276. Why AI Alone Can’t Fix Your CPG Deductions

Distributor deductions can quietly drain a CPG brand’s margins, especially when there is no clear process for planning, reviewing, and disputing them. In this episode, I sit down with Yuval Selik, co-founder and CEO of Promomash, to explore how AI is changing deduction and trade spend management.

We discuss where AI can save time, why clean data and proactive planning still matter, and why experienced people must remain involved in every decision. Yuval also shares why relying entirely on AI could lead to inaccurate accruals, incorrect promotional matching, unnecessary disputes, and damaged distributor relationships.

Startup to Scale is a podcast by Foodbevy, an online community to connect emerging food, beverage, and CPG founders to great resources and partners to grow their business. Visit us at Foodbevy.com to learn about becoming a member or an industry partner today.

Transcript

Jordan Buckner (00:00)
Distributor deductions are one of the places where CPG brands can quietly lose margin cash flow and visibility into what's actually happening in their business. We've talked about this extensively before, but you know, the real challenge isn't actually just knowing what the deductions are, it's creating a plan for them, how to review them consistently, how to decide what to dispute, and how to build a process that doesn't fall apart as retail distribution grows. So I want to kind of dive into this topic, especially as AI is coming.

Coming into this space and it's changing how people build processes around deductions, Trade Spend management. So I've invited Yuval Selik who's the co-founder and CEO of Promomash to talk about how AI is changing the way CPG brands are managing deductions and Trade Spends. And I also want discuss where experienced people are still needed to make judgment calls, manage relationships, and execute the right steps. You know, this conversation is gonna be a little bit more intermediate, advanced level on deduction and Trade Spend. So if you're still

still new, go back and listen to episode 98 where Yuval and I break down the basics. All right, Yuval. So I'm curious to learn how AI has influenced you and your approach to deduction management. And yeah, welcome back to the podcast as always.

Yuval (01:12)
Well, thank you. Of course. I'm looking forward to it.

I will say a few controversial statements here, so you should wait till the end or at least listen throughout carefully with intent. to set a frame, Because we kinda have to

look at deductions prior to AI and understanding

why they happen and how to manage them effectively and then we can get into AI, right? And so what I seen and the reason I got into the business of course 'cause I had my own brand, I understood exactly how where the pitfalls are. And

there's such a thing in life.

Where we make decisions reactively and proactively.

And I think that it flows through all aspects of life, including business and including deductions, right? So to start off, I think just setting the base about what proactive and reactive management is before getting into AI, because AI flows into, you know, takes over from that side of the business.

Jordan Buckner (02:01)
Yeah, let's do it.

Yuval (02:02)
So from my standpoint, many brands treat deductions like janitorial work. Okay.

It's always a cleanup. They always have a mop. They're always exercising, you know, some sort of look back. And they're always panicking. 'Cause what's happening is the money, you know, hits the account short. Three months later, somebody's digging through some emails, they're trying to figure out why. It's always reactive and they're always cleaning up after the fact. And

By then, half the paper trail is gone. So proactive deduction management is the opposite. That's where brands should know what a valid deduction looks like for every retail that they s retailer that they sell into. they've got the promo plans, the agreements, the contracts, the expected costs. They capture everything up front, right? So when a deduction lands, they're not investigating some sort of mystery. They're matching it against.

Something that they've already planned for.

The mindset shift is this though.

I think deductions

are in surprises for the most part. That's a little controversial, right? Because most people will say, I was always surprised, I was always surprised.

Yes, some are.

But we're playing the CPG game. So you got to accept that piece of it, right? We're not going to go into why, but you get the point listened to previous episodes. But for the most part, they're the cost of doing business. Right? And they're correct. And they're predictable if you do it right. And I see this happen all the time, Jordan. The brands that win aren't the ones who chase deductions faster. They're the ones who set themselves up so there's little

To chase. And that's when you catch the invalid ones early. They you know the brands

They led the valid ones through clean and that's really playing the offense instead of defense. And that's the foundation of deduction management where AI can come into play.

Jordan Buckner (03:42)
Yeah,

I mean I think that what you're talking about in terms of having a plan and a process in place early so that you are building that into your business and planning for it appropriately is key. I think the biggest thing, right? A lot of people are worried about

Either one, not knowing how deduction chargebacks actually align with their trade spend plans that they agree on, especially for early brand brands. And then two is being able to tell the difference between a approved or agreed upon deduction or chargeback or an invalid one because those can happen too. I think the third thing is you know, problems that happen in the system that you should be aware of, right? Like if you're having late delivery times, it's still your fault, but it's not something that should be.

happening but if you don't have a process to actually see that to know it's an issue it's hard to fix and so I guess I'd love to one you can do a quick overview of just like how like what are those kind of key processes that you need to stay on top of and then the ways that you've kind of built in processes to manage those and then I'd love to kind of go into from there does ai make that easier is it make it harder does it complicate things and where's kind of the role of people in there too.

Yuval (04:52)
Okay. So what you're really talking about is how do we try to avoid some of the deductions in the first place? And that's and that comes with planning.

Jordan Buckner (04:59)
Right.

Yuval (04:59)
and I think brands get that backwards. they plan the promotion.

And the deductions are always an afterthought, right? They have a contract and they don't think about it. They sign stuff and if you ask them what did you just sign, they have no idea what they just signed. And that's I wouldn't you agree? Like that's ninety five percent of the brands out there, right?

Jordan Buckner (05:16)
Completely. But now the other

thing that I realized, right, is like there might be 10 different charges that can happen, but brands don't realize how those can happen at the same time on the same invoice versus one at a time. And then so it just confuses them.

Yuval (05:29)
Yeah, I mean that happens because distributors and retailers know very well that if they give you plain deductions when they happen, right when they happen, it's easy to look for and validate, right? But if they wait three, four months and then they send you six months worth of deductions in one place, good luck. That ain't gonna happen. And so you need support. But

You know, again, I think from a brand's perspective, before they run anything, they should know what's going on, right? What the cost is on the back end. So practically, you know, when they're planning a promo or a launch or even a distribution program, they got to keep track of what's expected to spend. Now it could be a spreadsheet if they want, or it could be a fancy system, but a tool is just a tool. And so that's one controversial thing that I will say is that tools don't fix things. They help.

They support you. But if you can't get it done in Excel, you ain't gonna get it done in a six-figure TPM solution. That for that's for sure. It's actually gonna be worse because at least you know how to use Excel. You won't know how to use the TPM solution. And the wrong people are on it anyway. So you got salespeople managing trade, you have founders and CFOs managing deductions. I mean, these are the problems that we're facing every single day. And so everybody's chasing the next shiny new object, and they're seeing all these new pop-ups of

companies come out come about and we'll talk about that promising AI is gonna solve the you know their woes and problems, but that's a huge misconception because I don't care if you have the God of AI, Fable 55, you know, I don't care what model you have and I don't care what platform you have. If you don't have the people running the show, it ain't gonna work. It just ain't. We're we're not in we're

Jordan Buckner (06:56)
Yeah, I think that's a good point.

Yuval (06:57)
not in an industry that is clean.

This is not a clean industry. This industry has more holes than any Swiss cheese that you can find in Switzerland, right? That's where Swiss cheese comes from. So

Jordan Buckner (07:07)
Yeah.

Yuval (07:07)
that's that's the deal. That the the holiest Swiss cheese, it's worse in CPG. So you're expecting AI or a tool to fix that? No, you can't. Impossible. Mm-hmm.

Jordan Buckner (07:18)
So

talk about that let's kind of talk into the tools, right? Because like right now you kind of mentioned a couple. There's Excel, there's some TPM software like Promomash and some others. and then there's, you know, on the AI side of things, right? There's people integrating some of the chat and analysis tools into software. And then there's founders who are just going directly into the chat GPT or cloud and dropping in

their reports from distributors to kind of help an analyze those. And so I love your perspective, because you're in this every day on what that role of like the software is and is AI helping? Are you integrating that into Promomash in ways? are you not and what reasons and kind of your perspective on that part.

Yuval (07:57)
So we'll talk about Promomash and AI in a second. But

There's a hill that I'm gonna die on. And that hill is eighty percent of the ownership must be done by a he a human. Okay. Twenty percent is the tool. And I put tool as AI, as software, as algorithms, as ML. I mean, all of it is kind of

working in tandem. Promomash started using AI six years ago when we were coding with AI. We weren't using scanners, we were using machine learning and coding. We started that trend, this was kind of pioneered it six years ago. And we've enhanced it ever since. But never does a deduction invoice pass through an AI without a human looking at it. Ever.

Because I know it's gonna happen and we could talk about that later. But right now, the disservice that I see in this industry, and this happens every revolution, every technological revolution, you have a bunch of entrants coming in, taking advantage of the product, right? It happened in two thousand and six and seven with mortgages, it happened.

in two thousand with in nineteen ninety nine or whatever with with the internet. It happened in the industrial revolution. It happened when horses were replaced by cars. I mean, it doesn't really matter the type of revolution or technology.

Companies will come and say that this technology will completely transform and transform the world. And for the most part, the funny thing is, everybody's wrong. And I'll tell you why everybody's wrong. the biggest optimists are wrong because it's going to be way more than they ever imagined. And the biggest pessimists are wrong because it's never going to be that bad. So so y

Jordan Buckner (09:24)
Hey.

Yuval (09:26)
everybody's going to be wrong, and everybody is wrong right now. And my problem, I mean we could talk about this shit for like hours, but

The problem is when you have trillions of dollars funding companies that just put AI in their name or in their, you know, dot AI or whatever it is, you are setting yourself up for a very interesting next few years. Because even today, with the best models, AI isn't capable of managing this type of environment. It's too complex. It's too

Messy, it's too Swiss cheesy, it's too you have to understand the contextual aspect of so much in so many different departments and so many different relationships and a type of relationship and the type of brokers you're working with. Every broker is different, every contract is different, and you're gonna ask AI, a model, and a chat bot to do it for you. Good luck. And I'm telling you where the disservice lies. Every day a new company comes up on the radar, and you know them because they're probably calling you up.

Right? We do deductions. It's AI forward and we found the formula. And and as like five of them or six of them, like in the last few months. And there's a couple of, you know, ones that were about a year or a year and a half. The technology and the companies, the brands. So I'm not gonna name names, but all these new TPM and deduction management solutions that have launched in the last three years, let's say, they're still new. Everyone looks good on paper.

I can put a website up, put a nice story together, market the shit out of it, and look like I know my stuff. But the problem, and we've learned this hard pro mesh, the first few years is where we got burned because the back end had to catch up with our ambition. And so it looked good. Everybody thought we were great initially, but there was a lot of pain to get there. It took us six years, seven years now, to get to where we are right now. And we're still getting better and better and better. And I still would say that we're not a hundred percent.

So you take these companies that are like five months old all of a sudden managing 25% of somebody's trade in AI with no people and no support, and asking salespeople who are busy anyway to check on the platform. They don't know the platform, they don't even have the time to check on the platform. and the whole thing is going to blow up like it did in 2000. Now, in 2000, it was a financial blowup. I think with AI,

It's gonna be a data blow up. I think what's gonna happen is a year, two years down the line, because you work with Chad and Claude, and I love both. I mean, believe me, Claude is my best friend. I'm on it 24/7. But it tells me a lot of things that I love because it tells me how good I am, tells me how wonderful I am, it tells me how smart I am, it tells me it tells me how pretty I

Jordan Buckner (11:53)
Those are all true, you've all knowing you. There's nothing bad.

Yuval (11:57)
am, it tells me everything. And guess what? It's wrong partially.

Not about me, but everything else. He's wrong about it because ultimately what's gonna happen with brands, and this is where AI is gonna really shine in the next few years, they're gonna have an audit or they're gonna look at their books and their accruals are gonna be completely wrong. They're coding, their GL fund mapping, their promotional matching, their you know, the trade reasoning behind what's effective, what isn't effective. They're planning. Planning fully without really having a great

you know, team behind you and pla all of that is a recipe for disaster. I'm not saying it doesn't do good things, it does. But fully trusting these new companies that are coming out and saying, I have no people, I have AI, and you guys just manage it, it's scary because brands don't know how to use these platforms anyway and they don't know how to check it anyway. So that's the problem. They don't have the right folks. They don't have the trade experts and the deduction experts to check the system. So they rely on the system, they take it as f at face value

They manage what they manage, just like you would trust, you know, ChatGPT to give you, send you an email response or something like that. And it sounds great. But if it doesn't have all the context, what happens when it bombs, you know, UNFI with disputes? It disputes everything. Whatever it is. Hey, dispute this. What do you think the relationship with UNFI is going to be like if you're just disputing stuff without looking at it?

Jordan Buckner (13:11)
Yeah, I think the other thing that's so interesting within there too is right, like you learned this early on in in Promomash is that you need those experts who understand the relationships, how deductions work, what they actually mean, how it relates to the business. And it's not just the tool, as you mentioned, but it's the inputs to the tool plus how you actually make decisions based on what it's telling you, right? Because as you mentioned, there are a lot of things that are valid deductions, valid chargebacks.

but they might represent an underlying issue or challenge in the business. But then you need to take action to fix those, to really diagnose what the issue is. And if you don't have someone internally or a really good partner who's surfacing those, then those can get missed and not acted upon, which I think is really detrimental to a brand.

It can be really challenging. You know, I'm kind of curious, like I definitely agree with like the human approach. You know, where do you think AI can make the biggest immediate impact in deduction management?

Yuval (14:08)
So

Now that I got it off, you know, my chest on how I feel about AI. Don't get me wrong, I do love AI and I'm doing vibe coding and I'm creating prototypes and it's changing my business as well. But it's just it's heartbreaking. You know, at ten o'clock at night I'm arguing with AI that it's you know, it's just back and forth. It's really, really annoying. But I think the biggest

Jordan Buckner (14:27)
Yeah.

Yuval (14:28)
immediate win is matching deductions to planned spend. That's one thing that I could do.

fairly well. It's not fully accurate by any means, but it gives you a good head start. it could look at a deduction and suggest, right? it could say, here's the promo that this probably ties back to. But again, it needs a human. you don't want to just accept it because I've seen what happens if there's multiple deduction multiple trade plans that are very similar or very close and it just chooses one and it's like you're you're dealing with your curls.

And that's the problem. It's like if you don't get it right, your cruels are off. and so that's where the time goes today, right? Is just trying to match it manually with the salespeople, and salespeople should not be matching promotions in the first place. That's one rule that I disagree with, and I think every brand wants to do that, and I think it's the wrong approach, huge, hugely wrong. that's why Promomash we have our team match deductions on behalf of the brand. We don't want the brand's salespeople to match deductions.

they need to be selling, shaking hands, kissing babies, you know, promoting, asking for more money. That's what they're supposed to do. Let professionals do the rest of this stuff. but I think that that one of the reasons is you know, from a matching perspective, it takes hours, you know, just in a single account. So AI can do the first pass in seconds, gets you most of the way there. But even here, and I've tested this internally, there are gaps.

So even with our matching, you know, platform and what we're doing and we're building right now, there are a lot of the gaps. But in general, the thing is, and and I'll go back to AI in general.

The biggest impact is only available to brands that have perfect data. And zero brands have perfect data. And I mean that. Zero brands have perfect data. So you can think you have perfect data. You don't. I mean, maybe Mars and Hershey's has great data. I'll give you that. But for the brands you and I work with, zero brands have perfect data. So then

That's where the limitation is. So I think it can do it can aggregate, it can find patterns, it could give you insights, it can give you suggestions, it can give you things that are helpful. Right. Just like the difference between the first Word doc, you know, like Apple created Word or whatever it was, not Word, P what was Pages dic they created for

Jordan Buckner (16:33)
Pages, yeah.

Yuval (16:34)
right versus doing it on a typewriter.

Yeah, there's an if there's something to to be said about that technological leap. And that's where we are right now. It's an enhancement. It enhances what you currently do and it it helps you with it. But

Jordan Buckner (16:47)
Yeah,

and I think that, you know, one thing that I've

Yuval (16:47)
it doesn't replace it. That's that's what I'm trying

to say.

Jordan Buckner (16:50)
yeah, I think that one thing that I've seen as well is that, right, like if you have nothing, AI can either be really helpful or send you in the wrong direction, right? Confidently send you in the wrong direction. And so managing anything, but in this case trade span deductions, you have to have a baseline understanding of

which deductions are valid, what the contract says, how they actually work, how they align with your promotions. And that's like there's some information out there, but it really takes that experience of knowing and working with the distributors with these tools, like how things work, right? Like codes are changing with distributors and there's new programs that they launch and that affects how your numbers look. And AI is not great at catching on to those updates. And so it can make it seem like it can tell you

one story when the reality is different. You know, that said, I always say like if you have nothing, you've never done this before, you maybe don't have the resources to hire someone, like it could probably get you something that's a little, you know, that's better than nothing if you have the right mindset that it still might be wrong.

Yuval (17:50)
I'm not knocking AI at all. I'm just saying AI plus human in the loop is the way to go. AI is the game changer that this industry and every industry, you know, required. MIT did a study recently where they were they put AI to the test of some, you know, legal documents and some other things, and they found high double-digit hallucinations on all of the simple stuff that AI is supposed to do well. So

I hear people putting in their freaking contracts, not just CPG. I'm just saying like legal contracts through Chat GPT and saying, What do you think? And it gives them, this is great, or here's some changes, and they go with it without an attorney. Okay. Bad mistake. It's the same thing. It's not just CPG, it's every industry. It can help, it can speed things up. It could summarize. Like God knows how I ever managed my world without AI prior to

Fathom, you know, my note taker and every like it transformed my life, but it didn't replace me. And that's the big thing that everybody needs to understand. until AI becomes that good that it can replace people, you better be the one in the loop. You better understand what you're dealing with because you're gonna find yourself in in in deep doo doo if you don't. Just my take.

Jordan Buckner (19:00)
Yvonne that's a great way to

end. Yeah.

Yuval (19:01)
That's that's the hill that I'm dying on. I don't know. People will people will argue, but you know, if you don't have anybody arguing

with you, you're not making, you know, any noise. So I'm here to make some noise.

Jordan Buckner (19:09)
I love it. I love that pers

yeah, I love that perspective. And yeah, definitely having someone in the loop and not just a person, but the expert who knows deductions, how they work and how they build those relationships.

Yuval (19:16)
Expert, exactly. Yeah. Word word of advice.

Don't let salespeople do trade. Don't let anyone but deduction people experts do deductions. Everyone should do what they do best. Their job title, right? That's that's the trick. You you follow that principle, you'll be fine.

Jordan Buckner (19:34)
Love that. Yuval thanks so much for being on as always and talking through this. If you want to learn more about Promomash and how it can help your brand, you can check out the details in this in the show notes. Yuval, thanks again.

Yuval (19:43)
Appreciate it. Thanks so Jordan.