Numa OpenData services: inbox telemetry, SMS threads, and POS-adjacent order payloads

Feature modules we implement against Numa-style backends

Each module below names a concrete dataset or control plane, then states a single downstream action your team can take once the interface exists.

Core benefits for operators who treat messaging as operational data

Numa’s product surface spans restaurants, multi-site retailers, and—through NumberAI’s automotive programs—dealership service lanes. Public reporting around early 2025 highlighted new AI agents aimed at dealerships, including an appointment-booking agent and a call-intelligence agent that records, transcribes, and summarizes conversations for coaching dashboards. Those additions widen the dataset beyond SMS alone: managers suddenly expect call transcripts, disposition codes, and booking outcomes in the same governance envelope as texts.

Separately, metrics cited for 2024 in industry coverage described extremely high inbound call volumes handled through Numa’s dealership stack, alongside broadened DMS coverage—signals that integration buyers should plan for telephony metadata, not only chat bubbles. Your benefit is a specification that anticipates those fields instead of retrofitting them after the next product release.

• Dataset completeness — Engineering receives field dictionaries for conversations, rescues, assignments, translations, and—where enabled—voice intelligence artifacts.
• Operational KPIs — Product analysts join message latency with POS ticket IDs or repair-order numbers without manual CSV pulls.
• Vendor-neutral storage — Snowflake, BigQuery, or Postgres copies run on your retention schedule, not an ad-hoc inbox search.
• Faster procurement — Security reviewers get OpenAPI sketches, pseudocode auth steps, and explicit error envelopes before any production traffic moves.

Data available for integration (OpenData perspective)

The inventory synthesizes Numa’s published consumer-facing description with common enterprise messaging patterns. Exact field names depend on your contract, enabled modules, and any official partner APIs; we document what we observe under authorization rather than guessing proprietary schemas.

Typical integration scenarios

Each scenario lists a business context, names the data contract in play, and states how OpenData-style controls (consent logs, scoped tokens, replayable webhooks) keep the flow defensible.

1. Automotive service lane → data lake. Context: a dealer group already stores repair orders in a DMS-aligned warehouse. Data: conversation IDs tied to customer phones, optional call transcripts from the 2025 call-intelligence agent, and appointment outcomes from the booking agent. OpenData mapping: treat each transcript as structured text with immutable recording_retention_policy_id, mirroring how OpenFinance aggregators keep provable audit trails beside balances.
2. Restaurant group → finance close. Context: text orders must roll up nightly with POS net sales. Data: raw SMS, normalized SKU rows, void flags, and printer timestamps. OpenData mapping: append-only landing tables + idempotent upserts mimic payment-file ingestion used in OpenBanking reconciliation jobs.
3. Franchise CX compliance. Context: legal wants proof that promotional blasts only went to opted-in handsets. Data: outbound message IDs, template IDs, TCPA consent artifacts stored in your CRM, and Numa delivery receipts. OpenData mapping: join keys behave like PSD2-style consent receipts even though the underlying regulation is TCPA/CPRA.
4. Bilingual retail support. Context: stores in California serve Spanish-speaking shoppers via translated threads. Data: paired utterances, detected languages, agent edits. OpenData mapping: lineage fields (translator model version, human override flag) align with GDPR Article 5(1)(a) fairness and transparency expectations when EU tourists appear in logs.
5. Observability for pay-per-call hosting. Context: you consume our hosted endpoints instead of self-running workers. Data: per-call latency metrics, HTTP 429 backoff counters, and redacted payload hashes. OpenData mapping: usage metering resembles OpenFinance aggregator billing with customer-visible call logs.

Technical implementation sketches

Illustrative pseudocode only—final paths, scopes, and auth headers follow the surfaces your organization is contractually allowed to use (official partner APIs, exporter tools, or enterprise agreements). We never document bypassing authentication.

GET /integrations/v1/conversations?cursor=eyJpIjoiNDgifQ
Authorization: Bearer <M2M_TOKEN>
X-Tenant-Id: store_8812

200 OK
{
  "next_cursor": "eyJpIjoiNDkifQ",
  "items": [
    {
      "conversation_id": "cv_9a21",
      "customer_e164": "+15551234567",
      "channel": "sms",
      "assigned_user_id": "usr_404",
      "last_message_at": "2026-04-11T18:22:11Z",
      "tags": ["service", "priority"]
    }
  ]
}

POST /hooks/numa/inbound
X-Signature: sha256=9f86d08...
Content-Type: application/json

{
  "event_id": "evt_31aa",
  "type": "message.received",
  "conversation_id": "cv_9a21",
  "body": "Need a table for 6 at 7pm",
  "received_at": "2026-04-11T18:22:41Z"
}

// Handler: verify signature with rotated secret,
// enqueue to dedupe table on event_id,
// respond 204 within 2s or mark for retry.

POST /kitchen/orders/from-text
{
  "conversation_id": "cv_9a21",
  "raw_text": "2 fish tacos no dairy + large horchata",
  "location_id": "loc_771"
}

200 OK
{
  "order_id": "ord_5510",
  "lines": [
    {"sku":"TACO-FISH-2","qty":2,"mods":["no_dairy"]},
    {"sku":"DRK-HORCHATA-L","qty":1}
  ],
  "confidence": 0.86,
  "needs_clarification": false
}

Compliance & privacy

Business A2P messaging in the United States frequently implicates the Telephone Consumer Protection Act (TCPA) and related FCC guidance. Numa’s published service requirements state that customers—not the vendor—must determine applicability and maintain compliant programs, which means your integration must capture consent artifacts outside the chat UI.

Where California consumers appear in logs, the California Consumer Privacy Act (CCPA) as amended by the CPRA can govern disclosure, deletion, and opt-out-of-sale workflows for personal information retained in data warehouses fed from Numa exports. European contacts in bilingual threads may invoke GDPR bases and transparency duties; translation metadata therefore belongs in the same privacy inventory as message bodies.

Numa also discloses that SMS travels across networks that may not be encrypted end-to-end; architects should assume TLS to your API edge only and plan vaulting for any PAN-like tokens that accidentally surface in free-text orders.

Data flow / architecture

Numa mobile and web clients talk to NumberAI services over TLS. Authorized integration workers either call documented endpoints or consume sanctioned exports, writing into a short-lived landing bucket. Downstream ETL normalizes threads, applies PII hashing rules, and publishes curated views to analytics sandboxes. Finally, BI tools or CRM middleware query those views while retaining immutable raw archives for disputes.

• Node 1 — Client surfaces: Android/iOS apps plus browser inbox for staff.
• Node 2 — Integration plane: Webhooks, batch exports, or partner APIs with rotating credentials.
• Node 3 — Curated store: Warehouse tables partitioned by tenant and day.
• Node 4 — Consumer systems: CRM, QA scorecards, finance rollups, or observability dashboards.

Market positioning & user profile

Numa’s public positioning centers on small and mid-size businesses that still rely on voice lines but need modern messaging: restaurants texting pickup instructions, retailers coordinating repairs, and—through NumberAI’s vertical programs—automotive dealerships managing enormous inbound call volumes. Device coverage spans Android and iOS phones for frontline staff plus desktop browsers for shared inbox supervisors. Geography skews North American business telephony today, yet translation features imply cross-border clientele and multilingual staff, which matters when choosing residency for replicated warehouses.

Screenshots

Thumbnails below open a larger preview in-page. The grid stays compact until a stakeholder explicitly expands an asset, keeping the marketing story readable on first paint.

Close

Similar apps & integration landscape

Teams evaluating Numa frequently pilot or already run adjacent messaging and reputation stacks. Listing them here clarifies how datasets overlap and why a normalized messaging warehouse pays dividends even when vendors change.

API integration instructions

1. Inventory channels. Document every inbound path—native SMS, imported Facebook Messenger threads, webchat—because each channel changes consent language and retention.
2. Map identities. Decide canonical customer keys (CRM contact ID, loyalty ID) and enforce joins before message bodies land in analytics.
3. Choose transport. Prefer signed webhooks for near-real-time rescue metrics; schedule nightly bulk exports for historical backfills exceeding webhook windows.
4. Model authorization. Issue machine tokens per tenant with read-only scopes, rotate quarterly, and log administrator approvals in your ITSM tool.
5. Validate with shadow traffic. Run dual writes to a staging warehouse for two weeks, comparing counts against manual inbox samples before cutting over finance jobs.

What we deliver