running deep research on 100 companies = basically $100 gone. anyone actually solved this?

so quick context - im building a deal screening thing where i run a fast LLM pass (like 6 different scoring angles) on every company that comes into a batch. this part is cheap, can throw dozens of companies at it no problem. now i wanted to add an actual "deep research" step on top for the ones that look promising - either perplexity's sonar-deep-research or this custom multi tool research agent i built earlier for another part of the product. problem is both of these are basically priced (and timed) for someone asking about ONE company, not for running across a whole batch. did the math and if i run it on like ~100 shortlisted companies just once, its close to $100. and thats before i even need to re run anything lol. stuff im thinking about rn, would love opinions: - just do a funnel - keep the cheap screen like it is, only send the top N (say top 10-15) into deep research instead of the whole batch - a lot of these companies are in the same sector / competing with each other, so is full deep research per company actually wasteful? can that research be reused somehow instead of starting fresh everytime - maybe theres a cheaper "in between" option - not a full agentic deep research call, but more targeted tool calls that get 80% of the value for way less - if anyones actually run perplexity/parallel ai's task api at real volume, did you find any pricing tricks that helped - has anyone just built their own "lite" version of deep research thats actually cheaper but still useful genuinely trying to figure out how ppl running these pipelines at any real scale are dealing with this, not looking for a "just cache it" one liner lol. any real experience appreciated