People communicate differently, not controversial. Whether it’s the medium they choose or the way they write or speak, there are often dramatic variances. Turns out people are unique, go figure. Keeping that in mind, we looked at the way people were communicating in one of our mediums, in-app messaging.
Let’s say you’re going to have a business meeting with a potential partner. You sit down with that person and begin to discuss your business’s financials. You ask them if you can get their opinion on how to solve a certain problem. But for them to help you, they need to leave the room, go into a filing cabinet in the office next door, look at a document, and solve the problem there. And then come back. That is ridiculous, but it’s pretty close to how bookkeepers and clients used to talk.
Bench had a messaging feed with threaded conversations allowing bookkeepers to send links to a client. A client would click those links and go to a different page to leave comments on a transaction, letting a bookkeeper know what an unknown transaction was for.
Why should a client go to a completely different page of the app to fulfill a request from a bookkeeper? They shift contexts, forget instructions, and receive no feedback for their efforts. This is where Service Bubbles evolved from.
A Service Bubble is an enriched message that brings exactly what was needed from another page and puts it directly inline in the messaging feed. The client never needs to leave the messaging feed. Workflows pop-up over the feed and give direct feedback within the feed on progress and state. One place of truth for communication.
Service bubbles were the end state of the process. We began right from the foundation and I met with members of our internal bookkeeping team to walk me through the current state of communication and get their advice on where they’ve identified problems.
The bookkeepers speak to clients every day, so they were a great resource for summarizing a broad set of experiences.
After taking in all that information I worked on synthesizing what we learned into different areas of oppurtunity and mapped out the existing experience to get our product team onboard.
I setup sessions with engineers and product managers to work on ways in which we could better represent these tasks and workflows. The logical conclusion from our observations was to bring what was being discussed in the messaging feed, into the messaging feed.
While Bench does its best to take away as much bookkeeping as possible, we do still require client input for certain tasks. When a bookkeeper is not certain what a transaction is for, they’ll message a client and ask them for help identifying it. This is a categorization workflow. We identified categorization as the biggest oppurtunity to affect both the client experience and the speed in which bookkeepers could get answers to their categorization questions.
Of all comments related to transactions, 99% of comments were related to categorization. I took what I had initially worked on with engineering and product and started to map out a workflow for categorization.
A service bubble would launch the categorization workflow, and if the client decided to leave the workflow part way, the bubble would represent the state of the workflow.
The categorization workflow was about improving the process of having a client comment on what an unknown transaction was for. Rather than sending them to another page to go track down and leave a comment, we decided to overlay the workflow directly on top of the messaging feed. This meant they could quickly jump back into the messaging feed, but still re-open the workflow and jump back without losing progress. Our previous representations of transactions were in a table format, which didn’t afford many opportunities for quick scan and recognition.
I worked on developing a UI that would allow for as few moving parts while quickly going through as many transactions as possible. The only elements of the UI that would change would be the elements used to quickly recognize and categorize a transaction. We used a card metaphor to wrap text to allow for efficient space usage and quick scanning. The results turned out well and clients enjoyed the process so much they finished work exceptionally faster than the previous process.
I worked on designing a way of phasing the categorization workflow into three steps: commenting, suggestions, and auto suggestions. The first step which included only the ability to add comments, went very well. Clients responded much faster to bookkeepers, sometimes completing as many as 100 transactions in a single session, which would have taken multiple requests over months to complete with the previous workflow.
We’re currently on phase two in which the bookkeeper suggests categories that they think a transaction is, and the client just has to pick one. Clients really enjoy this because it makes it much easier to go through many items, and there is less need for writing out descriptions.
The final phase of the categorization workflow is auto suggestions in which the bookkeeper just marks what they don’t know, then Bench uses machine learning to guess what it thinks it is. If it’s not sure what the transaction is for, it automatically gives the client a few suggestions. As Bench records more input, fewer categorization requests are needed. This is the ultimate goal, no client work needed.
This general process will continue as we develop more service bubbles and more workflows. We’re continuing to move things of value into the feed to smoothen out conversations and build out a better way of communicating.