I was wondering about that, it is described as:

"where rather than hold the data, it is split in two so CertiVox has one half and the user has the other, and law enforcement would need both to access the data."

By data here, do they mean encryption key? If so then I'm not sure what splitting it in two does as you presumably have to join them together somewhere and GCHQ would presumably just attack wherever the complete key is available.

[Edit: Product details here, it appears to be Open Source: http://www.certivox.com/m-pin/]

Defining some terms: Alice, Bob, the company, and GCHQ (Eve).

Alice sends Bob stuff, via the company.

GCHQ sends RIPA requests to the company, which means that GCHQ get Alice's communications. Neither Alice nor Bob are aware.

So, a new scheme where Alice and the company need keys to decrypt means that GCHQ send RIPA requests to the company, and to Alice. Now Alice, and the company, are aware of the RIPA request. Bob isn't. But at least Alice can stop sending things to Bob.

I don't know anything about the implementation, but I would assume that the idea is to avoid having the service provider be a single point where complete failure can occur. Perhaps this new system offloads that risk to the customer.

But then I get confused, what service is the service provider providing if the customer still needs to carry around half of 'the data'?

Splitting it in two does not mean splitting it in half.

Found some detail of "splitting in two" to assure no single point of failure here https://certivox.org/pages/viewpage.action?pageId=10420248 "Distributed Trust Authority"