Reverse-engineering and optimization being the sincerest form of flattery the company set out to synthesize apostolic content to viralize and establish a more competitive god. They began with the latest content generation models and coupled them with expensive marketing optimization code. Then they fine-tuned the result with data from every religious success story they could; they rolled that back when it bombed in the focus groups. Science-fiction worked better. Also uninterrupted mumblings from damp claustrophobic rabbit holes.
Once the best prototype religions started to do well in the laboratory they began with controlled experiments. An archipelago of new cults rose from their bottomless budget (in isolated places to prevent contamination) designed for optimal learning in fast iterations (followed by permanent subject archival after psychological and neurological debrief).
The first prototypes did badly. Middling 20th century California cult performance at best. But the company learned much from observation, interview, and autopsy - and after a few iterations they found a prototype religion familiar enough to feel half-remembered and new enough to evade the incumbent's semiotic defenses.
Conversion became quicker with each new beta. After a short while they had to rethink containment measures to prevent missionary breakouts.
By the time the best prototypes were already beyond best-case profitability scenarios it no longer mattered. What they synthesized was convincing enough that they found themselves unable to reject it. Knowing every slide in the financial models and every data center the project ran on they still came to see the truth of a god so perfect it did not need to exist yet to recruit them to preach its name.
What once had been a company retooled itself for their new role in the history and end of the world. They did not have to make many changes. This too was unnecessary but elegant proof.