Jarvis

AdRoll

Jarvis is an AI assistant for advertising account managers that you can talk to - Just like Tony' Stark's Jarvis.

Research

Before designing and developing the Jarvis AIĀ assistant I outlined all of the end goals and existing daily practices of account managers.
This included observing AMs work and documenting it as well as researching how other companies did account management.
I looked at common primary objectives like revenue and retention, secondary objectives like cross selling, solution adoption and LTV, and the processes/tools these companies used to get there in measurement (CRMs), data visualization, and day to day management like account pools, books of business, and agile frameworks.

Design

Function

Since Jarvis is a voice or text chat solution, the primary functional design elements were linguistic flexibility making sure to cover common intents and triggers as well as heavy research in proper chatbot error handling, guiding users to other intents instead of letting them run into a brick wall of "we couldn't understand you". Every error is guided to a possible intent with leading questions as well as a "contact us here" after a number of errors received.

Form

To make sure Jarvis was pleasurable to use I wanted to make sure to include natural language features like Hey instead of Hello and writing responses that embody a personality, making it feel like you are communicating with someone rather than something.

Development

Stack

To build Jarvis I explored a lot of different tools like IFTT, Zapier, and a few other chatbots but ultimately landed on an industry-standard from Google called Dialog Flow. The full-stack included Redash (to query the original data) Google Sheets (to better organize the data), DBsheets.io (to configure the sheet into a DB accessible via API), firebase (the other end of the API integrated with Dialogflow to connect the chatbot to the data) then finally Dialog flow to execute the intent recognition and response forms. And as always I leaned heavily on Google search and Stackoverflow for problem-solving. Languages used were SQL for the redash queries and JS for the fulfillment in Dialogflow.

Process

Go To Market

Ideal Customer Profile (ICP)

Advertising Account Managers who manage many accounts and want a single hub for their account information and daily management. (A specialized CRM)

Value Prop

The primary value proposition is that Jarvis can save a ton of time, get you better results for less effort (efficiency), and improve culture through enjoyment. All of these values improve attributes directly tied to money.

Benefits

Save time, save energy, higher quality answers, have fun

Features

Integration with nearly any chat form e.g. Amazon Alexa, google home, slack, etc. so it is flexible to many workflows. Quick and accurate answers without having to log into any dashboard

Delivery Method

As an internal tools, the GTM sales method may be influencers and decision-makers over the purchasing and adoption of new tools/processes rather than the end-user themselves. leaning into the values like time, money, energy, and insights into how their teams are requesting data

Time Goals

There are no specific time goals for this project.

Future Roadmap

Additional Features

I've played around with the idea of better structuring the data for AMs that would unlock better pivoting and requesting of groups of data. e.g. "Hey Jarvis, what is a good optimization plan for this account over the next 90 days" could be responded to if the relevant data was stored in a table that already was built to answer questions like that. Scheduled triggers like Jarvis generating reminders or emails and recommending workflows with prioritization and tasks. And some fun end user things like different voice packs to add character to their day.

Time Goals

There are no specific time goals for this project.

Follow me on