Interrupted time series is the disciplined version of a before-and-after comparison. Instead of comparing two averages, it models the weekly or monthly history before the change, including its trend and seasonal pattern, and asks how far the results after the change depart from where that history was heading.
It is often the only option for an organisation-wide system with no possible comparison group. Its weakness is anything else that changed at the same moment. A reorganisation, a price change or a new policy announced alongside the AI launch will be credited to the AI unless it is modelled separately.
Three habits make it usable: a long run of regular data on both sides of the change, the expected shape of the effect written down before the analysis, and a control series, such as a related process the AI does not touch, wherever one exists. Even then, a benefit measured this way should be graded lower than one measured against a comparison group.