[Advice/Vent] How to coach an insular and combative science team

My startup was acquired by a legacy enterprise. We were primarily acquired for our technical talent and some high growth ML products they see as a strategic threat.

Their ML team is entirely entry-level and struggling badly. They have very poor fundamentals around labeling training data, build systems without strong business cases, and ignore reasonable feedback from engineering partners regarding latency and safe deployment patterns.

I am staff level MLE and have been asked to up level this team. I’ve tried the following:

- Being inquisitive and asking them to explain design decisions

- walking them through our systems and discussing the good/bad/ugly

- being vulnerable about past decisions that were suboptimal

- offering to provide feedback before design review with cross functional partners

None of this has worked. I am mostly ignored. When I point out something obvious (e.g 12 second latency is unacceptable for live inference) they claim there is no time to fix it. They write dozens of pages of documents that do not have answers to simple questions (what ML algorithms are you using? What data do you need at inference time? What systems rely on your responses). They then claim no one is knowledgeable enough to understand their approach. It seems like when something doesn’t go their way they just stonewall and gaslight.

I personally have never dealt with this before. I’m curious if anyone has coached a team to unlearn these behaviors and heal cross functional relationships.

My advice right now is to break apart the team and either help them find non-ML roles internally or let them go.