Personalization and Memory
Sam Lessin, 6 Dec 2017
One of the most important things a good assistant does is learn, remember, and properly apply the preferences of whomever they are helping. While this might sound intuitive, the difference between a good assistant who remembers some preferences, and an amazing assistant who deeply understands you and can creatively apply your preferences for you is enormous.
At Fin, we have been investing deeply in systems that allow us to do these things as well as (and sometimes much better than) any traditional assistant ever could.
Learning about you
“It’s easy to forget that when you ask for something, the person might not know all the things you know or make the same assumptions you make. You first need do the upfront work to explain your preferences in great detail.” — Andrew Kortina
The first week with any new assistant, whether human or virtual, requires a lot of investment. Having someone to support you is helpful, but if they don’t know you or your preferences, their ability to provide value is extremely limited. With a traditional assistant that has few prior clients, they may not know everything to ask right away and will spend a lot of time with every new task trying to understand you.
Intelligent applications like Fin can create detailed onboarding processes to learn about you upfront — everything from what people and places are important to you, to how you prioritize things, to preferred meeting times, and everything in between. With a larger client base, Fin understands all of the different pieces it needs to absorb right away. Not only that, Fin looks at interaction histories over time, continuing to grow knowledge about you. Training will never have to restart, unlike a human PA or EA, who may leave after a period, forcing you to hire and onboard a new one.
Remembering all your preferences
“We made some infrastructure investments in systems for remembering things, which almost seemed outlandish at the time. Over time, it’s become more obvious how important this is and it leads to our ability to do very good work.” — Sam Lessin
During and after the learning process, a lot of information is generated that an assistant must remember. Even the most trained humans can have trouble dealing with complicated context, despite using personal software programs. Really understanding the number of preferences for people, storing and structuring them, and learning from them (and from all sorts of entry points — email, calendar entries, interactions) is the really difficult part.
Having an augmented system allows Fin to go much deeper — remembering every important birthday, airplane seat number, meeting request, email exchange, and contact, and carrying it forward into the future. This is where Fin has made much of its technology investment, leading to a valuable memory system that every member of Fin — whether human or machine — can use to serve the customer base.
Recalling preferences to do work well
“When you mix machines and people, you can use humans for the judgment, which they’re good at, but then computer for the memory. Even for a human with a great memory, computer memory is still better.”
Once information has been stored in memory, it must be recalled quickly by an assistant in order to complete tasks in a high quality and efficient manner. Most human assistants can remember pieces of information that recur often, are very important, or have happened recently, but not necessarily distant or minute details.
It has taken us a while to figure out how to encode the information in Fin’s memory so that when you ask for something, the system itself is able to recall contacts, prior interactions, ideal workflows, etc. Internally, there is a balancing act to this — learning and storing as much as you can about a person, but then only presenting the necessary information for the task and nothing more, in order to reduce clutter and noise.
Externally, an assistant has to be able to recall information and either make intelligent decisions based on what is provided or ask for clarification, present options, or request confirmation. The key is to learn, remember, and recall enough preferences specific to a person to be able to do the former — take a task and complete it with no further communication needed. If you have to confirm or ask something obvious, you are just wasting a person’s time, or they may miss it completely.
With assistants, there is a risk-reward trade-off which requires a lot of judgment between known preferences that imply a set of outcomes or having to continuously ask questions to get to that set of outcomes. This is where Fin’s AAI (artificial artificial intelligence) really shines, as, with technology, there is so much to encode to aid with learning, remembering, and executing.
If you haven’t tried it yet, we encourage you to signup to get started with Fin and see how you can benefit from personalized on-demand assistance.