I agree with much of what's already said here.
1. Figure out the difference between your inherent and accidental complexity. The Cynefin framework can be a useful way of thinking about what types of complexity you have.
2. Attempt to reduce (or remove) the complexity at root before assuming you must model it. "The biggest tragedy in business is doing well that which should not be done at all." For example, we had an important user still on a legacy browser because they couldn't afford to upgrade their old laptop. Instead of adding a complicated legacy layer, we bought them a new laptop.
3. Think of the whole system, rather than just shifting the burden somewhere else within the system. This is why microservice architectures so often fail - engineers think they've reduced complexity because each service is so simple. Yet all they've done is move the complexity into the operational layer, where it's harder to deal with. Gerald Weinberg is a fun writer in this space.
4. In many organisations, a lot of people's salaries depend on their ability to add complexity, especially to processes but also to code. If you are working within such an organisation, you will have a Sysiphian task until you solve the organisational problem. Whether you can solve this problem depends on how much influence you have. This is why "skunkworks" projects so often work until they are re-integrated into the mainstream.
5. Once you've sorted out the organisational complexity, separated the inherent and accidental complexity, you'll be most of the way there. Keep reviewing and challenging everything. Complexity is a weed that just seems to grow unless you constantly tend to the garden.