A2 · Tool workflow · financial services, insurance, logistics, public sector, healthcare administration

Document intake pipeline

Invoices, claims and forms become validated records. Exceptions reach a person with the reason attached.

Customer job: Turn a stream of supplier invoices or claim forms into correct records without manual keying.
Today: Clerks key fields from PDFs and email attachments; errors surface weeks later in reconciliation.
Accepted result: A record posted to the system of record that matches the document and passes every validation rule, or an exception a clerk clears in under two minutes.

Incoming documents are parsed, extracted to a typed schema, validated against master data and business rules, and posted to the system of record exactly once. Anything that fails validation goes to an exception queue with the failing field and evidence.

Google Cloud / primary provider

What carries each step.

Eventarc starts a Workflows run per object; Document AI parses layout, Gemini 3.8 Flash extracts to a JSON schema, validation runs on Cloud Run.

Data and model estate in one place: Gemini on Agent Platform, Agent Search and RAG Engine, AlloyDB and BigQuery, Cloud Run for everything else.

Component	Service	Evidence
	Person
	Cloud Run	prior, ga
	Cloud Storage	prior, ga
	Eventarc	prior, ga
	Workflows	prior, ga
	Document AI	prior, ga
	Apigee (AI gateway policies)	prior, ga
	Gemini 3.8 Flash (Agent Platform)	source, ga
	MCP server on Cloud Run	prior, ga
	AlloyDB for PostgreSQL	prior, ga
	Review console on Cloud Run	prior, ga
	External system
	Cloud Logging, Monitoring, Trace	prior, ga
	Evaluation service (Agent Platform)	confirm

Customer view / judgement, not benchmark

Weigh what this customer values.

Remove providers that fail a hard constraint first: residency, data rights, identity estate, latency, operating owner, budget. Then weigh the rest. Scores are architect judgement on a 1-5 scale for the pattern as specified, recorded with a rationale. They are not benchmarks. A provider's consumption opportunity never changes a customer score.

Performance×1Does the default path meet the latency and quality budget for this workload?Time to value×1How quickly does a team reach a first accepted result with managed parts?Speed to execution×1How much undifferentiated wiring stands between the design and a running slice?Cost per accepted result×1How predictable and low is the all-attempt cost at the stated volume?Exit cost×1How cheaply can the workload move models, data or runtime later?Governance fit×1Do identity, residency, audit and policy controls exist without custom building?

Performance 4Time to value 4Speed to execution 4Cost per accepted result 4Exit cost 3Governance fit 5
Mature document parsing and regional controls; several managed services to operate.
Performance 4Time to value 5Speed to execution 5Cost per accepted result 3Exit cost 4Governance fit 3
Quick to build; no dedicated layout parser, so cost per page depends on vision tokens.
Performance 4Time to value 4Speed to execution 4Cost per accepted result 4Exit cost 3Governance fit 5
Well-trodden services with strong controls; more IAM wiring.
Performance 4Time to value 4Speed to execution 4Cost per accepted result 4Exit cost 3Governance fit 5
Prebuilt models shorten the path for common documents.
Performance 4Time to value 4Speed to execution 4Cost per accepted result 5Exit cost 3Governance fit 3
Low unit cost and durable steps; layout parsing quality must be measured on the customer's documents.
Performance 3Time to value 3Speed to execution 4Cost per accepted result 4Exit cost 5Governance fit 2
Full control, all reliability engineering in-house.

Provider ledger / shown after the choice

What grows on Google Cloud.

These meters are the provider team’s consumption view. They explain the cloud workload; they never raise a customer score.

Gemini input and output tokens
Cloud Run vCPU-seconds and requests
Agent Search queries and index storage
AlloyDB or Cloud SQL instance hours
BigQuery slots or bytes scanned

Take it with you / generated from one specification

Exact views, and a running start.

Google Cloud views

SVG diagram for decks and documents
diagrams.net file editable, shared IDs
React Flow JSON for your own canvas
Mermaid flowchart

Every provider

Structurizr workspace C4 containers, one deployment view per provider
Journey sequence (Mermaid)

One-click template

No Google Cloud template yet.

The same answer contract runs on these runtimes today; its conformance suite is the porting checklist.

Vercel: anthropic/claude-sonnet-5.5 with gateway fallbacks
Cloudflare: @cf/meta/llama-3.3-70b-instruct-fp8-fast
Google Cloud: gemini-3.8-flash on the global endpoint

Customer ledger / assumptions you can move

Cost per accepted result.

Every attempt is charged, so failed runs count. Value comes only from accepted results. Illustrative assumption for an accounts-payable team; acceptance is the straight-through share, measured in a pilot.

Primary metric: Straight-through rate at target accuracy. Documents posted without human touch whose key fields match a sampled human check, divided by documents received.

Monthly value: €39,488
Monthly cost: €3,675
Net: €35,813
Cost per accepted result: €0.42

Eligible tasks per month15,000Share attempted with the system90 in 100Attempts accepted without rework65 in 100Minutes saved per accepted result6 minValue of an hour of that work€45.00Run cost per attempt€0.05Fixed monthly cost (hosting, review, support)€3,000

Experience / drawn from the specification

What the person sees, including no.

Screens are rendered from structured blocks in the catalog, so a refusal or an exception is designed before it happens. Values marked fixture are illustrations, not measurements.

INV-55203, Contoso Packaging: total does not match line items

Field	Extracted	Rule
Net total	4,180.00	Lines sum to 4,108.00
VAT	794.20	Pass
PO	PO-11842	Pass

Use line sumReject document

Exception consoledesktop · exception

Received. Reference DOC-8F2A.

Parsing
Checking against your purchase order
You will be emailed if anything is missing

Upload receiptmobile · success

Straight-throughfixture

Exceptions openfixture

Oldest in queuefixture

Received
Parsed
Extracted
Validated
Posted

Pipeline boarddesktop · default

Team workflow / who carries what

Every lane leaves evidence.

Lane	Responsibility	Evidence it leaves
Customer	Submit the document and receive a receipt ID	Receipt and hash
AI	Extract fields with source coordinates	Field confidence and region
System	Validate, post once, route exceptions	Rule results and idempotency key
Reviewer	Clerk clears exceptions; corrections become labelled data	Correction log
Operator	Watch queue age, straight-through rate and per-supplier accuracy	Daily pipeline report

Release evaluations

What must pass before anyone ships.

Key-field accuracy
>= 0.98 on total, supplier, date, PO
Labelled set per supplier layout; exact match after normalisation.
Straight-through rate
Reported per supplier; no global target until baseline
Pilot week with sampled human audit.
Duplicate posting
0 under replayed events and retries
Fault injection on post step.
Exception age
p90 cleared within one business day
Queue telemetry.

Risks and controls

What would go wrong first.

A plausible but wrong amount posts without review.
Arithmetic and PO rules in code; low-confidence fields force exception.
Retries post twice.
Content hash run key and ERP idempotency key.
Documents leave the required region.
Regional model endpoints and storage; verified per provider.

Image recipes / concept art only

When to generate, and when never to.

16:9 · gpt-image-2 or a current Gemini image model

Customer moment for a proposal cover or portal hero.

Cinematic editorial photograph, late afternoon light, shallow depth of field. Turn a stream of supplier invoices or claim forms into correct records without manual keying. The moment just after the accepted result: A record posted to the system of record that matches the document and passes every validation rule, or an exception a clerk clears in under two minutes. Real workplace, restrained palette of deep navy, warm paper and one cobalt accent. Natural expressions, no screens facing the camera. Concept art only. Do not render product logos, cloud icons, UI text or architecture labels; compose exact labels from the specification in code.

3:2 · gpt-image-2 (consistent characters across panels)

Team workflow storyboard for a workshop or executive briefing.

Five-panel storyboard, clean ink line with flat colour, consistent characters across panels. Panels follow the lanes customer, ai, system, reviewer, operator: (1) Submit the document and receive a receipt ID (2) Extract fields with source coordinates (3) Validate, post once, route exceptions (4) Clerk clears exceptions; corrections become labelled data (5) Watch queue age, straight-through rate and per-supplier accuracy Each panel shows hands and decisions, not screens. Concept art only. Do not render product logos, cloud icons, UI text or architecture labels; compose exact labels from the specification in code.

Do not generate

Any image a customer will read for component names, data classes or provider services.

Do not generate. Export the compiled view for the chosen provider. A generated architecture image can invent services, misspell names and draw icons the vendor never published.

A design, not a deployment: no realization on this page carries a provider receipt. Generated from the pattern source at revision c43825efda38. Apache-2.0 for catalog content and exports; provider names are trademarks of their owners; Google Cloud icons retain Google rights.

Next pattern: Enterprise MCP tool hub