# Nokumi pricing benchmark and cost-sharing decision

Checked 19 October 2026. Decision: retain prepaid metered fees, recover permitted channel charges from the paying customer, quote attributable/shared costs before usage, and fund bounded free allowances from positive earned contribution. Do not promise that volume or micro-SaaS directory inclusion proves profit. Owner: founder; pilot constraints remain approximately Rp10m and 1–3 developers using AI. All Nokumi prices, costs, mix and subsidy parameters below are [A] proposals, initial confidence 40%.

## Dataset and selection

[F] Read the [public research dashboard](https://saas-scraper-research.pages.dev/)'s existing public `dataset_view`, dataset `micro`: 232 rows, 100 selected, with 70 eligible, 27 excluded and three pending review. Selected six eligible cases for overlap with API aggregation, MCP, prepaid credits, metered developer utilities, builder bundles or creator purchases. This is purposive, not a random sample or exhaustive competitive landscape. [Audit](micro-pricing-audit.json) and [linked evidence CSV](micro-pricing-benchmark.csv) preserve slugs, prices, source URLs, dates and limits. No raw financial records or private credentials were exported. The dataset label is not an audit of each company's current size/revenue. Its Indonesia-status field is not founder geography.

## Benchmark results

| Selected micro-SaaS | Current published architecture and price [F] | Implication for Nokumi [I] |
| --- | --- | --- |
| [AnyAPI](https://getanyapi.com/) | Unified API/MCP and USD balance, no required subscription; approximately160 free requests. Web-scrape example USD0.70/1,000 calls. | Closest architecture analogue. One key, credits and MCP are already available. Differentiate through Indonesian app workflows and recovery. |
| [BillionVerify](https://billionverify.com/pricing) | PAYG from USD0.001/check; selected20k/month-equivalent annual option (240k/year), USD96 paid annually.20 login-day free credits, up to600/month. | Distinguish annual commitment from monthly payment and verification from sending email. |
| [WhoisJSONAPI](https://whoisjsonapi.com/) | Basic USD49/month for100k credits; Premium USD129/500k; Enterprise USD299/2m.14-day trial. | Low usage still bears minimum-plan costs. DNS access needs higher tiers. |
| [Databuddy](https://www.databuddy.cc/pricing) | 10k events free; Hobby USD9.99/month incl30k; Pro USD49.99 incl1m. Initial paid overage USD0.035/1,000 extra events. | Free caps and graduated rates are familiar. Analytics events are not interchangeable with messages or MAU. |
| [ZocialComment](https://zocialcomment.com/pricing) | First100 comments/post free, three free exports/day. PAYG USD1/100 comments, USD3 minimum; USD9/39 three-day passes with per-post caps. | One-off buying can reduce commitment; “unlimited” still needs duration/output boundaries. |
| [AIWA](https://aiwa.codes/) | Solo USD20/month incl200 building credits; Agency USD149/month incl1,000. Agency extra credits USD0.10, top-ups USD5–2,000. | A builder can bundle auth/backend/integration generation. Nokumi must prove extra runtime-maintenance value. |

[E] In these six intentionally chosen cases, three offer a nonrecurring purchase path (AnyAPI, BillionVerify, ZocialComment); four have published monthly plans (BillionVerify, WhoisJSONAPI, Databuddy, AIWA). These overlap. Five explicitly offer an allowance or trial; four explicitly show MCP/AI-client integration in reviewed material. These are counts within six selected cases, not national prevalence, independent adoption or product-quality measurements. “Not established” does not mean unavailable.

[E] Normalizing only within the same provider/unit: AnyAPI's web-scrape example is USD0.0007/request, or USD7 for10k calls before other charges. WhoisJSONAPI Basic at full100k capacity averages USD0.00049/credit, but at10k usage of the same plan costs USD0.0049/used credit. These are different functions; no comparison to Nokumi's IDR/message/MAU fees or “X% cheaper” claim is valid. No FX rate or Indonesian demand ranking is manufactured.

AnyAPI's FAQ notes some failed wallet-funded requests may have at-cost processing fees, while another section says failed attempts cost zero. Preserve that scope conflict; do not copy a blanket free-failure promise. BillionVerify's USD0.00014 headline depends on volume/term conditions and is not the selected20k/month-equivalent annual option. ZocialComment's differing traction totals are excluded. No API calls, quality/reliability tests, account signup or purchases were performed.

## Who bears each cost?

| Cost | Recommended payer/allocation [A] | Guardrail |
| --- | --- | --- |
| Bank/VA or other permitted top-up channel | Customer pays the actual permitted channel fee and applicable fee tax, quoted before payment. | Fixed+percentage fees require gross-up; do not deduct an undisclosed fee after crediting. |
| QRIS MDR | Nokumi remains merchant payer; ordinary service prices fund legitimate operating costs. | No QRIS-specific surcharge, reduced-credit workaround or relabeled QRIS fee. Provider/legal review of final price architecture. |
| Dedicated Supabase compute/storage/add-ons | The relevant app/customer; managed quote includes attributable cost plus the proposed5% once, or BYO billed directly. | No hidden cross-subsidy for expensive dedicated runtime. No duplicate recovery in common costs. |
| WhatsApp/email/API consumption, tax and FX | Attributed to consuming customer; provider bill plus approved allocation and Nokumi fee. | Actual billable provider failures/minimums are disclosed, even when Nokumi success fee is zero. |
| Common vendor minimums, observability, billing/control plane | Allocate genuine common costs across forecast weighted usage, publish the quote/cap before use. | Use measured cost drivers, not raw incomparable WA/email/MAU units. Each invoice line allocated once; no surprise retrospective shortfall bills. |
| Free-tier control-plane operations | A bounded reserve from realized positive contribution, plus any explicitly budgeted launch seed. | No unused customer credits, provider pass-through or assumed future heavy customers counted as revenue. |
| Human setup/support/custom maintenance | Separately scoped quote. | FeeRp15 does not fund unlimited manual help. |

[F] [BI guidance](https://www.bi.go.id/id/publikasi/ruang-media/cerita-bi/Pages/mdr-qris.aspx) says merchant bears MDR. The [17 August2026 BI release](https://www.bi.go.id/id/publikasi/ruang-media/news-release/Pages/sp_2815926.aspx) extends0% MDR to transactions up toRp100k across merchant categories, effective1 October2026; UMI threshold remainsRp500k. This is MDR, not a guarantee every gateway's entire service is free. [Midtrans pricing](https://midtrans.com/pricing) lists QRIS0.7% and bank/VA Rp4k, before VAT/category differences. These sources do not resolve our contract/classification; obtain the applicable quote. Do not replace BI rules with a generic gateway page.

[F] [Supabase billing](https://supabase.com/docs/guides/platform/billing-on-supabase) applies organization quotas across projects and charges project compute independently. [Resend pricing](https://resend.com/pricing) has plan/domain limits and paid overage buckets; [Cloudflare Workers pricing](https://developers.cloudflare.com/workers/platform/pricing/) has base/usage limits. Include those dimensions rather than assume one shared unlimited account. Agreements for managed/delegated use remain an activation gate.

## Quote arithmetic and worked examples

[A] For a given homogeneous meter, quote **D=(P+F+V+S)/(1−r)**. P=attributable landed provider cost, F=Nokumi unit fee, V=variable shared operating cost, S=allocated common provider cost, r=non-recovered collection fraction. D is the final usage debit, the same service quote regardless of top-up method. Show P, shared-cost allocation D−P−F, F and the total before consent. This is not permission to disguise a QRIS surcharge. Apply no second common markup where a cost has already been recovered in a dedicated quote.

[A] Illustration:60% of credit funding through QRIS at a deliberately conservative0.8% all-in cost assumption,40% through allowed channels whose costs are fully recovered at top-up, so r=0.48%. The0.8% is not BI's MDR or a contract quote; model0%/actual category rates when confirmed. Shared vendor minimumRp500k/month over forecast1m comparable paid units gives S=Rp0.50/unit; V=Rp2/message; P=Rp500/message is hypothetical. Quotes lock for approved usage, and later quotes may change; actual shortfalls affect Nokumi's margin, not old user invoices.

[E] For10k utility messages: P=Rp5m, F=Rp150k, V=Rp20k, S=Rp5k, D=Rp5,199,960 after once-per-statement rounding. Shared recoveryRp49,960 covers modeled operations, minimum allocation and collectionRp24,959.808. Contribution≈Rp150k before human support, fixed payroll, CAC and Nokumi taxes. If this were the only10k platform units and the wholeRp500k common bill remained, contribution would be **−Rp345k**. Forecast scale is not achieved scale.

[E] For1m utility messages, same costs/bands: P=Rp500m, F=Rp10.23m, V=Rp2m, S=Rp500k, D=Rp515,202,975. Contribution≈Rp10.23m before reserve/support/fixed costs. Actual unrecovered collection rising to3% while that quote stays locked produces **−Rp2,753,114.25**. Do not reprice historical usage to repair the loss. Use `scripts/pricing_math.py` and public `pricing.json` to inspect assumptions.

[E] Bank example, before fee tax: desired creditRp100k + permitted fixed channel costRp4k → payRp104k. With percentage costs, gross payment G=(credit+fixed×(1+fee-tax))/(1−percentage×(1+fee-tax)); round only the final quote. No live all-in bank fee is asserted. QRIS payRp20k → creditRp20k, no extra MDR line or credit haircut.

## Heavy-to-light subsidy budget

[A] Earmark up to20% of **realized positive contribution after all attributable, allocated common and unrecovered collection costs** for free-tier operations. Assume a measured marginal free-workspace ceilingRp1k/month for this illustration, excluding provider charges, dedicated runtime and human support. This assumption is a ceiling to test, not the actual cost of1,000MAU+500emails or a promise to fund thousands of users.

[E] At the fully recovered1m-message case, reserveRp2.046m funds at most2,046 free-workspace-months atRp1k each; remainingRp8.184m is still before payroll/support/CAC/taxes. At10k within an already funded1m-unit pool, reserveRp30k could fund30; with only10k platform volume, negative contribution funds **zero**. If marginal free cost isRp5k, the1m case funds409, not2,046. Unused deposits do not increase capacity. Low-cost users should retain explicit fair-use controls and existing approved allowances; use admission limits/future offers rather than retroactive fees when budget runs out.

[A] Propose Rp200k launch seed from the existing Rp3.5m unallocated pilot contingency, not extra authorized spend or customer money. It is not spent by this task. Owner must approve an operating budget before activation. Heavy users are currently hypothetical; the seed exposes the launch chicken-and-egg problem rather than pretending it is solved.

The strongest objection: recovering more cost lines can destroy the simplicity that attracts beginners. Recommendation still holds if the customer sees **one upfront all-in quote with an optional breakdown**, the same service price across payment methods, and a hard spend cap. Compare that against BYO and a builder bundle; do not force prepaid users through a spreadsheet. If customers only want cheaper access to free provider tiers, this model adds little value.

## Risks, reversal conditions and next actions

- Risk: allocation forecasts overstate volume (high likelihood/high impact). Measure actual minimums/utilization weekly; pause managed offers that cannot cover shared bills at observed volume.
- Risk: top-up or shared charges are disallowed/confusing (medium likelihood/high impact). Founder checks provider agreements/BI treatment and tests quote comprehension before real payments; stay sandbox/BYO if unresolved.
- Reverse recommendation if no approved collection/cost-recovery arrangement exists: offer permitted channels/BYO rather than bypass rules. If two actual cohorts contribute negatively after full common costs, stop discounts, revise quotes or narrow scope. If0/8 qualified offers consume paid units, revisit segment/value before scaling ads.
- Founder, first pilot week: obtain quotes/permissions, map every invoice cost to exactly one payer or explicit subsidy pool, verify actual QRIS category and tax. Success:100% of invoice lines assigned without duplicate recovery.
- Developer, first two pilot weeks: sandbox quote locking, tier boundaries, gross-up, retries, reversals and invoice reconciliation. Success: no duplicate debit and bank net covers promised credits within disclosed rounding.
- Founder, day30–60: measure the actual contribution/subsidy reserve and free-account marginal cost before expanding free admissions. Success: funded grants cover measured cost with no reliance on deposits. Still open: WTP, support load, FX, provider resale rights, actual payment mix and cost-weighted allocation drivers.