15. My objective:
Predict the likelihood of reaching a revenue
target in each market
Desired solution properties:
1. Interpretable
2. Stand alone prototype
3. Get better with the data being collected
16. Project take-aways:
1. Probabilistic method of
estimating revenue
2. Gets better with the data Room40
is collecting
3. Prototype is driving sales of
business analytics platform
17. Plug in any revenue target to get the
probability of hitting it
Room40 came to me asking for help to make tools for non-profit business analytics
Quickly scoped from talking to to them and from research that the problem was
Built tool to help non profits from squandering millions of dollars. Mitigate risk
CEOs of major non-profits have to make big decisions with a lot of money at stake: like corportate business one of the biggest being growth
Unlike the corporate world where there is a lot of intelligence you can buy to help make sure you get your decision right, the NPO
I’ve been working with to help build a dashboard to help CEOs
Add insight logo
Add plant to text
Switch my name and room40 group around
Add predictive mdoel
They want to be everywhere – a lot of decision go into this
But biggest thing is to hit revenue.
Really what they need is some sort of revenue risk model
Where can you put the next Bell so as to service kids but make responsible financial decisions?
Make map faded yellow
Consider switching bell to Y and make joke how there are Y’s everywhere
Where can you put the next Bell so as to service kids but make responsible financial decisions?
Make map faded yellow
Consider switching bell to Y and make joke how there are Y’s everywhere
Census information has limited usefulness as a first pass which I can explain after. Need org specific metrics for census information to be useful
Build distribution for each market, well have something that is immediately useful.
Add west coast to right
Deep dive into how the suasage gets made? Think it is too imprecise
Human in the loop
7-10 what they are, sin’t important, what is important is that the human is make the decision, not the alorigthm and this is a constraint I need to be wary of.
Revenue is a heavy component. We want to help with that
Give slide a nice bakcground color
Include icons
List constraint on interpretability
Can I make this a negative space slide?
White splotch with background color?
Make a fun colour
Census information has limited usefulness as a first pass which I can explain after. Need org specific metrics for census information to be useful
Build distribution for each market, well have something that is immediately useful.
Add west coast to right
Deep dive into how the suasage gets made? Think it is too imprecise
Human in the loop
7-10 what they are, sin’t important, what is important is that the human is make the decision, not the alorigthm and this is a constraint I need to be wary of.
Revenue is a heavy component. We want to help with that
Give slide a nice bakcground color
Include icons
List constraint on interpretability
Can I make this a negative space slide?
White splotch with background color?
Make a fun colour
Make this the whole pipeline
Make opaque bars and thicker line
Cascade the the hists and random distributions to depict different markets
Add big picture next slide
Make errors appear on a slide by slide scale
Get shaded region under graph
Emphasize that model will get better and better with more data which fits into their current model of data aggregation. Model is easily interpretable.
Real power is that this never existed before. Company immediately incorporated my predictions into their model.
Include a fun pic from each of these places
Add mcgill, NYU and Columbia names
Add tagline about human movement: important because we don’t know how to make cylons, but more importantly because we don’t know how to make people better after stroke and CP
Tackled this issue at a wide range of places
Add subheading for description, encoding, and causal inference
Make this figure look good
Change to logos
Be prepared to explain itemized giving
Good but won’t improve as we get richer data
Make this figure look good
Change to logos
Be prepared to explain itemized giving
Good but won’t improve as we get richer data
They want to be everywhere – a lot of decision go into this
But biggest thing is to hit revenue.
They want to be everywhere – a lot of decision go into this
But biggest thing is to hit revenue.
Real power is that this never existed before. Company immediately incorporated my predictions into their model.