A forward deployed designer joins the customer's operation when an AI system goes in. The job is to redesign the steps around that system.
What a forward deployed designer does
The common failure is the 40 percent trap. Risk assessors using AI finish their part of a case in less time. Legal then reviews every case by hand. The customer gets an answer on the same day as before.
Designers at AI labs, defense tech companies, and enterprise startups do this work, along with anyone building a forward deployed team. Most of their customers are large bureaucracies.
Where the methods come from
The methods come from three places. Each was tested inside a national government.
Helsinki Design Lab started by defining the problem
Helsinki Design Lab worked inside Sitra, the Finnish Innovation Fund, from 2008 to 2013. It was one of the first design teams inside a national government. Every project started from the same question. We know there is a problem, but what is it? The lab's books, In Studio: Recipes for Systemic Change and Legible Practises, are online under a Creative Commons license.
The 18F guides taught teams to decide for themselves
18F was the U.S. General Services Administration's digital services team. Its Design Methods, UX, and De-risking guides were written so a team could use them alone. Teams used them hundreds of thousands of times in more than 200 countries.
Palantir built the deployment model
Palantir started sending senior people to work inside customer sites 20 years ago. They shipped in the first days, improved the work as people used it, and reported back to the product team. The largest AI companies now call this model their enterprise strategy. More than 100 startups are hiring for it.
Forward deployed design keeps two habits from Helsinki and 18F, stating the problem first and writing judgment down for the next person.
Where work gets lost
The deployment model splits the work between a person who finds the problem and a person who builds. Design teams call the move from one to the other a handoff, which is where work gets lost.
Why the customer's vocabulary is design work
An AI system built on a general model needs the words a customer uses for their own work. Putting those words into the system is called ontology work. It affects the outcome more than any other step.
Ontology work is content design, information architecture, and taxonomy. Enterprise software companies pay more to solve it than any other problem, often to people with no training in it.
How to test the design
Engineering teams write an evaluation, called an eval, before the work is done and run it throughout. Forward deployed design needs the same kind of test.
Three ways the practice fails
- It turns into a low-margin services firm whose field work stays with one customer.
- Staffed below senior level, it fails in ways that look like execution problems.
- The designer arrives after someone else has defined the problem.
What to learn first
Learn how product, engineering, and the people who fund the work make their decisions.
Sources
- Helsinki Design Lab and Sitra, In Studio: Recipes for Systemic Change, 2011