I spent approximately two years embedded at a large energy company in the Midwest USA.
As part of a scrum team with a product manager, a team of 3-5 developers, and two primary client-side design and management stakeholders, we worked on four key use cases to build automation tools to reduce call center load.
All of these bots are voice-only, and are integrated into an existing phone IVR system (press 3 for Billing, etc) at key moments, replacing an option where previously the customer could speak with a representative.
1. 'Move Out' Virtual assistant
Every year, the energy company 'EC' receives many thousands of calls from customers who are moving out of their address and would like to stop paying the utility bill.
We started by reviewing the requirements that EC had assembled and analyzed several flows that described their existing process. We talked to three customer service representatives and interviewed them to gain a deeper understanding of the challenges that arise during these Move Out conversations, and started to map out the current-state Journey Map. Then, we evaluated how to adapt that process so that a bot could guide a customer through a similar experience, looking for areas to verify instead of input, reduce data entry by calling data from the customer's account, and save customers effort in key moments. We made a Journey Map that reflected the Move Out flow as adapted to be effective as a virtual assistant. At this point we did informal internal testing to get a sense of how it reads and further 'Wizard of Oz' role play testing with the Customer Service Representatives.
We continued to iterate at this step, considering the order of events, refining what the ideal order is for the bot to present the questions to a customer so that the interaction can be as logical, easy, and effortless as possible. We had a question arise about whether to have the customer listen to a series of advisements in the middle or at the end of the call, and we suggested grouping similar actions and thinking of the conversation as having four thematic units, which helped us to make an effective decision and establish a norm for how conversations will be handled in the future. This is an image I created from an email I sent to the executive team.
Ultimately, we fleshed out the full conversation flows and ended up with a very large Lucidchart document. Since this was the first chatbot that we worked on as a team, we had some 'Forming' and 'Norming' growing pains which led to the 'Storming' that we then carried out for the next year and a half or so. This is an example of the 'Norming' that needed to happen: I created a structure to document the conversation flow that was thorough and logical enough to read, but not efficient enough in how it was arranged, and it ended up being absolutely gigantic and required a lot of tedious upkeep if any changes were made. I learned from this delivery methodology and reformed it in the next bot, leading to a standard deliverable format with three key artifacts: One or several full conversation flow journey maps, one or several interaction maps, and a spreadsheet with full conversation dialogues.
Bot 2: High Bill
Every year, the energy company EC gets many thousands of calls where a customer is surprised to see their bill being higher than expected, often convinced that their meter is faulty or that something is wrong. We were hired to build a virtual assistant that would provide some information to help the customer understand their bill, explain common factors influencing their bill, and to pull in specific data points to help them understand whether there is something wrong or how their bill could have ended up the way it did.
I am reserving this project for live discussion, but here are two of the three primary deliverable output documents: First, the Happy Path Conversation Flow Journey Map, which is the primary artifact we used to visualize and plan out the experience, and the Interaction Map, which we made late in the design process primarily as documentation to help our development team understand and implement the full conversation flow map, which can get complicated to read through and see all options exhaustively.
Bot 3: Promise to Pay
We were asked to create a bot to automate the company's last effort to ask customers to pay overdue bills. This process takes 21 days and afterwords, the company would sell the contract to a collections agency.
I am reserving this project for live discussion, but here is sample of the third of three important final deliverables, the spreadsheet with the full conversation dialogues.
Bot 4: Move In
To round out our fourth bot as a team, we were asked to create a bot to automate the Move-In process for customers.
I am reserving this project for live discussion!
I hope these work samples help to illustrate the work I have been involved in and process we have followed to reach innovative solutions for our clients.