group_by() splits a table into groups, and the next summarise() gives one row per group: orders |> group_by(region) |> summarise(units = sum(units)).
Group by two columns and you get one row per combination. Here is the trap: summarise() removes only the last grouping column, so the result is still grouped by the first. Later steps then work within those groups without telling you. Add .groups = "drop" to summarise(), or ungroup() afterwards, to finish with a plain table.
group_by() also changes mutate(): sum() inside it sees only the group. That is how you get each row's share of its own group's total.
One row per region and channel, returned as a plain table.