Ahmed Elsamadisi is the Founder & CEO of Narrator.ai, a data intelligence company that equips decision makers with personalized actionable insights.
Ahmed started his career at Cornell’s Autonomous Systems Laboratory focusing on human-robot interaction and Bayesian data fusion as well as building algorithms for autonomous cars.
What initially attracted you to AI and data science?
I fell in love with how people make decisions. Starting with psychology, to social engineering, and finally to how we reason about uncertainty. This led me to dive into Bayesian mathematics and the world started making more sense. I decided to embark on a journey to replicate how we make decisions.
You’ve had a phenomenal career including having worked at Cornell’s Autonomous Systems Laboratory, could you share some highlights from this time period?
Cornell’s ASL was a lot of fun! From our Autonomous Car to a fleet of mobile robots, I got to experience building, programming and testing algorithms and hardware in real settings. My favorite moment was a project I led to see if we can play a game of 20 Questions with all the students at Cornell. The game was simple, a robot is looking for an object and it can ask Cornell students yes/no questions to help it find the item. One tiny twist, humans can lie.
In this situation there is no real information, nothing that is absolutely true. I worked on an algorithm that could fuse information from people and sensors to make better decisions. This project later got picked up by Business Insider and got known as “the robot who can tell lies.”
These moments where data and algorithms can do something that you could not easily imagine, is what makes these projects phenomenal.
The idea for Narrator originated from your frustration of working with data at your previous employer WeWork. What were the issues you were facing with star schema modelling?
Every company uses a star schema for their data models. It makes sense! You build tables that represent a set of questions that you want to answer and then you give it to people to plot it. The challenge is that questions are constantly growing and changing and thus the series of tables you build are never enough to answer all possible questions. The only solution is to build more tables, which causes a deviation from the source of truth.
I always google “air flow data modeling” to show people what the best case scenario of a modeling layer is and it is complex. Hundreds of tables depending on each other.
That being said, that was the only option. At WeWork, I talked to many unicorns and saw the same situation that we were facing. We spent millions of dollars on data tools. We built hundreds of models. Data was still losing trust and failing to answer questions in a timely manner. Then every 1-2 years we would rebuild the system using the newest tools but the same approach.
If every company that implemented a star schema ended up needing to rebuild their system then there is an issue with the framework.
How did you come across a better solution to work with data?
Initially, I took inspiration from data blogs. In a blog, a company can tell a story of a customer and an analysis without ever showing us their data model. They use customers, doing actions in time to explain any analysis of algorithms. Effectively these are all concepts that everyone understands (a user viewed the website, then booked a meeting). I wondered why that structure wasn’t used in data if it seems to have the potential to represent anything. The short answer is this data structure is really not queryable by BI tools (there is nothing to JOIN on).
Our data model, which we call an activity schema, seemed to have the potential to really change the world. It could allow every company to have 1 single data model that can answer any question. Every company could have the same data model — thus analysis and algorithms could be shared across companies for the first time. And finally, it could create a common way for data people to work together and talk about data.
I thought this could revolutionize the data industry so I left WeWork in 2017 with the goal to make the activity schema queryable.
Narrator output data is formatted quite differently than most of us have come to expect, could you explain what Narratives are, and how they are formatted like a story?
Narratives are actionable analyses in a story format.
We, as data people, are so used to dashboards. However, dashboards are a means to an end. Dashboards show you data and you need to interpret it yourself:
- Figure out how to read the hundreds of different visualizations.
- Combine the data in your head to figure out what is actually happening.
- Create a story in your head that makes sense of the pieces of data.
- Decide on an action to take.
The challenge with this process is that every person who looks at a dashboard will come up with a different story and different recommendation. This is a natural bias.
The goal of Narrator was to drive action based on data. As we iterated we saw that our customers needed stories and interpretations to make decisions and that is what we gave them.
Narratives start with a clear goal. They make a recommendation. Then show you the key takeaways that you will get. Finally, they go through the analysis. Each section tells a story about something we are learning and data is used as supporting evidence.
Show 100 people the same Narrative and they all come back with the same recommendation and interpretation.
Narratives also understand that your business is changing so they are constantly run every week to see if the recommendations and interpretations are still valid and will update themselves accordingly.
After going through all this learning and making Narratives these incredible action-driving tools, we realized that most major consulting firms had also learned the same thing. McKinsey gives you similar presentations and approaches instead of just dashboards. We’re working to provide this level of quality for a fraction of the price.
What are some of the benefits of using Narrator for data engineers?
Data Engineers are our biggest advocates! I think it is because I started as a data engineer that Narrator was built with them in mind.
In Narrator, modeling data is really easy — you just map concepts from the source of truth to our data model. We run a bunch of tests to make sure it will work and then you push to production. Narrator handles the migration, syncing the data and ensuring everything is fast and easy.
Within Narrator, you can quickly assemble any dataset you want in minutes and give it to anyone requesting it.
We also thought about all the little things that you need and made sure you got it:
- Full transparency in all processing and control to pause, cancel, run_now, etc..
- Ability to put alerts on the raw data before and after the transformation.
- Full log of all changes to the query.
- Full visibility to every update that happens in the table.
- 1 click to materialize any dataset or send it to Google sheet.
- Setup quick webhooks to send the data to any system.
- Reach out to support and you will get a data engineer to help you.
- Every query Narrator generates is READABLE and transparent.
- Debug any data with quick timelines of everything that happened to a specific customer.
- Build 1 dataset and create many aggregations on top of it (no need to copy the same query in a CTE).
- If you swap out any of the building blocks, then we reconcile all your data for you.
There is so much more that data engineers will love!
How much time is saved doing reports with Narrator compared to traditional modelling?
Honestly comparing Narrator to traditional modelling is unfair. In Narrator, all your data is modelled in just 1 day instead of months of planning like in a star schema. Changes in Narrator are low-risk and easy. Creating any table is instant in Narrator because you don’t have to worry if there is a foreign key to tie things together.
In terms of Analyses, Narrator has a library of expert hand-written analyses that you can instantly run. These analyses are well thought out, tested on multiple companies, update based on your business changes, leverage algorithms to interpret the data and are beautifully presented in a story format. This level of work is often done by teams of data people over weeks (which is what we do, but we do it once and then make it instantly available to everyone).
Is there anything else that you would like to share about Narrator?
Narrator is different.
It is a bit hard to wrap your head around how and why it works but once you get started with it, it will click and you will never be able to go back.
I am excited by the world that Narrator is making possible and welcome anyone to reach out to learn more!
Together we can make better decisions that will create a better world.
Thank you for the great interview, readers who wish to learn more should visit Narrator.ai.
- Pexip Collaborating with NVIDIA to Create Immersive Video Meeting Experiences
- Sean Byrnes, Co-founder and CEO at Outlier – Interview Series
- Microsoft Buys Nuance For $19.7 billion
- Deep Neural Network Can Screen for Skin Disease on Laptop
- AI Systems Might Prefer Human Language Instead of Numerical Data