Telstra uses AI to reconcile billing during a migration between systems
The comparison starts between two databases
When migrating customers between platforms, the Australian carrier Telstra needs to reconcile charges and payments recorded in the old and new systems. In a post from October 27, Microsoft reports that the company adopted an artificial intelligence agent to find inconsistencies, recommend corrections and open cases with additional information for human review.
According to Microsoft, the solution supported more than 272,000 service migrations and reached accuracy above 95% on recommendations in production. The post doesn't define the basis of that calculation, the period measured, the types of error considered or how many recommendations were changed after review. The solution’s period of operation was also not reported. The numbers should be read as results reported by the vendor, not as an independent error-reduction rate.
AI recommends; review stays with people
The agent cross-references information from the two platforms to flag differences, suggest a corrective action and create cases with context for human review. The source doesn't say the system changes charges automatically or makes the final decision on the customer's account.
This limit is part of the operational design: when a record affects a person's payment, balance or service, the team needs to be able to check where the data came from, understand why the alert was raised and record the decision made. The recommendation only helps if it comes with enough context for that review.
The problem can exist outside telecom
The case is specific to Telstra. The reading for other companies is an inference: a migration of ERP, CRM, billing platform or customer base can also require comparing records that don't use the same rules or formats. The case doesn't show that the same results would repeat in those environments.
Before automating, the company can pick a bounded process and measure how many discrepancies are found today, how long the team takes to handle them and which ones affect customers or cash. It also needs to define which system is the reference for each piece of information, which alerts require review and who approves any correction. These criteria help compare the current routine with a controlled test.
Sources
Microsoft, 27 Oct 2026. https://blogs.microsoft.com/blog/2026/10/27/the-platform-approach-to-becoming-frontier/