CRM, ERP and Payment Integrations That Don't Lose Orders

2026-11-14

The most expensive mistakes in business systems are rarely visible on screen. The site works, the Pay button works, the customer gets an email. Then it turns out the payment went through but no order was created in the CRM. Or it was created twice. Or ERP stock updated an hour late and a product that was already gone got sold. Managers reconcile spreadsheets by hand, and the owner trusts no number.

Integrations between a website, CRM, ERP, accounting system and payment providers are a core part of my work. Here is how I design them so the data matches and a failure in one service does not break the whole process.

Typical problems clients come with

Symptoms of a bad integration
SymptomReal cause
Duplicate deals and contacts in the CRMNo idempotency: a repeated webhook or double click creates a new record
Payment exists, order does notWebhook handling crashed and there is no retry or reconciliation
Stock does not matchTwo systems both think they are the source of truth and overwrite each other
The integration "sometimes" failsAPI rate limits, timeouts, errors with no logging and no alerts
Nobody knows what brokeNo event log or monitoring, only customer complaints

None of these are fixed with "one more script". They are fixed with integration architecture.

First: who is the source of truth

The first question in every project is which system owns each entity. Customers may live in the CRM, products and stock in the ERP, prices in accounting, payment status at the payment provider. Once this is written down, most conflicts disappear: each system writes only its own data and reads the rest.

Then I describe the exchange contract: which fields, in what format, how statuses map between systems, and what to do with fields that do not exist on the other side. It is the boring part, and it saves weeks of debugging. I wrote separately about why the data schema is a contract: https://zavada.dev/blog/sql-is-the-contract.

Reliable event delivery

Between systems I almost never make direct synchronous calls at the moment a user acts. Instead:

Incoming webhooks are verified first (signature, origin), stored in the database and only then processed. If processing fails, the event is not lost.

Outgoing changes go through the outbox pattern: the database write and the event to send are stored in one transaction, and a separate worker delivers them to the CRM or ERP with retries and backoff.

Every operation has an idempotency key. A repeated webhook from a payment provider, or a resend after a timeout, does not create a second payment or a second deal.

Queues absorb load spikes and respect external API rate limits. Events that still fail after several attempts go to a separate dead-letter queue for review instead of disappearing.

Payments: a separate level of responsibility

With payments a mistake costs money, so there are extra rules. Payment status never comes from the customer's browser, only from a server-side webhook or a request to the provider. Amounts are stored in minor units, never floats. Refunds, partial payments, subscriptions and chargebacks are modelled as states, not exceptions.

Daily reconciliation is mandatory: the system compares its transactions with the provider's statements and highlights differences. This catches the rare case when a webhook never arrived. More on payment platform security, including crypto, in https://zavada.dev/blog/payment-platform-security-fiat-crypto-2026.

What I integrate with

Example systems
TypeExamples
CRMHubSpot, Salesforce, Pipedrive, KeyCRM, Bitrix24, custom CRMs
ERP and accountingOdoo, BAS, SAP Business One, Microsoft Dynamics, warehouse systems
PaymentsStripe, PayPal, LiqPay, WayForPay, Monobank, crypto payments
LogisticsNova Poshta, Ukrposhta, Meest, international carriers
OtherGoogle Sheets, telephony, email marketing, marketplaces, banking APIs

If a system has no proper API, we work through file exports, an intermediate database or careful scraping, but always with a log and reconciliation.

Monitoring and support

An integration without monitoring breaks quietly. Each one has an event log that shows what was sent, when, where and with what result. Alerts fire when the error queue grows, the success rate drops or an external API changes its response. An admin screen lets a manager see a stuck event and retry it without calling a developer.

How a project runs

1. Process review: the path of an order from the website to shipping and payment, every system involved, who works where.

2. Data schema, sources of truth and the exchange contract, agreed with the team.

3. Implementation one flow at a time, with a test environment and reconciliation on real data.

4. Launch alongside the old process, monitoring, then ongoing support.

Scope and cost depend on the number of systems, the quality of their APIs and the complexity of business rules, so I quote a price after the brief and process review, with a written plan and milestones.

Conclusion

A good integration is invisible: orders, payments and stock simply match, and when something breaks you know before the customer does. That takes explicit sources of truth, reliable event delivery, idempotency, reconciliation and monitoring.

If your data drifts between the website, CRM, ERP and payments, or you are planning a complex system with these integrations, see how I build web applications: https://zavada.dev/web-apps. Describe your systems in the brief form: https://zavada.dev/contacts?service=web-apps

Need a developer for the long run?

I can start now and work with you for months or years, not hand off after two weeks. Send me a message and we will talk through your project, requirements and next steps.

Ask about experience or services