Overcoming AI Obstacles to Embrace Transformative Technology

He describes AI as being central to both disrupting business constructively, and meeting the needs of consumers and customers to create value for all stakeholders. However, it’s that last part about stakeholders that we're going to dig into today. There's a lot to talk about. 

As we begin the main part of our discussion, Cheryl and Conor, welcome. Let’s begin by introducing yourselves and sharing a bit about your roles at your company.  

Cheryl Perkins: Hi, I’m Cheryl. At Innovationedge we have a group of champions that focus on strategy, growth and innovation to help demystify things like AI, machine learning, and predictive analytics. We focus at a strategic level, but often end up helping execute and build capabilities at a tactical level, in order to make things happen.

Conor Keane: Spring Global sells Salesforce automation software and merchandise to the consumer goods industry. I'm based in Denver. The company has been in business about 20 years, and features some larger brands as customers.

For the longest time, the business has been transactional. However, in the last five years, we've started following the trend toward massive investments, artificial intelligence, and applying that to retail execution processes.

Johnston: Conor, I'd like to start with you, why does AI mean so many different things to different people? Especially how it's being shared with business partners in the industry?

Keane: Well, first of all, I should point out that I'm not a technologist, I'm a process and product guy. I don’t view it as the beauty of the technology, but rather the value we can extract from it. Obviously, stats are the bedrock of this. 

In my view, data science is stats, computer science, and technology joined together to do something useful. Then, machine learning is trying to look for patterns in data sets. Artificial intelligence may be a step further than that, where it enables the machine to make its own decisions and generate its own data. 

It’s a spectrum of sophistication with AI being the most sophisticated. For the most part, AI and ML are being used interchangeably, and I don't take great exception to that. I find I'm much more focused, in terms of the value we can extract from it.

Johnston: As part of today's discussion, we're going to focus heavily on field operations. We know, hence the title of this webinar, that CPGs are collecting large amounts of data, especially in the field, but acting on the data is still a roadblock. Conor, can you tell us why this is.

Keane: Our focus as a business is on field execution, so our perspective tends to be fairly tactical. Acting on data remains a roadblock for the same reason it's always been — people have more data than they've ever had. One of our customers recently shared that they have up to 500 reports produced every morning in Power BI. It's simply too much information to use, and as a result, only five of the reports are actually in daily use.

The second problem is timeliness. 

Johnston: Cheryl, what is your impression on this, especially from what you see from within the broader industry?

Perkins: The challenge is that translation part. You have all this information, but it's not just repeating or telling what those observations are. You have to ladder it to get to the insights that are actually actionable. That's a skill set that not everyone has, and a lot of people struggle. You can try to train it, but again, it gets back to how you look at disparate pieces of data, patterns, and then determine what can come from that. To Conor’s point, it's time consuming, uncomfortable, and not easy to do. 

Part of the challenge is that people repeat what they saw on the data instead of mining it and getting real insights that turn into action and deliver the results. If you don't ladder to get those insights, then you're acting on the base data and not getting the real value. You're not capturing or delivering the real value from the data.

Johnston: Our report talks a lot about people, process and technology. When it comes to this, where do people fit into the advancement of AI? Especially the role of the employee and the associate, why are they so important? How is this best communicated with the employees?

Perkins: If we look over the last 15 years or so, the technology now goes beyond any single function, and is a combination of art and science. You've got to create this interdependency among the teams and make sure that you don't have data silos to be able to leverage it broadly, improving experiences and getting better results.

What's absolutely critical is having the engagement of those employees. In the field, the store, wherever they are, they have to be actively engaged. Now, to do that, we'll discuss what enables this to work and what disables it. What's important is what I call having a “team of team” approach, where you're all engaged at the beginning, but the roles of those engaged will change or evolve over time.

This is where strategy and leadership come in. 

If you pick those right people in the field, you start doing things with them, pilots, etc. They bring people they know on their networks, then they bring others, and it becomes this cascade of people getting engaged to roll out the plan. They're critical to make it happen.

Johnston: Conor, what are your thoughts, especially in your experiences?

Keane: I would agree. In our experience, we've seen resistance, too. How does AI show up in the daily life of a field rep? It’s in an algorithmic recommendation that shows up in his app, “please deploy this assortment.” Looking at the statistics, 90% of those recommendations are ignored, at least from what we see. Largely because it's not tied to compensation, not explained, feels manipulative, and these reps reject it.

When we're trying to teach an algorithm what best practice is, we identify who the best people are, capture their expertise, the tools, and why they do what they do. Why didn’t they do other things that others do that aren’t predictive? Then, we use that knowledge to inform the algorithm, which uses a phrase, amplifying expertise.

How do we amplify that expertise across the workforce, so that everybody benefits in the form of training and gamification? Those high-end, skilled employees, the best of the best — which easily 10-20% of the workforce — what do they know? How should we capture that and feed it back in the algorithm? We call that amplified expertise as a methodology. That is really important.

Lastly, these people are very smart, much smarter than people give them credit for. We teach them what the algorithm is, what it's trying to do. Rather than feeling manipulated, we engage in the process and educate them on what we're trying to do. 

If it's framed properly, they'll adopt it. If it's framed inappropriately, they reject or resist it. That's been our experience.

Perkins: That's why it's so important that leadership, executive leadership, middle management gets involved to set the context to your point, and understand that things will evolve. Their role will evolve. It'll change over time, but they're an important part of it.

It's not the data alone. They have to translate it and take action on these insights. It's important that the top-down piece and the context around it is set, so the strategy and objectives are clear to those in the field.

Johnston: Not only is informing the employees and associates important to speed adoption, but doing so turns many of them into AI evangelists. Then, they're spreading AI good cheer across the company and acting as positive beacons for it.

Conor, your company has worked with some very large companies. I was wondering if you could share some insight about how they were able to do just that? How have they been able to demonstrate the value of technology to their employees?

Keane: In terms of automation and intelligent automation, we've had a lot of experience with customers through deploying apps, field processes, and HQ tools. Our basic approach has been to try and strip out friction, so the sales rep feels benefit in daily life. They feel there's less steps on the app, or they can get through their visit quicker. Everything is in one place and it's presented to them when they need it.

It's stripping friction out of the process, which we've been able to demonstrate. How people feel that in the field, they're able to get more done with less effort and less stress — that's a very practical example. A few drinks customers have seen in both the number of visits per day, but also the quality of the experience and relationship with customers get deeper. They have more time to talk to customers and spend less time heads down in the app.

That's anecdotal evidence, but we can see it in sales numbers. You can see productivity increase usually 2-3% in terms of sales following the rollout of the technology. Those are two basic principles that we see.

Johnston: Cheryl, do you have anything you can add with your work and your experience?

Perkins: We see it in the sales when we launch new products, etc. For example, Coca-Cola took a unique approach on how to leverage AI years ago. The power was not putting it in one siloed location within the organization. Instead, in order to get transformation, they needed to go broader, and make sure that the learning and frameworks were leveraged across the organization to fuel the capabilities and growth. That's where the power comes. 

They found a way to pilot and link to new products launched, so there was an obvious output. It wasn't data for data’s sake, but data linked to creating a way to get buy-in in the organization. This started in 2009, I'm sure many of you questioned the customized soda fountain machines, which as a consumer, I love.

They collected a lot of valuable data and were able to look at all these flavor combinations to understand what makes a product launch successful — they look at, use, and mine the data for insights. They successfully launched one of my favorites, the Cherry Sprite, nationwide as a new product. The key is that they knew they had this method of collecting data, linked it to something that would be an obvious output, and could see the success.

Now, they're seeing endless opportunities of personalizing products for target audiences. 

Then, make sure all those involved are trained and the capability is built. As we move forward in this area of moving away from being overloaded with data and making actionable insights count.

Johnston: I’d like to build on the idea of getting overloaded with data, and get a little granular. Let’s talk about the growing value of predictive analytics. This is something that's not new, but still not considered mainstream. What do you think, what's holding them back?

Perkins: You're absolutely correct that it is growing. There's many reasons why predictive analytics can add value. First, leaders and organizations that have figured out how to use it are developing more confidence in their plans as they put them forward. It's much easier to make decisions if you can rely on the data. 

Second is having a lot of live, accurate data, which increases speed and agility of decisions. It helps increase speed of moving new products to market, making it easier — depending on function and industry — to do that. 

The third thing to watch is the transformation of the planning process, which has only started happening the last couple years and is very positive. It's feeding in early to say, "we're seeing the need to do things differently in this part of the organization, but it also impacts another part of the organization." We're seeing lateral increases in collaboration across the various functions. 

Integrated planning is starting to make a difference, using analytics and transitioning faster to these insights that are actionable. It's interesting because this is where predictive analytics has real value. For example, H&M analyzes store receipts and returns in an effort to evaluate purchases in every store. They're now able to promote and stock differently, based on specific store locations because they've collected enough data on who's in the store, the guest, and what they need. 

A certain skirt or floral pattern may do better in urban stores than somewhere else. They're changing the entire inventory to match what the customers want. They’re able to ask, "Why is it not working in some places and it is in others?" You have to be willing to look at the data and use it to predict, rather than default to what you think the answer needs to be.

For H&M, they're dramatically changed practices on what they promote and stock — those are big pivots. If you are not willing to pivot, or not act based on the data that you have, then predictive analytics is not going to provide value.

Johnston: That's a great example. You mentioned developing an AI-ready culture within a company, Conor, what’s your take on this?

Keane: There are a couple of things. We built an algorithm for a customer in South America that is a bottom-up revenue sales algorithm based on store level and mix volume pricing, as well as other factors like weather, etc. We've also deployed this in the United States and have found, over time, that it can predict revenue.

If the data set is big and clean enough, we can predict fairly accurately, about 90 days out, +/- 4%, assuming the business is fairly predictable. Of course, COVID upset data sets a bit, as you can imagine. However, in general, these algorithms are very predictive. 

Imagine you had that algorithm telling you what your sales are going to be for the quarter and it's bottom-up. People still have not learned to trust it. Although, over time, as the accuracy is realized, they learn to look at it. Again, it is getting used to the fact that this predictive algorithm is there and learning to trust it.

One of the barriers is the data set. COVID perhaps messed with the data set, but also the ability to get clean data, clean the data once it comes in, normalize it and get it to the form where you can predict accurately — that's a process not everyone has the skill to do. It’s certainly a barrier. But people want to use this. Once they rely on the prediction, they use it to tie into planning, as Cheryl noted.

In our case, they might look at doing what-ifs. If there is a particular problem, or revenue is trending off, there are two or three options. First, run what-if scenario planning on the options, as to what to do about that. Then, execute on the action and see what it is, in terms of prediction for the quarter. Tying it back into planning, once they learn to trust the instrument, then they have to learn to tie it back into daily life and daily planning. That’s a process that will occur over time and people will learn to trust it.

Perkins: What's important is to continually reevaluate. One thing we're all learning in this space is that everyone's in learning mode — you need to assess what's working and not working. Although you’ve deployed a certain way, that doesn't mean it doesn't change or that you don't need to learn to pivot. What's critical is what's working, what's not working, and being ready to pivot. 

You've got a wealth of information, but people get overwhelmed by the data. How can we help with that, turn this data into actual insights, or what's not working, and how do we track that over time? Those are critical enablers to overcome everything we're talking about in terms of the barriers.

Johnston: There are other ways companies can develop AI-ready culture — the report digs into a few examples. What advice can you offer? Conor, why is AI different from integrating other new types of technology?

Keane: Rolling out an AI project is exactly the same as every other technology, but it’s also totally different in subtle ways. I mentioned earlier that large companies have spent quite substantial sums on optimization algorithms, particularly around assortment or price, for example.

These algorithms have been proven and tested, then get down to the field and the person. They ignore the algorithm recommendation because there's nothing in it for them. 

First of all, they think they know the root better than anybody, but the second thing is they're busy and don't know why they're supposed to deploy it. Then, it has to tie it back to compensation or influence them to use it in the form of gamification. What I've learned when rolling it out is that it helps to explain to that level of person why you're doing it and their direct manager.

Looking at the stats, 90% of users in the field are 22-32 years old, and 90% of them are men. Half of them are avid gamers and are well used to taking recommendations from the gaming context.

They rely on the machines to consult and educate them. Most of them are extremely smart, especially in technology and their root. To engage them, in terms of what you're trying to do and how the technology works, is a valuable exercise. They're very capable of understanding the concepts. 

Again, we have learned to explain what we're trying to do, even at a technical level, and engage mentally in it. For us, that has been valuable. Now, you could argue that's Change Management 101, but it's also how it's tied into gamification, compensation, and other things. It's change management like we've always done it, but with a new spin on it.

Perkins: It is different in terms of the training, and the fact that a lot of the training has to be learned by doing and through the incentives. The incentives vary based on the industry. To your point, Conor, who is out in the field and what incentivizes them to stay motivated and inspired? But there's a lot of similarities. 

What are the goals and expectations? Then, go to middle level management and make sure they've got the right mindset. Be selective who you start with pilots, focus on a few small pilots, celebrate those successes, and make sure they can deploy the plan. Then it comes to team members — the advocates, ambassadors, agents of change.

Understand the type of training that's needed — it will vary based on the industry or what you're trying to get — what types of data to collect, how to train them to do that. The biggest gap is training the why, the what, the how, but not the translation piece. Are they able to get in and take the data, ladder it to get to those actionable insights? If you can do that, you're going to create a cohesive shift from the top-down throughout the entire organization.

Culture changes and this change management piece is critical. However, the way we train and how we incentivize varies a lot compared to other initiatives.

Johnston: Cheryl, the report talks about the benefits that AI can provide for better and more seamless consumer experiences. Do you think AI is becoming table stakes for CPGs right now? If they're not leveraging this within operations will they be left behind when it comes to consumer experiences?

Perkins: Absolutely. It's critical for these experiences. Take yourself as a consumer or those around me. We're not content with just getting a product with functional benefits or getting convenience anymore. We want to be inspired and engaged. Even more so after this whole COVID experience. We're seeking a more immersive experience, but we're not going to give up convenience, so it's an “and” not an “or.”

A lot of companies are seeing that they have to continually look at new ways to transform the shopping experience and make sure they're leveraging insights  — they have to be nimble, move quickly. They have to improve execution in the field, while still doing the tasks they normally do. Making sure to have the right software and technology solution is critical, but then again, you’ve got to maximize sales and remain competitive.

When the customer or consumer pulls up, the groceries are right there, ready to put in the car. They also have the Walmart+ plan, which is pretty cool. If you don't know about it, it's based on the app and all that, but you don't even have to checkout at the register anymore.

You can scan as you shop, paying as you go. When you're done, you just have to go and leave. What's interesting about that is that at any time I can go into the app, check my profile — talk about instant gratification — and see how much time and money I've saved. 

That's the improved experience. It’s not just about what's in the cart, but how I'm being engaged throughout the entire experience. We have to pay attention to those services and start looking at how we create that immersive experience to keep people coming into the store.

Johnston: To your point, we are all consumers. As we have these seamless consumer experiences using technology, we bring those same expectations to our work technology. The technology we use daily should be as easy and enjoyable as what we're using at home. 

I would like to talk more about examples. Can you talk more about CPGs being innovative in field operations? What are they doing that's granting them success? Who's standing out? We don't have to name names here, but if we can, that's great. What examples can you provide of companies that are being innovative, Conor?

Keane: An example of innovation would be image recognition. It’s a great technology, which can be used in the course of a normal merchandising audit to do plan compliance or audit for discount. One of our customers is using image rec to replace some of the tasks that were previously done with manual survey. 

Then, using the AI engine to be smart about interpreting what the photo is showing, and turning that back into recommendations for the rep in real-time. With this, they can then be consultative back to the store owner and make recommendations for merchandising, pricing, or placement. 

The visit still takes 40 minutes in this instance, but the amount of time spent talking to the customers has doubled. The store owners value this interaction — they love it, find value in it, and want more of it.

I've seen them buy more, as a result. That's an example where someone is using a very tactical approach, but shifting the nature of the daily life of a rep, replacing labor intensive tasks with the machine. 

If you follow that upstream, the corporate level can now be smarter on individual stores, individual regions, what they want, how they might run a promotion, and even customize promotions to zip codes or particular sectors of the marketplace. The raw data captured in the photo has opened up new possibilities for both the rep and upstreaming in management.

Perkins: Another example to share is virtual fitting rooms at Levi’s, GAP, Brooks Brother, and Old Navy. They're using this as a way to advance the online platforms, making buying products more convenient and accessible. It helps when you're able to see something on yourself, in your home, etc. Wayfair is another good example.

Consumers are able to say, "I like how this looks, but I'm not sure how it'll look on this specific table. Is this the right table within my room?" Now, you're able to virtually put it within your space and see how it fits. It makes for more informed buying decisions. You're able to assess the fit more quickly and feel more confident in the decision. All of those things enable a good, smart decision.

Johnston: What best practices can you offer to CPGs that are just getting started, or still have a lot of heavy lifting to do when it comes to leveraging the power of artificial intelligence across the enterprise? Going back to the title of this webinar, how can they stop drowning in data and actually act upon the insights they're collecting? Cheryl, any thoughts on this?

Perkins: In the end, it's top-down and bottom-up, but you have to have buy-in and executive leadership to ensure that middle management is engaged, helping define how to deploy. It goes back to change management and culture evolution, understanding what types of data you need, how to leverage that data, and then make sure you're doing the right training and capability building.

It's hard because now there is such an influx and outflux of resources, that you'll often lose somebody that is good at a part of the AI process. Either collecting and predicting or translating into insights, you lose that and could have a breakdown. 

Those sound like generic things, but they're important to balance between the strategy and the tactics, to make sure that you're doing something with this awesome amount of data and are able to action it by getting to an outcome that delivers on value.

Johnston: Conor, what can you add to that? What advice do you give to your customers?

Keane: We often advise picking a project — it could be to roll out a particular product. I want to roll it out in the following manner, with the following timeframes, and measure it. They want to roll that out using a machine learning-based algorithm and measure the impact of it. 

Our advice is always to pick a specific project and roll it out end-to-end. Roll it out into the field, and see what happens in terms of adoption. See what you need to do to increase adoption and adherence, then gather the data back from the impact in the field. 

Continually teach the algorithm, as to what actions produced what outcome, and the data has to round-trip back into the algorithm so it can learn and get smarter. Therefore, we instruct or advise and guide to do a complete project end-to-end, get data flowing in a virtuous circle, and learn. The reason we do that is to try to educate people that the algorithm itself gets smarter.

Sometimes you need a solid block of time, even a year, for the algorithm to start to be reliable and predictive. We educate on that too, which is to build end-to-end and let it run, watch it get smarter, and enjoy the process as it gets smarter. That's very tactical.