Welcome! What an honor to kick this off,
my colleague Pan Shen is at the table here where the reps from HAVI, here, and
we’ve, as John just mentioned, we’re honored to be sponsors for this user
conference. Unfortunately our counterpart at McDonald’s Mike Kramer was not able
to attend. He’s a last-second cancellation here but we’d really like
to tell a joint story of a partnership and business success story for a really
challenging promotion that reached into 14,000 restaurants at a very real and
challenging way. And I hope you learn a lot through my talk but if there’s one takeaway that I would hold on to if I were you it would be this: we live in a
really exciting time for simulation, do we not? Simulation has been around for 40
50 plus years but it doesn’t really look the same, does it? There are things like
Apache spark out there that everyone’s all of a sudden using. We’ve got predictive
analytics, and I I see those dovetailing into simulation, and it makes simulation
a more exciting, powerful, useful platform for all of us. So, with that, I’d like to… I’ll just talk
real briefly about HAVI, a little bit about who we are.
You’ll learn a little bit more at the breakfast tomorrow, we are a globally
based privately held company. Had anyone heard of HAVI before this morning? Okay
one, two, three, four. It’s actually more than my models predicted.
We.. we are the probably the largest least heard of five billion dollar global
company. We’re just kind of obscure and there’s a good reason for that. We grew up as
sort of the silent partners, supply chain and packaging partners, to McDonald’s. Way back in the day Ray Kroc and our founders partnered together in a
distribution and packaging service. So, today we do quite a bit of supply chain
management, we do packaging services under the supply chain analytics, we have
an analytics practice that’s where I reside. We won’t go too much more into this
you’ll learn a little bit more tomorrow as I mentioned. Let’s talk about
McDonald’s though. Normally, if I’m going to be talking about an innovative
fortune 500 company we’re the first ones that come to mind… Apple Google Facebook…
but I’d like to argue that McDonald’s is actually quite innovative in their space.
You may not see the same pace of innovation but they’ve actually been a
market leader in quick service restaurants. So you think back to Happy
Meal. It seems like every restaurant you walk into now you can get a toy for your
kids. A lot of that was due to the… the pioneering of McDonald’s. In the early
2000s they introduced this notion of having a.. almost full service coffee
substation within your restaurant, that was pretty new at the time, the focus on
value menu, and then today we’ll talk quite a bit about a little thing called
all-day breakfast. That was launched about this time last year actually, it
was the number one most requested customer feedback. They said, you know what, you… I like what you guys are doing but I really want to be able to have breakfast –
whenever I want it, and they said …ok Let’s go for it, let’s try it. And that
was a really challenging problem, we’ll talk about that in a little bit. Through the years McDonald’s has also grown tremendously, but also spurred a lot of
growth in the restaurant quick service industry and we like to think it’s no
small coincidence that HAVI has been a partner since pretty much day one in
that journey. So let’s think through some of the
complexities facing McDonald’s or another chain of comparable size when…
when you want to make one of these innovations, right!
What about it’s something as simple as a new shake flavor. That’s given the fact
that they have… there are 14,000 stores and the supply chain underpinnings to
support that. …that is not a small undertaking, right? So I’ll give you a
concrete example. A few years ago you may remember a promotion called
mighty wings. Anyone try that by the way? Their chicken wings at McDonald’s and
you think okay that’s a pretty standard product. It actually took 12 months lead
time to work with the chicken suppliers just to ensure that there are enough
chicken wings in the United States to be able to run this promotion. So, hey we’d
like to run this promotion, you know let’s plan it, oh, you’re gonna have to
wait a year, probably a year and a half to do it. So, it’s just one anecdote.
Another example is… with the launch of all-day breakfast last year… What do you
think the most needed ingredient was there? Bacon, good guess.
I saw some on the table over there… Eggs! Just about every sandwich has an egg in it. Unfortunate timing because there was an
avian flu outbreak last year. So your supplier capacity goes to a pittance of
what it was, right? So these are just some of the challenges on top of that you’ve
got along with the scale you’ve got regional preferences for example the
mighty wings did very well in some regions and not so on others uncertainty
the mighty wings was a brand new product we it’s very uncertain demand difficult
to forecast okay so let’s forget promotions what if
we have McDonald’s wants to introduce a whole new platform are very fundamental
change to its business and every sufficiently large company goes through
this I think I saw that Walmart is in attendance here and whatever you want to
bring a system-wide change to your company that that requires so much
lead-time so much planning decision making all that and so just to give you
a concrete example when McDonald’s introduced McCafe line in 2001 that was
a whole new foray into a business that trip was going to pay off or not right
one of the problems with that is you’re purchasing very expensive equipment you
have a lot of regional variation your demand so some store operators were
ecstatic about that others not as much they weren’t sure they were going to get
the return on their investment and so we’ll take a dive down that path in a
little bit and see what is what do we recommend for getting an optimal return
on some of those capital expenditures like equipment additional I should note
that the complexity of the stores there’s 14,000 of them they actually
refer to them internally as snowflakes because each one is a little bit
different from the next they might have the same footprint and actually a lot of
them do not have the same footprint but they have different equipment they have
different demand patterns regional preferences so it makes the work of
modelers like you and I actually pretty difficult so either you can try to
simulate the whole universe or you have to sample on cluster intelligently okay I talked about a lot of the
challenges and problems right well this is where we come in and hobby provides a
lot of planning and forecasting solutions that’s the bedrock of our
supply chain management offerings as well as supply chain and network design
so you got to be able to contain your costs in your supply chain and we offer
services around that we also do quite a bit of marketing and promotional
analytics and these all feed and you’ll see in a
second they fed into the simulation studies that we do you may not
necessarily agree that you need all of these but I can assure you they will
help enhance the quality the results and the savings of your simulation models okay we have you believe that it’s not
enough just to have the software such as an analogic or a tableau it’s not enough
just to have even to have the right people that’s challenging to write
getting good analysts in the door getting good modelers but it requires a
disciplined framework as well so I’ll just walk you through the the framework
that we use this is applicable to simulation but it’s also I think more
broadly think about any data modelling and analytics exercise that you could
undertake so this is the process that we approach problem-solving with so you
start with the hypothesis so might be something and as we’ll see in a second
like I think that introducing this new type of grill is going to cut my wait
times down by 30% the next is data validation do we even have the data what
is the completeness of it look like then we can actually run our models and we’re
going to run in this case simulation models get a results iterate on
scenarios and unfortunately this is where a lot of people stop and we think
this is you want you want to draw that arrow don’t you you want to draw that
greenair around to the front but I’ve seen a lot of cases where that just does
not go well right what do you think the intermediate step is what’s missing here verification I love it yes
business validation is what I’m calling it here so this is where you take that
model and you bring it back to your climates in the interim update and you
say what do you think this pass the sniff test will say oh no no no you
can’t you can’t even fit that piece of equip you can’t fit that new grill in
half of the stores ok back to the drawing board so we think of this as an
iterative process so we may have to go back refine our hypothesis or scrap it
all together come up with a new one right so this is the mindset we use
approach this problem and really all our analytics challenges at hobby okay let’s
talk about the case study let’s actually get into the simulation work that we did
for partners at McDonald’s about a year ago they launched all-day breakfast as I
mentioned overcame a lot of challenges I mean these a lot of these kitchens have
been around for decades and they were not they just weren’t built to handle
both breakfast and lunch time demand right it was it pure cutover so you stop
this process and you start the other one additionally some places like you could
imagine a college campus all-day breakfast is a very hot commodity that
there is a high demand for it but they’re places you might not get so much
alright so where do you roll it out do you selectively roll it out there’s
regional differences as well in the south you can imagine probably in
Nashville people prefer their biscuits for breakfast of the preferred and not
so much the Egg McMuffin but they want the biscuit so they had to overcome
these challenges and they did and it was it was really a smashing success they
pulled it off and there’s some public figures out there that it boosted
same-store sales as much as three or four percent which is really big so
immediately of course you know you you lasso yourself to one good thing and you
say okay how we can how can we do this again can we do it again next year and
that’s immediately where they went to and they said okay this is great but
let’s let’s expand it you could get a very limited scope of breakfast items
last year when we launched let’s try to offer all the breakfast sandwiches so
immediately you’re introducing more complexity more items you have to have
stocked and cooked all day there’s competition for the grill space so
there’s a cooking bacon Canadian bacon eggs
burgers on the same surface so you got some logistical challenges there and on
top of that they said okay it’s good let’s go put it in a test market but
what they found after running it in a few dozen stores was why we better we
better go do some additional work because they were coming back with a lot
more questions than answers so this is where they came to Hobby and
collectively we decided on said you know what let’s simulate this let’s simulate
the kitchens and see what what sort of equipment and what sort of labor we can
toggle to solve this problem so what sort of equipment investments do
we need to solve this problem and correspondingly labor and then kind of
fine-tuning the balance between those as well so this is a diagram of a typical
McDonald’s kitchen and I like to think of this is actually a small production
facility or a manufacturing plant all right we have deliveries of raw
materials we have customer fulfillment metrics we have to make on the other end
we’ve got a production line right down the middle and we’ve got
work-in-progress inventory workers all that so it’s not really all that
dissimilar from a factory and we get some of the same challenges and
complexities okay so this is a walk you through this
this is diagrammatically sort of the variable the decision variables and
parameters that were in play for our models regrettably we cannot pull up the
model because it’s a proprietary McDonald’s model but the decision
variables here are primarily going to be equipment and labor so we tune those we
we can actually drag and drop various pieces of equipment in to find the
capacities the layout and just to give you a sense for how big this model as we
had about 50 parameters six data tables so you can start seeing how just how
many millions of permutations we have to deal with here and as you all know that
can be somewhat daunting right just the sheer number of permutations where do
you even start in addition we had to keep our customers satisfied
so we kept track of things like what’s the average wait time how fresh how many
seconds off of the grill is my average burger how much waste am i producing so
we had to keep those in check and those are really kind of our objectives so with all those permutations and
possibilities were to even start well one of the advantages of working with a
company as large as McDonald’s is they actually have their own physical test
kitchens that are built out they’ve got about three of them and so what we were
able to do was we were invited into these test kitchens and we ran some
baseline scenarios and we’re able to calibrate the model to those baseline
scenarios to make sure that things were lining up so we check things like making
sure the inventory levels are the same in the real world and in our model as
well making sure the those customer service metrics we were talking about we
got to make those are key we’ve got to make sure those are very similar and so
we ran we were actually in the test kitchens for about four days running
some some scenarios that we’d recommended so try this piece of
equipment don’t try that one and one nice thing about the test kitchens is it
it really helps with the business validation right if your model comes up
with a wacky configuration that doesn’t fit in a physical test kitchen it’s not
going to fly right so that was a luxury kind of a unique luxury that we had in
this study the only problem with physical testing is it’s expensive right
almost a week’s worth of testing was a couple tons of thousands of dollars so
that’s in truth we that’s why we all have jobs in the room is it not you
can’t build out everything we can do things with any logic so much faster so
many more iterations and scenarios than can be done in the physical world right
so we have jobs so at the end of the study what we were able to provide was
we were able to Ron recommend additional scenarios that
they had not even conceived of with their initial physical testing through
that we’re also able to identify some savings in equipment cost so like huge
capex savings so when you mutex thousand dollar capex savings and you multiply
that by 14,000 that becomes a big number right and we’re excited because what
we’ve doubled mostly in the kitchen and the drive-through we think there’s huge
opportunity for for some use use of simulation and any logic specifically
within a restaurant right you have there’s some aspects of pedestrian flow
modeling in the front right I actually learned they the tiles look in next time
you go to a Wendy’s or McDonald’s anything the tiles will be different
colors and different densities based on where pedestrians walk I didn’t know
that there’s more durable tiles in the in the queue there are also course
opportunities for inventory modeling at the store level as well as DCs we do we
how they do some work in menu analysis if you run all-day breakfast what’s the
impact to the rest of your items are you going to start losing Big Mac sales of a
sudden so we do halo and cannibalization analysis that’s also well-suited for
simulation and then with the introduction of any logistics we believe
that this is really exciting because now you can put that restaurant model in the
context of your supply chain you can solve for things like delivery and
routing optimization right so that’s a really exciting new frontier as well so I really believe that again we’re
living in exciting times for simulation it’s we’ve got a good job we’re starting
to see a lot more data even in the in the restaurant industry we’ve got video
data now we’ve got sensor data on machines and on top of that we’re also
seeing a lot of customer specific preferences right you log into Amazon
and it’s tailored exactly to you most quick service restaurants now have an
app and they son offers targeted for you right so what does that mean what
implications does that have for us there’s more data to deal with there’s
more models to be run because they’re tailored to specific consumers or even
segments so we’ve got a lot of work to do but I’m excited to see these things
come together and be used successfully in simulation and ultimately is going to
lead to better results for our clients all right that’s all I have any any
questions I just have a question one of the scenarios when you run these models
there’s a lot lot of labor differences between the different scenarios do you
have in when you did the modeling did you have any automated routines to be
able to balance the labor or did you do that with manual iterations it was
typically we would run time windows at a time so we would start we would set with
a sort of a starting set of staff position in the kitchen and then run it
for say two hours maybe like a Saturday morning brunch period where you’ve got a
lot of traffic in the store so just more manual but we we’ve worked with models
for that have that you have sort of a automated configuration of Labor hi so
my question is like how did you account for the different sizes of kitchens
because each kitchen different cases without different sizing and right this
would affect a lot of things right so we had one of the neat features about this
model is we were able to go in and rearrange the kitchen layout so we could
take like a CAD drawing as the backdrop and shuffle the equipment so that
allowed us to account for multiple kitchen layouts and you know we we’re
not going to be able to capture all of them just like with the demand data you
have to make some kind of an 8020 rule what are the most important layouts to
capture what’s the typical demand cluster look like and so we want those
thank you