Where should an AI roadmap start? With the operational work you’ve already offshored. To outsource a process, companies had to document it, standardize it and measure it, and that’s the same groundwork AI needs.
Someone recently asked a bank CEO a question almost every company is wrestling with: There are hundreds of things we could do with AI. Where do we start?
His answer surprised me.
Start with the work you’ve already offshored.
Not necessarily the process everyone complains about. Not the flashiest customer-facing use case.
Look at the boring operational work you sent somewhere else years ago.
| Related: Why Smaller Banks May Have an AI Advantage
The biggest obstacle to AI isn’t AI
Why? Because one of the biggest obstacles to AI isn’t actually AI.
It’s the mess underneath it.
As the CEO put it: “Your data sucks.”
Take something as basic as closing a loan.
At many banks, it isn’t really one process.
One office does it one way. Another office does it slightly differently.
A team inherited through an acquisition has its own procedure.
Someone has a spreadsheet nobody else uses.
Someone else knows an exception that was never written down.
Humans are remarkably good at navigating that kind of mess. Machines aren’t.
Offshoring forced companies to make work legible
Now, think about what had to happen before you could send a process to a team thousands of miles away.
You had to write it down.
- Define the inputs.
- Define the outputs.
- Standardize the workflow.
- Reduce local variation.
- Create some way to measure whether the work was being done correctly.
In other words: You had to make the process legible to a stranger.
It turns out that’s pretty good preparation for making it legible to a machine.
A different way to build an AI roadmap
There’s a wonderful irony here.
Companies spent decades standardizing work so it could be performed more cheaply by humans somewhere else.
Without realizing it, they were also preparing that work for AI, which suggests a different way to build an AI roadmap.
Don’t start by asking, Where could we use AI?
Start by asking: Where have we already done the hard work required for automation?
The answer may be sitting in an outsourcing contract written fifteen years ago.
Because the hard part of automation often isn’t the model.
It’s turning five different ways of doing something into one.
Your offshore map may already show you where to start.
AI roadmap FAQ
What is an AI roadmap?
An AI roadmap is a plan for which processes a company will automate or augment with AI, in what order, and what has to be fixed in the underlying data and workflows first.
Where should a bank start its AI roadmap?
With standardized operational work, often the back-office processes it already outsourced or offshored. Those processes usually have documented inputs, outputs and quality measures.
Why do so many AI projects stall?
Usually not because of the model. They stall because the same task is done differently across offices, acquired teams and private spreadsheets, with exceptions nobody wrote down.
How can you tell if a process is ready for AI?
Ask whether a stranger could do it from the documentation alone. If the inputs, outputs, workflow and quality checks are defined, it’s a strong candidate.
