In this video, we're interested in how you take

an existing product and you make it better and better.

Increasingly, one of the big advantages of an existing product over

its newer competitors is that it has lots and lots of

data about what users do with the product,

how the product performs for the user,

and they are able to use that data to make

the product and the overall user experience better and better.

And I think that there's three things that you should bear in

mind as a product manager with regard to data science.

One is how you instrument data collection into the way that you

build the product and the way the interfaces you have with the customer create data,

so for support or when they call in.

It's really important to make sure that you're creating a data,

can't do data science without data.

Two, is just like you need to bring

your development team strong inputs about here's our customer,

here's what we see them doing out there,

and here's the qualitative and quantitative data we have at what they do.

You want to bring those same strong narratives to

your data scientists because that will help them form

strong hypotheses and focal points about

what avenues they pursue with the data science they're going to do with you.

And then three, your data science

will help you with this but zero in on questions of interest,

what will we most like to know about the user or be

able to do for them because that's how you will literably

create a rich interface with

your data science team or whatever data science resource that you have.

So, there's a kind of general view of what are the basic jobs of data science.

There's four of them. The first one is it can give you descriptive outputs.

So what are these are clicked?

And who was that user that clicked that thing?

And how hot is this piece of equipment in the field right now?

And how fast is it working?

Or where is our truck,

or where is our Uber driver?

These are descriptive things.

And the second level,

these get more abstract and more intricate as we go along.

The second level is diagnostics,

so who clicked what?

And under what conditions does this part tend to break?

Or it was working really

fast and it's been working really fast at 3:00 PM for the last 200 days,

all of a sudden, it stopped, so maybe it's broken or not working.

Where does this car go over the course of the day?

And then predictive is when

we are able to look at patterns and predict what's going to happen.

So we say, if we'd have a bunch

of this type of users on this site where we're going to start posting ads and we know X,

Y, and Z about them,

can we predict now what type of that they're most likely to click on.

Or if this piece of equipment is working for a certain amount of

time and it's got a certain age and it's got a certain heat or whatever is relevant,

when do we think it's going to break?

Or where will this truck be at 4:00 PM today or tomorrow or whatever?

And then prescriptive is the most refined and

it's where just the machine intelligence could just tell us what to do basically.

So, what edge should we run for this user?

Just pick one. Or when should we go and replace this part?

Based on all the cost and predictions about when it's going to break.

And, is this calling the right route or should it take a different route?

Based on a tree fell over the road or something like that.

Let's loop through these for the example of cooped up LLC.

We'll look at this for their H1 business where they're selling

industrial feed and watering systems to existing factory farms.

An example of a descriptive application of

data science would be that they have sensors instrumented into the feed and

wandering equipment and they're able to tell how fast it's working or how hot it

is or how many parts

per million of antibiotics are running through the water feed system.

A diagnostic would be,

it looks like this part is broken because we know that

all the other things are doing X and Y

and it looks like it's broken because it's doing this unusual thing.

Predictive would be we think that based on all these factors that we know about,

this part is going to break at 3:30 PM two weeks from now and we have we predicted that.

And then that might be valuable because then we could then prescribe to them,

hey, in the next week,

we really need to come and replace this part so that it doesn't break.

And wouldn't that be great because they could avoid downtime

and having to move chickens around things like that.

So those are some examples.

My recommendation is if you want to learn about data science, go learn about it.

It's so fascinating and there's a lot of

great e-learning platforms or

programs at regular universities where you can learn about it.

As a product manager,

it's good to think about the kind of things that it might be

able to do for you and how you get that data,

how you present the right narratives to

your data science resources and then how create a rich interface with them

where you can start to make your existing product better

based on the data you're able to collect from your users.

Applying Data Science

Digital Product Management: Modern Fundamentals

Rencontrer les enseignants