AI chat bot- Conversational Design
Lost and Found Bot- Seektron.AI
Problem
Multiple buildings on Campus
Multiple floors in each building.
Students visit many places in a week.
Lost and found items stored in multiple locations.
Solution
Centralized database of items found on campus
Chat bot that assists in locating the item - integrated on IU mobile
North Star — Number of items matched
Research
Competitive Analysis | Secondary Research | Qualitative Interviews
1.
28
items lost/month
2.
$178
Avg price/ item
3.
" Student Don’t always remember where and exactly when they lost the item, but they describe the item in detail ”
Methodology
Similar bot called — LOFO bot, Security and privacy barrier before presenting details of items found. Integration with university/ establishment database is our USP
Secondary Research using YouTube conversations
Interviewed Front desk workers to understand how a typical student who lost an item behaves. Role Played as a student who lost an item to understand how much help they receive.
AI is a collection of IF statements written on a massive scale
Prototyping - Voiceflow
Conversation design
We started here,
mapping out conversations…….Needed so much testing and iterations.
Tested, Iterated and improved ✨
1. Novice users
New users had no idea what jargon Seektron uses for entities like size/ color..
So we made sure they are guided when Seektron doesn't understand them.
2. Error prevention
Different users refer to each building by different name (SOIC,Luddy, IT building…)
40% of users failed a task with spelling errors and refereeing to a building with different names.
So we decided to include limited options to chose from.
3. Security
As mentioned earlier.. Market research revealed that a security barrier is crucial before providing any information related to item/location.
Crimson card/ ID details are requested before revealing crucial info and to keep track of information access.
4. Empathy
lastly lets be nice and keep users comfortable.
Most users were uncomfortable when we asked for specific location.
They might not remember the exact location. Might feel like a wrong input might alter the results.
So we replaced the questions to be less direct and more empathetic.
Impact 🚀
Chat to call Handoff Rate
40% reduction in call handoff rate compared to the first iteration increasing the efficiency of seektron.
Chat containment Rate
70% increase in item match found with the database. This improved the north star metric.
