Quantiphi

Their pitch was 'we'll build you custom AI solutions' — which worked beautifully until OpenAI started shipping what used to require six-month Quantiphi engagements as pre-trained APIs you can call from a Slack bot. The consulting hours are still billable; the custom models are increasingly not.

Data scientists, ML engineers, and model architects — the exact roles they built their headcount around — are being compressed by Cursor, Claude, and AutoML platforms that ship production-ready models faster than their Discovery Phase PowerPoints. They're hiring PhD statisticians to compete with chatbots.

Enterprise relationships and Google Cloud partnership credentials are real assets — CXOs who've spent millions on their recommendations don't switch vendors casually. But the moat is consultative trust, not technical irreplaceability, and trust erodes quickly when the new vendor's demo solves in 20 minutes what took Quantiphi 20 weeks.

They're pivoting hard toward 'AI governance' and 'responsible AI implementation' — which translates to 'we'll help you manage the Claude deployment that replaces what we used to build custom.' It's the consulting equivalent of teaching people to use the microwave that put your restaurant out of business.

Enterprises still need AI strategy and implementation. They don't need armies of data scientists to build what OpenAI ships out of the box — the question shifted from 'how do we build intelligence?' to 'how do we deploy intelligence?' and that's a much smaller, cheaper conversation.