Team
-
1 x Product manager
-
1 x Product designer
-
2 x Front-end engineers
Role
-
Research
-
Lead Designer
-
Design System
The problem
Creating skills for your framework is a very time consuming and difficult task, which requires alignment across multiple teams and input from experts in appropriate disciplines. It's important to make sure the skill covers all the correct requirements and is written accurately with clear expectations and actionable examples in the right tone of voice, at every level of the skill.
A skill can be relevant to every team in an organisation such as Communication but may require unique examples of how that skill could be used for each individual team to make them relevant. Other skills might be completely unique to a certain team, writing code or visual design for example.
The Progression library has a lot of prebuilt skills to get customers started but the chances of that skill being usable straight off the shelf is very slim and will always require a certain amount of editing. So the same problems are not entirely solved by this approach.
A third option between writing a skill from scratch and using the library is using our AI skill generator. This is a basic form which will use ChatGPT to create a skill for you. This has proved very popular with customers but the accuracy of the results is a little hit and miss and it's quite hidden away at the moment.
Goal
We wanted to help customers create skills from scratch in a much easier way. Making them unique and relevant from the start but doing a lot of the heavy lifting around writing examples and scaling the skill to different levels. We want to leverage the current AI skill generator and ChatGPT to help.
If quicker for customers to write skills and scale them without losing any quality or relevance, the path to value will be clearer. Organisations will be able to complete their teams quicker and then Progression will hopefully be adopted by other teams, increasing subscription value.
Ideation and Implementation
I started to create some very quick prototypes of how this flow would work. I wanted there to be only one way to create a skill and incorporate the AI elements into this empty skill, giving customers the option to generate some parts of the skill and write others themselves.
I put together a few simple patterns slotting them into existing skills modals and flows. When you first opened an empty skill we presented the option to generate the whole skill or just start from scratch and generate each individual section. Regardless of which option you could always regenerate the information at any time. I worked with engineers on prompts and polishing these flows in real time. We adopted a very iterative process and the main challenge was getting quality outputs from prompts.
We worked through a few iterations and leaned heavily on other tools which had started to incorporate similar AI patterns to edit forms inline. As we wanted to work quickly get something out there it seemed reasonable not to stray too far from this at the moment.
Designing for AI is an interesting challenge as it's a new way of working and there aren't any established patterns to work with. Most tools using it at the moment tend to use it in a similar way, so it was a experience to take on the unique UX challenges that come with it, and how these challenges differ from Progression and other tools.
The quality of the output has to be worth the work in order for the skill builder to be successful. It has to be quicker than writing it yourself or editing an existing skill from the library but if the output isn’t getting you 90% of the way there you have the exact same problems as before.
As with the original skill builder we found the outputs to be somewhat hit and miss with some skills feeling accurate and useful and other times the examples produced either too specific or too general.
Another issue was that examples and descriptions didn’t span all skills levels. The description and examples in level 2 should be an escalation of the ones in level 1 and not entirely isolated requirements. Again, sometimes this worked well and for other skills not as much. We had to work on the prompts so they took into account content which had already been added. This was tricky but ultimately paid off, with the quality of the output increasing.
Things such as company tone of voice are hard to add to a simple prompt but we hoped to explore this type of concept at a later date. With the idea being you could create a tone of voice style guide which would feed all AI inputs in the app.
Impact
Shortly after launch there was steady increase of people who started to use the AI features to help build their skills. Examples and the skill description being the most popular generated parts of a skill, both showing an increase of roughly 100%.
"This has saved me about 2 weeks worth of work!"
It proved the power of AI to assist in complex and long winded processes. This got everyone in the team really excited at what we could use it for next and how we could expand on this type of concept.
