A simple method we've seen is adding a binary COVID/non-COVID variable to the data for training and inference.
You could also bring in more detailed public COVID-related data - whatever metric you think may have had an impact on the particular revenues or expenditures you're predicting. Could be cases, cases per 100k people, deaths, etc.
You could also bring in broader economic data that may pick up on the effects of COVID at various times over the past couple years, like the BEA's consumer spending data.
Feel free to reach out to your CSM if you'd like to talk details with our team.