Perspectives.
AI in real workflows, and where it stalls.
Short, sourced articles on putting AI into one workflow at a time, with a named person at the decisions that matter. Written by Doreen Chee, founder of Humancraft AI.
Most AI pilots stall because AI is added to a process designed for people, so it waits at the same handoffs and gets worked around. In Camunda's September 2026 survey of 1,000 process decision-makers, 72% said process-related challenges had caused AI initiatives to fail, at an average cost of $1.55 million per organisation. Pilots that reach daily use redesign one workflow with the people who run it, agree a baseline before building, and put a named person at each decision that matters.
Read the articleThree finance workflows where AI pays first, and a person still signs
Supplier invoices, overdue invoices and blocked orders: where AI pays first in finance operations, what it should never do, and what to measure.
AI in customer service: drafts, not autopilot
How to use AI in customer service without losing the human: ticket triage and reply drafts, status updates, and field service triage, with a person approving.
AI in regulated work: how to keep a named human accountable
How banks, wealth managers and regulated firms keep a named person accountable for AI: a three-tier control model, an AI register, and three worked use cases.
The relationship manager's client pack: an AI workflow compliance can sign off
How banks and wealth managers can use AI to prepare client reviews: a sourced draft pack, compliance flags, the RM in charge, and every version kept.
Retail AI that starts with stock, not content
How retailers can connect marketing to inventory with AI: flag promoted lines that will sell out and slow lines that need a push, with merchandisers approving.
Before you add AI to operations, ask the people who run the work
Why operations teams should map the real workflow with the people who run it before adding AI, and how assisted and autonomous modes differ.
Whose agent touches the money? What JPMorgan's no to outside AI agents means for your workflows
In one week the largest US bank refused outside AI agents access to customer accounts, and three US regulators asked banks whether they can control and shut down autonomous AI. What bank and fintech leaders should do with that this quarter.