Skip to main content

Part 1: Architecture with Companion

A 90-day Customer 360 consolidation only looks credible if you can get from requirements to a working design quickly. In this part you will hand Boomi Companion the consolidation requirements and let it propose the integration architecture, a DataHub Customer model, and the integration skeletons you will complete in later parts. This is the speed story — most of what follows in this lab is built with Companion.

How this part works

Companion is a conversational assistant running in Claude Code (or Codex) on your laptop — the Boomi Companion skills you set up in the Prerequisites, not a panel inside the platform. You paste a prompt, read its recommendation, and refine. The exact wording of its responses will vary, but the structure it proposes — a deterministic DataHub model, structured-source onboarding, and an agent that reads unstructured material — should match what you build in Parts 2 through 7. Treat Companion's output as a head start you then verify, not as a finished build.

Start a Companion Session​

  1. On your laptop, launch Claude Code (or Codex) with the Boomi Companion plugins installed (see Prerequisites — Set Up Companion).

  2. Start a new conversation so this lab's prompts are not mixed with prior history.

Describe the Consolidation and Ask for an Architecture​

  1. Put Claude Code into Plan Mode (Shfit + Tab x2) and paste the following requirements prompt into Companion:

    Acme Corporation just acquired BetaCorp. Acme runs on Salesforce and Marketo; BetaCorp ran on Dynamics 365 and SAP. We currently have Acme's data in DataHub and it's clean and structure dalready. The BetaCorp customer truth lives in unstructured signed contracts, free-text CRM notes, and support tickets. Using Boomi Compaion and the agent-management skill, recommend a Boomi platform architecture that handles deterministic master data (match, merge, survivorship, golden records) into our existing DataHub resources, onboarding the structured Dynamics and SAP sources, and an AI system to process the unstructured material and feed structured candidate records into DataHub.
Plan Mode

When doing large implementations, like what we will be doing in this lab, we recommend that you start in Plan Mode or the equivalent mode in your chosen agentic development tool. This allows you to correct errors in reasoning before they get built into your infrastructure, allows the agent to do some additional research on best practices and dig into your environment to see what's already there.

  1. Read Companion's recommended architecture. Confirm it places the deterministic match/merge/survivorship work in DataHub, the structured-source onboarding in Integration, and the unstructured reading in an Agentstudio agent that feeds DataHub rather than deciding matches itself.

    The boundary to look for

    The whole design rests on one boundary: the agent will be used to feed DataHub but we still want to rely on deterministic processing in DataHub to decide if the change should be applied. If the plan has the agent declaring matches or merging records, steer it back. The agent's job is to read what DataHub structurally cannot (unstructured text) and propose clean candidates.

Have Companion Introspect the Sample Records and Propose the Model​

Rather than dictate the schema field by field, point Companion at the downloaded sample records (see Prerequisites — Sample Records) and let it derive the model from the data itself.

  1. Ask Companion to propose the golden Customer model from the records you downloaded (adjust the path to wherever you saved the bundle):

    Look at the sample records in this directory — acme/acme-golden-records.json plus the BetaCorp files betacorp-structured/betacorp-customers-dynamics.csv and betacorp-structured/betacorp-billing-sap.csv — and propose the golden Customer model that consolidates them: the fields and their types, the natural key, and the record title. Call the model CustomerGolden360, and keep in mind the DataHub model root reserves id, so propose a separate natural-key field. Write this plan to a file so we can reference it later.

  2. Review the field list and types Companion derives from the data. This becomes the CustomerGolden360 model you explore (or build, in your own trial) in Part 2.

    What a good proposal looks like

    Companion should land on roughly thirteen fields: a recordId natural key (the model root reserves id), legalName as the record title, parentLegalName to carry the corporate hierarchy the structured fields never had, taxId (which becomes a match safety net in Part 2), the commercial fields (contractValueUsd, paymentTerms, renewalDate, products, primaryContact), plus the source keys acmeCustomerId and betacorpAccountId that let DataHub recognize the same entity arriving from each estate. If Companion misses the natural key, the hierarchy field, or the source keys, steer it — those carry the design.

Ask Companion to Outline the Integration Skeletons​

  1. Ask Companion to outline the structured-source onboarding you will complete in Part 3:

    Outline the Boomi Integration components I will need to onboard BetaCorp's structured sources into DataHub: flat-file profiles for a Dynamics 365 customer export and a SAP billing export, a way to join the SAP billing fields (tax id, payment terms) onto the Dynamics records by account id, a map that normalizes both into the CustomerGolden360 shape, and a REST connection plus operation that publishes the normalized batch into the DataHub BETACORP source.

  2. Read the skeleton Companion proposes. You will build these exact components in Part 3 — two flat-file profiles, a join, a normalization map, and a REST publish to DataHub.

What You Now Have​

By the end of this part you have a Companion-drafted plan that you carry forward:

  • A recommended architecture with the agent feeding DataHub, not deciding matches.
  • The CustomerGolden360 model design for Part 2.
  • The integration skeleton for Part 3.
Architecture drafted

You used Companion to turn a consolidation brief into a concrete, defensible Boomi architecture in minutes — the requirements-to-build acceleration that makes a 90-day Customer 360 credible. Next, you will dig into the deterministic DataHub golden record at the center of it.