The robbed-bit signaling thing is really a red herring: it was a problem, but not THE problem. At the time analog 56kbps modems were popular, it was already possible to negotiate 64kbps/8-bit clean DS0 channels over T1 — you just needed a PRI or an inter-machine trunk to do so.
In fact, if you were an ISP at the time, you generally needed those modems connected to T1 PRIs or IMTs anyway to support your ISDN customers economically. (My memory may be getting fuzzy on this, but I also remember that certain access concentrators with k56/v.90 digital modems would only negotiate the higher analog speeds on a PRI or IMT bearer channel.)
What there was absolutely no appetite to do, on the part of the voice carriers, was beef up the bandwidth available on POTS lines by developing new ADCs, getting rid of companding, or engaging in labor-intensive “facilities grooming”[0] for a POTS subscriber, in the way they routinely did for T1[1] circuits.
0 - Telcos call their lines and equipment “facilities.” Grooming, in the context of outside plant (telco terminology for the physical cabling that goes to subscribers) means to select wire pairs that exhibit good noise and echo characteristics and to remove elements, like poor splices and bridge taps, that impede performance.
1 - T1s, unless delivered to a customer via an optical transport network, use the same copper lines as analog POTS lines. Originally, T1s required two pairs (four wires). This eventually was replaced with HDSL4 (two pairs, but better distance and repeater characteristics,) then HDSL2 (single pair.) As the names suggest, the encoding used was a DSL variant.