Enterprise AI starts on a hated workflow

Enterprise AI is a job, not a genre. Models used on work the company already pays people to do. Claims. Credit. Knowledge work. Customer ops. If the use case only exists in a town hall, it is not enterprise AI yet. It is a demo with catering.

I am tired of programmes that begin with a tool catalogue. Copilot, ChatGPT, a private GPT, a vector database, a slide about ethics. Nobody named the workflow. Nobody named the owner of the old way. Then we wonder why adoption is "cultural."

It is not cultural. It is managerial. People will use a model that removes a hated step. They will not use a model that adds a login to a process you refused to kill.

Start on the work

Pick the workflow with the worst cycle time. Time it. Put a model on the ugly part. Keep a human on the bit that can hurt a customer or a regulator. Measure the hours. That is the first enterprise AI project. It will not impress a conference. It will impress the person who used to copy-paste until 7pm.

Corporate AI training belongs here, not as a side quest. If you will not teach the people who do the work, you are buying a licence for a ghost. AI training is the how. This page is the where. AI transformation is the sequence of wheres.

A CAIO can help. They cannot replace a process owner. The interesting appointments fund literacy in the first 120 days and then pick a hated process. The rest are listening tours.

Banks are a clean example. AI adoption in banking shows up as hiring and training before it shows up as a product launch. When a HK or SG bank staffs AI-for-teams, write that. Do not write "banks are embracing AI." One is a story. The other is a press release.

Sprints, not slideware

An AI design sprint is how you find out whether the job is real. Two days. A model in the room. Same kill criteria as any other sprint. If you leave with a demo and no decision, you ran a show.

Infinitev used AI to generate landing pages and test ideas fast. That is enterprise-adjacent in the best way: a tool in service of a learning loop, not a transformation narrative. Steal the loop. Do not steal a claim we did not make.

I would put productivity next to this hub on purpose. AI that writes better emails for a meeting that should not exist is a faster waste. Hours come back when rooms die. Models help after that, or as the way you kill the room.

Collective Campus trains AI skills for teams and leaders. One link. The newsroom will keep naming companies in AU, NZ, HK, SG, the UK, the US and Canada that are actually staffing the work.

What we publish here

AI training without the slideware. When a CAIO funds AI training. AI design sprints that ship a decision. The playbook.

If you cannot name the job the model is replacing, you do not have an enterprise AI programme. You have a chatbot and a budget code.

Risk teams will ask about hallucinations, data leakage and vendor lock-in. Those are real. They are also how programmes stall at the policy layer. Answer them with a narrow slice: this dataset, this tool, this human in the loop, this audit trail. Then run the slice. A 40-page AI policy with no slice is how legal feels useful while the business copies data by hand. I want the policy. I want it attached to a job.

Vendors will try to sell you "enterprise AI" as a platform that includes search, agents, copilots and a roadmap to autonomy. Buy a job. If the platform cannot do the job in a trial on your ugly process, you did not buy enterprise AI. You bought a tour.