows marked carried are the proposal’s own figures, unchanged. Only the frontend line moves.
| Item | Production | Development | Status |
|---|---|---|---|
| Backend application, including the background worker (Fargate, 0.5 vCPU / 1 GB) | $18 | $6 | carried |
| Frontend application (Fargate, 0.25 vCPU / 0.5 GB) | removed — no longer a container | ||
| Static hosting for both applications (S3 + CloudFront, certificates free) | ~$2 | ~$1 | added, replacing the row above |
Database (RDS PostgreSQL, single-AZ; t4g.small / t4g.micro) |
$31 | $14 | carried |
| Load balancer | $19 | not used | carried |
| S3, SQS, ECR, Route 53, Secrets Manager | $6 | $4 | carried |
| CloudWatch and data transfer | $9 | $4 | carried |
| Threat detection (GuardDuty) | $5 | $5 | carried — not enabled from 9 October 2026 (turned on if Empire Teas asks for it); kept in the total so the figure still holds if it is, and the bill is about $5 lower per environment while it is off |
| Sub-total | ≈ $90 | ≈ $34 |
Total AWS: approximately $124 per month, against the approximately $140 in §A.2.2.
Two things account for the difference of about $16. Roughly $7 is production’s frontend no longer running as a container ($9 removed, $2 of static hosting added). The other $9 is the development environment: §A.2.2 stated its sub-total as approximately $43 while its own rows total $33 — the arithmetic gap the proposal’s header flagged for re-derivation — less the $1 of static hosting added there. Both corrections favour Empire Teas.
Storage grows with use, as §A.2.2 noted: production starts at 50 GB and autoscales, adding roughly $6 per 50 GB per month. At the record volumes above this is not expected to be material in the first two years.
4. The temporary acceptance environment
Updated 18 November 2026: one environment now serves both rounds.
UAT runs in an environment of its own. It is built by the same automation as the permanent environments, so it is the same system rather than an approximation of it.
One environment serves both acceptance rounds. It was created in November 2026 for Phase 1’s round. It is kept, with its data, until Phase 2’s round ends, about the end of January 2027. Then it is removed. Keeping it between the rounds keeps the test data. The records Empire Teas’ testers create are still there for later Phase 1 releases and for Phase 2.
It is sized like the development environment. It costs roughly $9 to $14 a week, based on what the development environment costs to run. That is about $100 to $155 in total if it runs for all of the roughly 11 weeks. It may be paused between the rounds, with its data kept, which costs less. It appears on the AWS invoice in November, December and January. It is not a standing monthly cost: once it is removed, its cost stops.
5. AI usage
Unchanged from the proposal’s §A.3, and confirmed against current published model pricing in the chosen region:
| Line | Figure |
|---|---|
| Expected usage at the planned volumes | approximately $26 to $31 per month |
| Recommended budget | $50 per month |
| Billing alert | $150 |
The models are served through Amazon Bedrock inside the project’s AWS account (Empire Teas’ own), so this appears on Empire Teas’ AWS invoice, on consumption, rather than as a separate vendor bill. Controls are unchanged: an administrator can switch any AI feature off without a deployment, and a hard ceiling turns AI features off rather than allowing an unbounded bill. Every affected screen keeps its manual path, so a switch-off degrades the feature and never blocks the work.
Updated 16 November 2026: a month in which a show is planned adds about $7.50 for a 500-exhibitor plan (Section 6): about $4 of AI, which stays inside the recommended budget, and about $3.50 of web search, which is nil while Exa’s free monthly credit covers it.
6. The trade-show capture application
Confirmed in scope for V1. It runs on the same AWS platform — same account, same region, same security services, same backup design — and introduces no additional hosting vendor.
One change against §A.7: lead notification is inside the capture app only — a notifications list and an on-screen notice, with no email (updated 12 November 2026; until then it was email through Amazon SES, a few dollars a month). The proposal carried the WhatsApp Business Platform at $5 to $25 per month; that is no longer part of the design, so the line and its vendor relationship both fall away.
| Line | Proposal (§A.7) | Now |
|---|---|---|
| Lead notification | WhatsApp Business Platform, $5–25 | Inside the app, $0 |
| Incremental AWS | $25–40 | $20–35 — static hosting rather than a shared container task |
| Research (web-search) data source | still to be selected | No longer a capture-only line — research is now a requirement on every lead in the CRM as well as in the capture app, so it moves into the base below. About 4,600 searches a year (updated 16 November 2026): the CRM’s leads, and for six shows a year about 150 captured leads and a 500-exhibitor plan each (the planning line below). Nil in any month inside Exa’s free monthly credit of about 1,430 searches, which a month with one large show stays within (at most about 1,030); about $32 a year were every search paid; $0 if served by AWS in-account. Section 7’s monthly figure assumes every month stays inside the free credit, as it does at these volumes. Empire Teas chose Exa on 10 November 2026. It runs on its free credit for now, in an account Altrium holds for development; before production, Empire Teas holds its own Exa account — design/ai/lead-research-source-options.md |
| Planning a show — research and ranking of every exhibitor on the plan | not in the proposal | About $7.50 for a 500-exhibitor show, a cost per show that is not in the incremental total below: its AI fits inside Section 5’s AI budget of $50 a month and its web search in Section 7’s web-search line. About $4 of it is AI on Amazon Bedrock, on Empire Teas’ AWS invoice, and about $3.50 is web search, which is nil while Exa’s free monthly credit covers it. A month with one large show stays inside that credit; a second large show in the same month needs the monthly search limit raised, and its searches are paid at $7 per 1,000. Six shows a year is an assumption: Empire Teas has not yet said how many shows it attends or plans |
| Incremental total | $30–65 | ≈ $25 to $45 (unchanged at this rounding when the few dollars of email fell away on 12 November 2026) |
7. Anticipated monthly total
| Line | Indicative monthly cost (USD) |
|---|---|
| AWS infrastructure, both permanent environments | ≈ 124 |
| AI usage, via Amazon Bedrock on the same AWS invoice | ≈ 26–31 · budget 50 |
| Lead-research web search (the CRM’s leads and the capture app) — Exa, chosen by Empire Teas on 10 November 2026, on its free credit for now | ≈ 0–2 — nil while the free credit covers the month |
| Source control and build pipeline (GitHub with Actions) | 0–50 |
| Anticipated total | ≈ 150 to 207 per month |
| Trade-show capture application, once live | ≈ 25–45 on top |
| Anticipated total once the capture application is live | ≈ 175 to 252 per month |
Against the issued §A.8 figures of approximately $165 to $215, and $195 to $280 including the capture application at its proposal estimate. The re-derived figure is lower at both ends.
These are estimates of the cloud and AI costs on Empire Teas’ own AWS bill, on consumption. (Until 29 September 2026 this line said they were Altrium’s costs behind a fixed USD 650/month fee under SOW ET-002, which Empire Teas declined.)
Excluded, as in §A.8: Empire Teas’ existing Microsoft 365 licences; the domain registrar fee; and any hardening option selected below.
8. Optional hardening — unchanged and still deferred
Available at any time without rework, exactly as §A.2.3 set out:
| Option | What it adds | Monthly cost |
|---|---|---|
| Multi-availability-zone database | Automatic failover in under five minutes instead of a one-to-two-hour restore | +$30 |
| AWS WAF | Filters malicious web traffic before it reaches the application | +$15 |
| Cross-region backup copy | Protects the backups against the loss of an entire region | +$8 |
| AWS Config | Records configuration over time and alerts on drift | +$12 |
Altrium’s recommendation is unchanged: run the baseline from go-live, and add the multi-availability-zone database and the WAF — about $45 per month combined — once the system is carrying real business. On the re-derived baseline that steady state is approximately $195 to $250 per month, or $220 to $295 with the capture application live.
Against all of it, the annual Pipedrive licence is removed once Pipedrive is decommissioned.
9. How this is confirmed
- The figures above are Altrium’s engineering estimates on the assumptions in Section 2. AWS bills on consumption; the first full month after go-live is the first actual figure.
- Cost visibility is built in rather than reviewed after the fact: an AWS budget with the alert thresholds in Section 5, and per-feature AI switches an administrator controls.
- If a figure in this annexe proves materially wrong in the first months of operation, Altrium reports it with the actual invoice line rather than waiting to be asked.