As you suggest in the question I think an improvement would be: "impact of -org X- -trying- to influence governments vs direct work"

Some key considerations:

• Gov interventions' numbers in my experience generally have much better expected values in back of the envelope models and then often look much less good when you add a bunch of additional intuitive discount factors.
• Attribution is highly uncertain with policy interventions.
• This said, I expect many of the highest return per dollar funding opportunities to be in policy and research.
• What's the tractability/chance of success of the government intervention for the given team? It might actually be very very low.
• Is this team more likely to do more harm than good? There are many ways of doing damage or damaging future efforts. I think this rules out a bunch of approaches that would otherwise justify their low tractability and
• How time pressured is the intervention? Potentially risk is worth it. My general rule of thumb currently on this is something like "major risks are very rarely worth it, be very careful and wait for great te
• Sometimes "direct work+ " is the best policy intervention. Do something the government could adopt really well, show that it works and has public support, then try lobbying for it. This said, it seems like quite a few charities just strap this argument on to what they were going to do anyway, which is fine so long as they've thought it through properly, just something to look out for.
• Lobbying / technical assistance etc opportunities are generally going to be fewer and further between.