00:00:00
Recently, Ford Motor Company brought back 350 engineers, humans, because the AI process they
00:00:07
had in place was not leading to the standards that Ford wanted in vehicles coming off their
00:00:13
assembly line. Starbucks went away from AI-powered app for its inventory system because too many
00:00:20
mistakes were happening. Santiago Galino is a professor of operations information and decisions
00:00:26
here at the Wharton School, and he joins us to try and explain why that could be developing a
00:00:32
gap in the process in the retail sector in using AI. Santiago, great to talk to you again. How are
00:00:38
you, sir? I'm doing very well. Very good to see you, Dan. You know, it's interesting because so
00:00:42
many people thought AI was going to be the panacea, make everything work perfectly. Companies
00:00:48
would probably be able to cut costs as well, but we're seeing, at least in retail, that there are
00:00:55
some of these instances where maybe it doesn't run as perfect as a lot of people expected.
00:01:02
Yes, yes. I think that that's what we're seeing now, and I think in part is a natural evolution
00:01:10
of these new technologies. I think many companies are moving from experimentation to implementation,
00:01:18
and I think this reminds me of a phrase by Yogi Berra that, in theory, there is no difference
00:01:23
between practice and theory, but in practice there is. And I think that companies are seeing
00:01:30
that firsthand, and I think it's a natural evolution, and it's good to see that companies
00:01:37
are trying and learning. You mentioned in the article that recently ran about Starbucks and the
00:01:47
changes they made. Take us through the problems that they had because I think for a company like
00:01:52
Starbucks, the last thing they want to have happen is something like inventory end up impacting how
00:01:59
their employees work, but also the downstream impact that probably occurs with the customer
00:02:05
that would also occur as well. Yes, so I think it's an interesting example because it shows a
00:02:13
couple of, I guess, red lights or things that companies should think carefully about when
00:02:20
rolling out these technologies. If you're an outsider to the retail industry, you might not be
00:02:26
aware that inventory accuracy is not a trivial problem. You will assume that every retailer
00:02:34
knows perfectly well the number of units of each SKU in the store. That is not the case,
00:02:44
and not because these companies don't try really hard. It's because it's a very difficult
00:02:50
part of the process to keep track of. Companies make a lot of investments and efforts, and so
00:02:56
I will say it's not surprising that Starbucks tried to bring technology to this specific
00:03:02
step. Now, the challenge is that accuracy in this particular case matters a lot, and so I think that
00:03:11
inaccuracies in the count of inventory, inaccuracy on the feed that is informing now the
00:03:19
forecasting algorithms, can escalate very quickly. With the additional point, which I think many of
00:03:25
the articles and conversations with store managers highlight, is that it creates a negative loop
00:03:32
because the moment that the employees learn that the technology cannot be fully trusted because
00:03:39
it's inaccurate and making mistakes, then the employees start to put less effort in implementing
00:03:46
the tool correctly, start to send wrong signals, and that starts to spiral out and making the whole
00:03:53
process fail. So I mentioned the case with Ford Motor Company. To see a company like Ford bring on
00:04:02
AI into their assembly process and not get the results that they wanted to get, mistakes happen,
00:04:11
and then have to bring back human engineers, it really does talk to that component of
00:04:16
how humans are still going to be important even with AI kind of in the mix. Absolutely, and I think
00:04:23
it's kind of an effort from managers to understand what AI is really about, and I think that
00:04:32
because of the excitement and the hype and the inflated expectations, I think there was,
00:04:39
and there still is, around the misunderstanding of what AI can and cannot do, and I think that
00:04:45
this kind of pendulum between going all-in and coming back with the human-AI collaboration
00:04:54
is a healthy exercise, and I think that we should not land hard on companies like Starbucks or
00:05:05
Ford Motor Companies in the sense that I think it is a risk not to experiment, not to try to learn,
00:05:13
at the same time it's like, well, how fast, how quickly can we scale, have we done the
00:05:21
exercise of going through the process in detail before throwing in a technology that, as we all
00:05:27
know, is still very much evolving. So the companies really, to a degree, were so excited by the
00:05:35
technology and the potential of what it could bring, they maybe jumped into the water with
00:05:41
both feet a little too quick, correct? Yes, and I think it is also a conversation
00:05:48
internally at the organization of what is this technology actually bringing to the process
00:05:54
and to the customer ultimately, and in the case of Ford Motor Company, you can imagine that
00:06:01
quality mistakes in the car can be an extremely risky proposition. It's not that you got the
00:06:09
wrong drink at Starbucks, easy to fix, not a big deal. If you start to have errors in the process
00:06:18
when you make a car that cause accidents, now we are at a different league, and I think that's also
00:06:25
an important consideration when you think about what to scale and where AI at this stage can bring
00:06:34
value. So is the process benefiting from it? Is the customer valuing this technology in one way
00:06:42
or another? Well, doesn't that also go to the question of brand? And it's something that I
00:06:47
know that marketing professors and marketers have looked at a lot in the last few years,
00:06:51
how companies are more focused on brand than ever before. And when you have these mistakes,
00:06:57
and as you mentioned with Starbucks, an issue with the stocking and having the right
00:07:03
amount of product in place has an impact on the customer, or I should say on the employee.
00:07:09
It could potentially have a downstream impact on the customer, and the customer gets a negative
00:07:13
feel for it, and they don't maybe go back to Starbucks as often as they may have normally done.
00:07:20
Absolutely. And I think that failures in the process, failure in the product, failure in the experience are always going to have an impact on the brand and the willingness
00:07:31
of the customer to visit again. I think it's also important to recognize that for the most part,
00:07:39
customers are not going to reward you for bringing fancy new technology to the store.
00:07:45
You can think of Trader Joe's as an example. They use bells to send signals inside the store,
00:07:53
one bell, I need one more cashier, two bells, I need the store manager, three bells,
00:07:58
and everyone is happy with that. I think it's an idea that we need technology when it's needed.
00:08:06
Are the bells a good solution for Trader Joe's? Absolutely. They have a footprint of the store
00:08:11
that allows for that. If you're going to ring bells at Costco, it will be a problem because
00:08:16
you're not going to listen to that everywhere in the store like you can do at Trader Joe's.
00:08:21
So it's a combination of the right solution, the right technology, and what the customer is actually
00:08:27
going to value from that technology in the store. So which is more important than in that entire
00:08:34
process? And I assume it is the recognition of the managers, the company, of what the expectation
00:08:44
needs to be of just how much you can gain from having AI in the process. Yes. And I think that
00:08:51
if we think about the steps we were talking before about experimentation, moving into
00:08:57
implementation, I think when you make that transition, the next step is, okay, what is
00:09:02
the return on investment here? Because all these investments, technology more broadly, AI in
00:09:08
particular, they are not for free. And I think that that consideration comes to play, especially
00:09:16
when you roll out these technologies at scale, they start not to give you the expected results.
00:09:26
And I think that that is, I think, the third step in the evolution of these implementations.
00:09:32
Experimentation, implementation, and then looking for the value these technologies are actually
00:09:37
creating. So has the value of return on investment changed then within these companies because of
00:09:46
the dynamic of AI? Is it as important to companies as it has been in the past? So I think when I talk to companies, you can recognize, I guess, two main camps. One that are
00:10:00
kind of almost arguing for a FOMO effect, like we cannot be out of this. We need to go all in.
00:10:09
Eventually we'll figure it out how this is paying itself, which this is my personal take. I think
00:10:17
not the right approach. I think other companies are constantly experimenting, learning what is out there and trying to align those investments and learnings to their strategy
00:10:31
and then making a commitment and a decision that is consistent with the overall strategy of the
00:10:37
company. And this, I mean, can vary a lot, even within an industry like retail, like not every
00:10:44
technology is for every retailer. Santiago, great to talk to you today. Thanks very much for
00:10:49
your time. All the best. Thank you. Santiago Gallino, Professor of Operations, Information
00:10:54
and Decisions here at the Wharton School.