Applied AI Doesn't Work
Summary
Vas argues that enterprises fail to gain AI ROI because they automate inefficient workflows instead of redesigning them, a mistake mirroring 1990s IT failures described by Michael Hammer.
Applying AI to bad processes fails.
Using AI to mechanize existing workflows merely makes 'garbage faster' without addressing the underlying queues and handoffs that cause delays.
most companies apply AI onto their garbage processes, and the end result is just making garbage faster.
Individual satisfaction masks organizational stagnation.
The Microsoft 365 Copilot trial saw 72% user satisfaction, yet Excel analysis was slower and PowerPoint quality dropped by half.
We did not find robust evidence to suggest that time savings are leading to improved productivity.
Varick delivers value through re-engineering.
By mapping actual work via process mining and interviews, Varick deployments achieved over $100M in value, including a 90% drop in payroll corrections.
In total, the value delivered across this set of deployments was over $100M.
Process ownership beats system consolidation.
Attempting to consolidate ERP or CRM systems first wastes years; agents succeed by having one process definition and one owner regardless of system disagreement.
Agents replace complex step sequences.
Properly redesigned workflows can collapse a 25-step process into three agents and two human checkpoints by categorizing steps into deterministic, agentic, or human buckets.
a 25 step process becomes 3 agents with 2 human checkpoints.
Highlights
I learned most companies apply AI onto their garbage processes, and the end result is just making garbage faster.
Read the original ↗Save to your Sponge
Summarized with Sponge. Quotes are the author’s own words.