Promotion Optimization and AI in action


Optimizing promotions is one of the most compelling use cases for AI. Category managers can improve forecast accuracy, automate marketing plan creation and then optimize for maximum effect. Dipu Mukherjee, VP of product management CPG solutions shares with Mike Troy, Editor-in-Chief of Retail Leader how promotion optimization is disrupting retail.

  • Read Video Transcript

    MT: Hello, Mike Troy here with Retail Leader at the Symphony RetailAI Conference. Wow, like, is that right? Like I’ve got to stick with my original thing. Okay, I’m doing it over.
    Hello, Mike Troy with Retail Leader here, at the Symphony RetailAI Xcelerate Conference in Irving, Texas. Joined by Dipu Mukherjee. Your focus is on promotion optimization, which is a huge topic, because there’s so much waste in promotions. What are you seeing, what are you hearing from customers about how they’re using your solutions to solve some of these promotion optimization…eliminate waste basically, to be more effective?

    DM: So first thing is, we’ve seen three different things. One is, the one effective way of creating the plans, more of an automatic plan creation, to save time for the category managers. So that’s one. Number two is, how do they create a better forecast? Because their forecasting is not giving them the big bang for their buck. And so our AI models are providing forecast generation that is top of the notch. And it gives a better accuracy of what the predictions should be. So that’s two. And number three is, how do we get a recommendation or optimized plan so that they can take it to market and execute? So these are three things that they can do, to be able to reduce waste and get a more of an automatic, accurate plan for execution for their different categories.

    MT: So save time, improve accuracy, enable execution. I get the saving time and improve the forecast, but how are you…how do you impact execution? That would seem to be outside of the scope of the technology part?

    DM: Correct. So…so once the promotion plan is created and it is optimized by our AI driven models, it will give the list of promotions with specific tactics that the category managers can adopt and enable them. And that doesn’t happen. And what normally happens is, they take the year ago plan, what they’ve done last year and make some tweaks, and basically, implement them. But that may not give the most bang for buck and could be a lot of…lead to a lot of wastage, right. So what we’re doing, is giving them the opportunity to adopt our solution, which is looking at billions of combinations of price, tactics, stores, UPCs, you name it, timeframe, and it gives…it’s not humanly possible to do that many combinations. So our tools are doing those billions of combinations and coming up with the best solution for the category managers to adopt. And that’s what makes the difference in saving money, as well as giving them the opportunity to grow the category.

    MT: You came from the CPG world, before you joined Symphony?

    DM: Correct.

    MT: What did you see then and what are you seeing now? Are companies figuring it out?JM: So companies are trying to figure it out. In my CPG job, what we found was accounting managers would want the ability to do a predictive analysis of forecast generation and optimisation, playing around with the solution options, to create what if scenarios, to see which one would be better, and then go to a negotiation table with a retailer, to say, these are the promotions we want to promote in your stores, and give us the space. And then the retailer will do multiple deliberations to come up with the right, acceptable plan. What has happened today, is that they also are struggling with their forecast efficacy. They are also struggling with what the optimised plan should be, in order for the retailers to accept it and then go and make that happen for them, for their category. Because the CPGs are looking at a few categories themselves. For example, if you look at PepsiCo they are have salty snacks or beverages, and they are interested in that, for the right promotions in a customer like Kroger. So how do they do that, is their challenge. And they again also do it from the same principle. Start with the year ago and then tweak. This process will give them the ability to say, okay, we’re going to give you the best possible plan with different combinations and come up with the best one for return.

    MT: So is this solution more geared toward the retailer or the CPG company, or is it either/or?

    DM: So it’s either/or. So what we’ve done is, we’ve created an opportunity for the CPGs to propose promotions, and if they don’t have our planning tool, they will propose the promotions, however they come up with it, and it will come into our solution, and the retailer would basically take the promotions and run the optimisation and see what…whether it makes sense from a funding perspective or not.

    MT: So the retailer could…so you come to me with some data, say, hey, I want to run this promotion. Can I run it through the tool? No, not going to work.

    DM: So then they can go and iterate with the CPG…

    MT: They can modify it.

    DM: …to be able to say…yeah, modify. The other option is, if the CPG has our tool, they can actually run their proposed promotions through our tool to come up with what the optimised solution should be. So then what happens is, the CPG has our tool. It gives them the optimised solution and they can recommend to the [inaudible 05:28] who has our tool as well, to say, we have a common platform.

    MT: Okay.

    DM: I’m using your platform to come up with this and it should basically give you the bang for the buck and the target objectives that you’re trying to achieve.

    MT: How do I know, how do I validate what you’re telling me, is going to really work, because I’m depending on the tool? Is there some way to like test it?

    DM: Yeah, so…so the best way, what we’ve done is do kind of like an A/B testing. So the…that helps in adoption as well. So you basically come up with an opportunity to say, okay, give me a set of stores, give me a set of promotions, give me events along the year, I’m going to execute that in specific stores and let the actuals come out after the execution is done, compared with…so you have a forecast, you have actuals, you compare with our process and you compare that with the manual process that the CPG or the retailer is doing. So once you’ve compared that, that gives you the delta of accuracy.

    MT: Okay.

    DM: Saying, how far are we off, between the tool versus the actual?

    MT: Then I could also determine like what my objective is, like am I trying to drive traffic?

    DM: Exactly.

    MT: Am I trying to go sales…?

    DM: So the multi-constraint optimisation does exactly that.

    MT: The what?

    DM: Multi-constraint optimisation.

    MT: The multi-constraint?

    DM: Constraint optimisation. So what that does, it lets you set an objective. So your objective may be maximising income on track. So you want to maximise income on track, you select that objective and then you say, I may do that, I don’t want my margin to go negative, right. So you can also set that.

    MT: You put all these rules in there.

    DM: Exactly, exactly.

    MT: Okay.

    DM: It’s a set of business rules, a global set of constraints that you can set and say, this is what I’m going to run my optimisation for. So when you run those optimisation proposals…constraints and objectives, you get the best outcome possible.

    MT: Okay. And then I’ll look like a hero to my boss…

    DM: Exactly.

    MT: …with gross sales and true profit.

    DM: And it also helps to narrow down the combinations.

    MT: Okay.

    DM: Because if you have no constraints, the combinations are going to be billions and billions of combinations. So now when you put objectives and constraints, your solution…

    MT: And then is it integrated with CINDE, so I could just like say, what’s the most effective promotion I could run for salty snacks…

    DM: Absolutely, absolutely.

    MT: …around Super Bowl?

    DM: So CINDE is going to generate these billions of combinations of plans and highlight the ones that make sense for the category. And the category manager’s going to say, okay, I can do this for you on traffic, I can do this for maximised sales, I can do this for maximum margin and that’s the beauty of it, is that, that’s not humanly possible for any category manager from a time perspective, to do. The system will do it for them.

    MT: Right. Sounds good, doesn’t it.

    DM: Absolutely, my pleasure.

    MT: Thank you.

    DM: Thank you.