Architecture
customer management: connecting the information system, the agent and the WhatsApp bot
Each customer has one card: what they asked, what was promised, and what is still open. The article ties that to the work of an accounting office, without numbers or clients we do not have.
This guide is general information, not legal, medical, tax or professional advice, and it is not a substitute for reviewing the actual case.
customer management is what this article is about, even when the technical tool has another name. For an accounting office the day looks like this: documents on WhatsApp, questions to the client by email, and report status with whoever grabbed the file. Each customer has one card: what they asked, what was promised, and what is still open. A new request opens on the card, not in a group. Status is updated where the team works. Customer management is the information system. The AI agent drafts on top of it, and the WhatsApp bot files what the customer wrote. A salesperson sees their customers. Management sees the picture. Finance sees what the invoice needs. We do not build a fake sales pipeline. We put in what the business actually does with a customer. The rest of the article is how a software house builds that without leaving the team with a demo.
CRM automation is the circulatory system of revenue operations: leads become opportunities, opportunities become invoices, support issues update health scores, and marketing segments react to product telemetry. When integrations fail, teams debate dashboards instead of coaching sellers.
Declare a system of record per entity and lifecycle stage. Ambiguity (“Salesforce says active, billing says churned”) is a design defect, not a philosophical debate. Document resolution rules with timestamps and authoritative writers.
Field mapping is where projects die quietly. Maintain a data dictionary-source column, transformation, default, validation, owner. When vendors deprecate API fields, contract tests against sandboxes should burn red before production silently stops syncing.
Bidirectional sync is harder than it looks: concurrent edits, offline mobile updates, and batch imports race. Choose strategies-last-write-wins with audit, source-precedence by domain, or manual merge queues for high-value accounts.
Respect API ergonomics: pagination cursors, rate limits, bulk endpoints, and webhook retries. Queue-based workers smooth bursts; naive loops trigger 429 storms that masquerade as “random CRM bugs.”
Identity resolution ties contacts, leads, and users across systems stable keys-not only email, which typos and plus-addresses fracture. Plan for merging duplicates and handling GDPR erasure across copies.
Monitor like a tier-one service: lag between systems, record deltas, webhook delivery failures, and auth token expiry are leading indicators. Alert the integration owner, not generic IT, when sync health degrades.
Privacy propagates: marketing automation must honor unsubscribe flags, product telemetry must respect consent metadata, and exports for AI features must consider data minimization.
Change management includes training revenue teams on new required fields-otherwise automation enforces garbage. Pair technical rollout with playbooks.
In summary: CRM automation thrives on clear ownership, documented mappings, realistic conflict resolution, API-savvy workers, and operational monitoring. Treat it as product infrastructure, not a one-time integration project.
You can come back from here to a plain conversation: Bring what a customer looks like for you: the fields people actually fill, not twenty fields nobody touches. There is no “who spoke to them yesterday”. The card says. AI agents, an information system and a WhatsApp bot stay the same offer, with human approval, from an official Meta provider.