Tip:
Highlight text to annotate it
X
Raise your hand if you do not agree
that adding data to your world view
enhances your ability to understand a problem...
excellent...
I'm delighted to join you all this evening.
I'm a recent transplant to the Bay Area
I was dragged kicking and screaming out of Chicago
I don't get a lot of sympathy for that because apparently
I've landed in a very nice place!
I lead an R&D team at Accenture Technology Lab
So it's fun to be here at SAP Labs
We're sort of Accenture's version (of SAP Labs)
Which is the R&D Lab whose primary goal
Is to look at technologies that are anywhere from a year or two,
to maybe three or five years out from widespread adoption and use
by ourselves and by our clients
I started out as an analytics person seeking to understand and
and identify patterns and trends and activities in large data sets
working on SaaS and the mainframe
If you were the DBA at my client
you'd see me coming and either run the other way or start throwing things at me
After that I went on to SPSS, which is an analytical software company
where I spent a couple of years working on their data mining workbench
where I got deeper into the algorithm side of the business
and started to understand the types of techniques and approaches we could take
to make those things scale over large relational data sets
so we started looking for technologies that would unlock the value of that data
in a way that was useful and didn't necessarily allow another terra-data salesman to retire.
Nothing against terra-data, but if you try to use it for your meter data
you will allow another terra-data salesman to retire
the last couple of years we've been looking at a lot of these technologies
and seeking to understand; you may be aware that Accenture has a fairly healthy
data management and BI practice; how do these technologies
impact that world?
and what possibilities to they open up?
I don't know if you can see it so well in the light but that is a very
tattooed arm and it says unstructured down here
and that's a nice elegant hand, and the tattooed guy is
putting the wedding ring on the structured data's hand.
How do we bring those structured and unstructured worlds together right?
If you're obsessing about methods
and trying to out if a neural net or a support vector machine
is going to better predict 'x', you're trying to hop lines.
Now, most man-down solutions typically focus on that
one point of data and some heuristics around how long people stay put where...
they don't tend to bring in the knowledge of the process that is executing or a
holistic view about how that process is executed over history...
What's interesting about this to me is that,
the debate has been about structured versus unstructured,
or sequel versus no-sequel.
the debate should be about geometric versus scale-out
Scale-up versus scale-out
Scale-out lets you be... encourage use
it lets you seek analytically driven insight that brings value to your organization
and to every corner of it.
in a way that's economical and valuable.
This, I think, really does change everything from a data management point-of-view
It's a very valuable thing to be able to integrate these worlds
and as BI Practitioners and as Data-Management Professionals
I'm sure you all, like I have, have spent lots of time
integrating data sources to do analytics to be able to make observations like this, right?
We got our start building accounting systems
and, for sure, if somebody gives me twenty bucks and I give them a beverage
or we have a business exchange, its probably pretty important that you and I agree
that I gave you twenty bucks and you gave me a cheeseburger or something right?
So, this really, I think, points to a fundamental change
in how we manage data.
So if we agree that we should, and that we've got some raw capabilities
that make it economical to do so,
that doesn't actually solve all of our challenges that we need to solve
to realize the promise here
and actually drive data-driven decisions making further and further into our enterprises.
What we'd really like is
is the ability to put something in the middle to marshall data around our organization
in a way that is efficient
If we introduce a counter-veiling force into the organization
In the form of a Chief Data Officer
which shares the unfortunate CDO acronym with
Collateralized Debt Obligation...
...but, getting that organizational power and interest
in data as a strategic asset
so you're not just thinking of it as esoterics
one of the forces that we suggest clients introduce
when their CIO organization is one of those very very costly ones
Some of them are more business oriented and its not really necessary
so its not a blanket prescription by any means,
but having that counter-veiling interest in preserving as much strategic value in data
as you can is an important thing, I think, for organizations to have.
Data is becoming one of the most strategic assets an organization can build up.
I mean, if you look at Netflix
Obviously they've learned a whole lot about how people interact and think about movies
And even if we were able to blueprint and copy their entire IT infrastructure
With no data, we'd still be a long way behind right?
If I am running the distribution center and I am worried about things like outages...
and the current electrical state of the grid, chances are my queries are going to span a lot of devices
asking for their instantaneous status, right?
A very wide and shallow query if you will.
If on the other hand I am more interested in predicting a customers next fifteen minutes
of consumption
now I want to look at long histories right?
a narrow and deep query pattern
And if I've optimized my data store for one, I've probably just de-optimized it for the other right?
Ideally, we would like to abstract data from its storage for the use of applications
which is to say, its the age-old promise of SOA, except this time we really mean it.
Bezos famously said at Amazon, "Thou shalt not build a point-to-point interface"
"Thou shalt not build an application specific data-store."
"You will build data platforms and expose that data as services"
"and if anybody wants to touch it, they'll do it with a service call."
And that's how they manage the complexity of having, literally, 500 different data products
everytime you pull up a page on Amazon.
It's crazy how many service calls are actually constructed by that...
The problem is, that...
If I want to put my data to incremental use
Another use in a traditional, relational data-based system
that scales geometrically
my last terabyte of processing and storage
is going to be geometrically more expensive than my first, right?
If we automate these we can get to that state where we are able
to take a posture where we are constantly evaluating data.
Figuring out which data actually adds to our ability to analyze the situation
and make a better decision.
And then bringing that insight through to that decision process.
We need to re-think our approach to data-management
and reporting in the context of this value chain
given the capabilities that we now have.
You literally can take a charge card and spin up a Hadoop cluster on Amazon
relatively cheaply and prove something to your boss
before your boss is any the wise that you've done any of this
And I'm not just saying this because my team might have done it to me before
It's not just for the fortune four-hundred companies that Accenture builds with
It really can - and you see it happening today - startups managing
really huge data loads profitably... because of these technologies.
The thought I'd like to leave you all with is back to my familiar story of making fun of the DBA
And sort of blowing that out all the way up to the CIO
and saying this: You know a lot of organizations these days, the IT department is a barrier
to unlocking the value of data
The formalisms that allow us to manage financial transactions don't need to apply to every last drop of data in an organization
Otherwise you'll have a data warehouse where good data goes to die