Decisions in the age of digitalization | Michael Feindt | TEDxKIT

Decisions shape our lives and our world, from love to career to everyday business operations. Can technology help us make better decisions today?

As founder and Chief Scientific Advisor, Prof. Dr. Michael Feindt is the mind behind Blue Yonder – one of the world’s leading companies for predictive applications. During his long years of research at the world’s biggest elementary particle accelerators, he developed the NeuroBayes algorithm, which now serves for purely data-driven forecasts of probabilities of future events in many areas, and therefore as a basis for the automation of operational decisions in the fields of purchasing, distribution, research and development, manufacturing, and finance.

This talk was given at a TEDx event using the TED conference format but independently organized by a local community. Learn more at http://ted.com/tedx

Okay, decisions in the age
of digitalization.

To make one thing very clear
at the very beginning:

Most important human decisions
can and will never be automated.

They are done by gut feeling,
under extreme uncertainty.

I'm not going to talk about these.

Don't blame me for that.

0% automation on decisions like,

these are examples from my life:
marrying my wife,

declining a tax-free,
permanent CERN job offer,

instead of a KIT professorship,

or founding Blue Yonder.

In retrospect, these were
all very good decisions,

but they weren't clear beforehand.

So I'm not talking about these things.

But I claim that 99% of all operational
decisions in enterprises

can be automated.

And they will be automated.

They will also be simultaneously improved
by quite some margin.

That is what I am going to report on.

And these are especially repeated,
regularly repeated, similar decisions.

Let's take an example from retail.

Assume you are a store manager
and you have to decide:

Of these articles,
how many do I have to order

so that I have enough tomorrow,

but not so many
that I have to throw them away

once the shelf life is passed?

Ten? Zero? 100?

That's a decision many people
have to make every day, again and again.

Or, again, you are store manager:

Do you want to decrease
the price of this product today? Yes?

By 5%? By 10%? By 20%?

That's a decision that people
have to make very often.

Or to send this expensive catalog
to that customer or to that customer?

Is it worthwhile? Will it be good for him?

Will it be good for us?

These are questions I'm talking about.

Don't you believe the 99%?

First, you have to know how most
common decisions are taken in real life.

And that's about something like that,

so 90% of these decisions are:
Do nothing, or do what we always do.

What people always have done,

even before I came here

and wanted to bring into this business
everything I have learned at university.

First, you are told, "No, we do it
this way - ever, always."

About 9% apply business rules
in one way or another.

But many of these business rules,
if you really look at them,

are not very good.

Almost none of them
has a real proof of value.

[For] only about 1% of decisions,

[does] somebody sit down,
use his brain, and think about it.

And even this 1% is far from optimal.

For this, you have to know

that the human decision-making
system has two systems.

The so-called system one
is fast and intuitive,

but it has many biases.

System number two is slow and rational,

but it is hardly used.

Why? Because it's work, it takes
a lot of energy to use it.

Most of the time we don't use it,
we use system one rules for everyday life.

And even, and especially, scientists
and experts use their "gut feeling."

They are not always thinking
and deciding rationally.

And it's very important to know,

system one, the fast one,
cannot speak statistics.

It cannot really judge risk and chance.

That's a big problem.

Daniel Kahneman, Nobel Prize winner,

has done research his whole life
about how we think.

His book, "Thinking Fast and Slow"
is really worth reading.

He describes all of this.

He describes how our system one
makes decisions,

and it is all explained by evolution.

So it was right: we are here now
because we did it that way.

But there are many, many effects,
for example, the IKEA effect:

We value something higher

when we have worked
for it ourselves a bit.

There are many
so called "cognitive biases."

And it's very funny to see

that somebody, like myself,
who thinks he is very rational,

has all these biases every day.

So, it's like that,
we can't do anything against it.

So, our question was:

How can we make more rational decisions
than we are actually doing?

The origin of this idea
comes from CERN and KIT,

so from where I do my research work.

The development is now done
at Blue Yonder.

The basis is really big data,
Bayesian statistics, machine learning,

data science, stochastic programming,
causality reconstruction.

In other words, it's really science,

how scientists work
in order to get insights from data,

and then optimize decisions on the basis
of these insights for the future.

It's a purely scientific
thinking and ethos,

that is also not true
in most enterprises, as I have learned.

We are trying to do
this technology transfer

to reinvent business processes
with scientific methods.

This is a key.

The key to better decisions
in the digitalization era is:

data and algorithms.

So, first, you need
relevant objective data.

So, it's mostly your data
or your company's data.

And you need good,
mathematically correct predictions

for near future events on the basis
of this data, including uncertainties.

That's a very important point.

And this is handled by an area
that is called "predictive analytics."

Then you also want to know, if you decide
this or that, what does it mean?

So you need cost functions,
utility functions:

What is the consequence
of this or that decision?

And then you want
to optimize this decision

using the prediction
and the utility function

to give a recipe for what you should do.

That is called "prescriptive analytics."

And finally, the King's discipline:

If you [have got this] far,
then you can automate it,

then it's completely automated.

When can we use predictive analytics?

In between purely random
processes, like the lottery,

and completely deterministic processes,
like a mechanical pendulum.

Most processes that we are interested in
are in the continuum between those.

So they are partly predictable,

but there's a lot of randomness
and uncertainty inside.

Most people think that the world
is very deterministic. That's not true.

There is some determinism inside,

and it's our work
to get this component out.

But there is always lots
of uncertainty inside,

and in evaluating the value of a decision,

we also need to take
into account the uncertainty.

That's a very important point.

So, that is what
predictive analytics can do.

Going back to our store manager
and his problem:

How many articles to order?

There may be many influencing factors

on how many articles
will be sold tomorrow,

from weather, weather forecast,
price, promotion,

day of the week, holidays,
seasonality, and so on.

But the store manager has
to make his decision based on,

or take his information from,
many individual events:

namely many articles, many locations,
many points in time.

And all together, there are many
such decisions to take.

The output of predictive analytics

should be something
which is not only a value -

it will never be 80 pieces,

because we don't know whether
80 pieces will be sold tomorrow -

but a distribution
of all possible futures:

79, 80, 81, 82, and so on.

The output is a probability
for each of the possible futures,

a probability distribution.

If you have that for each
of the single cases,

then you can calculate
what the value of a decision is.

That's prescriptive analytics.

Use the predictions to take decisions
on the basis of utility functions,

which are somehow known
or have to be found.

That is still a management decision:
What is a value of this or that?

We have to decide what is a value,
what we want to optimize.

Prescriptive applications give recipes,
prescriptions, for every single decision.

The recipe might be:
Take alternative two, buy 38 tomorrow.

For some very fine tasks,

the mathematical and technical complexity

of the whole process,
of prediction and decision,

can already be outsourced completely,

and you can buy it in some cases
as a solution available in the Cloud.

With predictive analytics, we very often
can attain two mutually exclusive goals.

That's very interesting,

because usually in management decisions,
you cannot do it.

For example, in purchase
order optimization,

you can have an increased availability
during the season,

and simultaneously reduce what remains
at the end of the season,

for example, in fashion retail.

Or you can reduce the out-of-stock rate
in your meat shelves,

but simultaneously,
you can reduce the meat waste

at the end of the shelf lifetime.

Or you can save 35%
of your marketing budget,

but still have more customers
to buy your products.

These are examples that are possible
with such technologies.

We have also automated
the work of scientists,

namely of my own science
as a particle physicist.

So we have seen that
usually a PhD student or researcher

has to do very similar things,

and once you understand it,
you can try to optimize all this.

We have found that usually
a PhD thesis on particle physics

can be done by 72 decisions,

and they all can be optimized.

This was done for one
of the most successful experiments

in experimental particle physics,
the Belle experiment in Japan.

All the data having been
collected over ten years

have been reprocessed with such a program,

and the output of it was twice as good

as 400 physicists have done,
by hand together, in ten years.

That was the work of machine learning
or artificial intelligence,

and three PhD students corresponds
to about 500 normal PhD theses,

and corresponds to another
ten years of data taking,

which would have cost 700 million euros.

Very often we have too many decisions
to be made by a human.

For example, 130,000 articles,
five productizes,

three storage locations, two predictions
40 days in advance, and so on to do,

and immediately you have some
100 million decisions.

so it's always good then to automate.

Because even if you have
100 million recipes,

who's going to read all the recipes
and then decide what to do?

That is where we see that automation
is the ultimate goal of what we want.

The impact of decision automation
can, for example, be seen here:

So, this is the out-of-stock rate
at a German supermarket chain

at constant overall stock level

as a function of time.

So, the original out-of-stock level
was about 7%,

and then when the project started,
you see that black line there,

it got less, that's what we want,
it's good if it's small,

and then suddenly, one day,
we really automated it,

and no humans were allowed
to do anything anymore.

Then you see, the result
was much better, and constantly better.

So that's the good or bad thing.

So, three examples at the end.

Automatic fresh-meat replenishment
for supermarket chains

does work very well,

and in some German supermarkets,
it is done automatically.

Also dynamic pricing:

in retail, it's something
that is appearing more frequently.

With the Internet, it's normal already.

It's also coming
in brick-and-mortar retail.

It's not bad even for the customer, right?

Because a retailer can never fight
against the customer,

the customer has to want
to come and to buy,

so it does not mean
that the price is always higher;

it just means the right price,
at the right time, for the right article.

So most retailers do not fight against
the customers, but against competitors,

and the competitors also use
methods like these nowadays.

So finally, another example from research,

from the collaboration
of Blue Yonder and KIT,

we have built such
a decision system onto one chip,

and this now holds
the world record in decisions,

it does eight billion
decisions per second,

and that is for the next generation
of particle accelerators.

So much data is produced in the sensors,
that we cannot read them out anymore.

So directly on the sensors,
there's a chip which decides:

"Okay, in this event,
this part of the response

seems to be important so we save it,

we put it in a computer,

and the others are not
even read out anymore."

So, these are things that are coming.

And here, it's very clear,
here it's simply time,

no human would have the time
to make the decision,

it has to be extremely fast.

What I'm talking about here is not
the far future, but it exists now,

and for sure, it will evolve further.

So, my prediction is that
quite some repetitive routine work,

but also white-collar work,

so it's not only the workers
whose jobs will change in the future,

but also, like me, white-collar work
will be automated.

Should we be afraid of it?

I think not, it is normal innovation,
it makes us stronger,

it makes our strategic decisions
more efficient and also sustainable.

I think that's the way
towards a better future,

and I am completely convinced
that it is unavoidable. It will come.

We don't have to discuss it,

it's only a question of time,
it will come, for sure.

Thank you very much.

(Applause)