Insider Update: 10/24
Version 0.1 of Babbage Insight is ready, and we have started selling it as a service. If you, or someone you know, wants to check out, please let us know, and we'll schedule a demo.
This is an email we sent out to “friends and family” of Babbage Insight. Recording it here for posterity, and for possibly better sharing.
TL;DR
If you (or someone you know) wants to explore it, please let us know via email, or schedule a demo with us!
Actually - here are some quick screenshots to show you what we do.
A quick “demo”
You initialise our system with the metrics you are tracking, along with SQL statements to generate them. You also need to give us some more metadata. Something like this:
In a preprocessing step, our system takes your metrics, understands your data schema and then generates some auxiliary metrics.
And then, at the defined frequency, the algorithm runs, finding exceptional insights in your metrics (including the auxiliary metrics). And produces reports like this:
This whole thing was automatically produced, from one simple run of a script! The first line here is the “ticker”, which shows the metric (right now we’re in a very initial version where there is only one main metric to track - we’re fixing this very soon), put in context (too high / too low / trend / forecast / etc. ).
Below that you see the key “stories”, that could be either of the main metric or any of the sub-metrics. Again this is still very early days (and you might recognise the graphs as being generated using ggplot2 ) but notice that we have a headline there talking about the key insights. The key regions of the data are all marked out (we need to improve our LLM game to improve these annotations, of course).
Let’s look at a report for another day:
This is a synthetic data set (downloaded from Kaggle) that we’re reporting here, so there isn’t that much insight in the data. But you can see that the ordering of graphs is different - because the significance of insights are different.
Here is another day’s report:
You notice that these reports talk about what is exceptional. Soon they will also have the why (this is work in progress).
Actually - a part of the “why” has been tangentially covered already, since we have shown exceptional insights in the auxiliary metrics as well, but we are working on formalising this, and also finding other ways of explaining why some things are the way they are (using correlations, unstructured data, etc.).
What do you think about this? And once again, do you think you or someone you know can use this in your job? Give us a shout!
Selling
Over the last four months, we’ve been building what you see above. And in the last couple of months, we’ve started selling it.
Some of our “insights” so far from the conversations we’ve had with potential clients:
When we say “AI data analyst”, people immediately think “copilot” / “Q&A system”. We have been spending more time than we want to telling people that this is NOT what we are building. I guess we need to change our positioning. Or maybe now that our demo is ready, lead with it
We are explicitly NOT doing free POCs - if the customer is not spending money, it doesn’t really prove any concept to us. This no doubt introduces friction in the sales process (and some of our conversations have ended because of this), but this is necessary so we don’t unnecessarily slip later on
We had initially thought we’ll target companies of annual revenues of ~$10-100M range. Based on our conversations over the last couple of months, we believe that our ideal customer is likely to be larger - of the $100-250M range (₹500-5000Cr range for Indian companies). So we need to rethink our sales strategy (assuming we had already thought of it)
In general, we’ll probably be better off selling to companies with smaller data teams
Any insights on navigating the above would me much appreciated!
Thank you!
Happy Deepavali / Halloween to those who are celebrating!
Karthik & Manu
Co-Founders
Babbage Insight
congrats Karthik on the launch