A working CRMin one day.

On 20 October the company’s main CRM suffered an outage. Clients write in four channels, and managers needed a place to answer them.

The company already had a cabinet holding a copy of its deals, and since 19 October its Inbox had been reading MAX, Telegram and VK, with replies to clients switched off. Managers still did their daily work in the main CRM. The goal for the day: the managers work in the cabinet.

The same deals, the same conversations, payment for clients. And every change must reach the main CRM, so the two systems never drift apart.

At 11:20 a transfer from the main CRM pulled in the latest data: the cabinet held 4,186 deals, and reconciliation found 0 discrepancies. The transfer runs every 10 minutes.

By 11:22 the cabinet was fixed as the working replacement for the duration of the outage. Managers received access, and replies to clients were switched on for MAX, Telegram and VK.

By 12:27 Avito joined the shared inbox. From then on clients in VK, MAX, Telegram and Avito are answered from one window.

The window behaves like a messenger: Enter sends the reply and new messages appear by themselves. The reply goes out in the channel where the client wrote.

Before switching on we tested the exchange on a test deal. The stage, budget, note and task from the cabinet reached the main CRM. A budget changed there later came back to the cabinet. The echo duplicated nothing.

At 12:43 the sync went live. The rule was chosen together with the company: if a deal is edited on both sides, the later edit wins. It decides field by field, so a budget from the cabinet and a stage from the main CRM do not erase each other.

The cabinet sends only what differs from the last agreed state. A repeat run duplicates nothing. It calls the main CRM no more than twice a second.

By 13:08 two more functions went live: “Delete a deal” and “Merge clients”. Test and spam deals are now deleted, and a person who wrote from different messengers becomes one client. Both took about 1 h 15 min. At first the transfer from the main CRM overwrote the merges. We spotted it in testing and fixed it in the same release.

By 13:45 two buttons appeared in the reply box: “Template” and “Payment link”. 16 templates came over, all from the main CRM. A manager answers with ready text and adds a short prepayment link. Both buttons took about 1 h 20 min.

Clients forget to pay, so the cabinet reminds them itself when the link was issued from the cabinet. For a link valid for one to two hours, a reminder goes out 30 and 10 minutes before it expires, in the same conversation and channel. Before each reminder the system checks the booking: if it is paid, the remaining reminders are cancelled. Nothing goes out between 22:00 and 09:00 Omsk time.

The first test run exposed a fork: in the test the conversation was closed and the deal deleted, and the reminder was cancelled. For a deleted deal that is right. But closing a conversation means removing it from the list, not turning the client away. The fix was made at 16:55.

The second run on a test booking went all the way: both reminders arrived, at 18:01 and 18:21. At 18:29 the reminders went live for Art Quest. Up to deployment they took 1 h 05 min.

A plan for a CRM like this was estimated at 155–195 hours of work, or 8 to 10 calendar weeks. We built the foundation on 15 October in 2 h 50 min of working time: core, transfer, screens.

On the 20th four features were added to it, the ones you cannot work without when you replace the main CRM. All four took about 5 hours of working time in total by the time log; the sessions partly ran in parallel.

By 29 October the cabinet holds 4,312 deals (one of them deleted) and 9,723 client records. The cabinet’s inbox holds 747 conversations from four channels.

The exchange with the main CRM works: 12 cabinet records reached it through the send log, each marked done. The log remembers what went out and when, so after a network failure a record is never sent twice.