Simple. The article states they use the UK's death registry. The problem is they have no way to flag an erroneous match for longer than 12 months.

“Massive scale” is a distraction because society is even more massive. For example, if you’re talking about school pensions you have a school system with employees and that means you could do things like have someone check ID in person, and perhaps require a non-relative witness if there’s any question. Doctors offices could similarly be enlisted to certify that person X is the patient they’ve been seeing.

The problem with this is simply that it’s not absolutely free, and the people promising savings through outsourcing need it to be free to make their promises happen. Cutting budgets everywhere removes the slack you need to deal with things like this.

You'd think if they are able to reliable set up a contribution scheme that works at scale that they could set up a system to ensure they only pay out to individuals that are still alive.

The possibility of impersonation isn't limited to pension pay-outs but many other things as well. A once-per-year on-site visit would work and would just be a cost-of-business item, and that cost could be lowered substantially by collaborating with other entities who take an interest in such information (banks, governments etc).

They send pension checks out to specific addresses and accounts at exactly the same scale. Does it seem unreasonable to add 'attempt to call the pensioner' into the 'maybe they are deceased' part of the verification protocol?