Sean McCarthy: Let's just get started by having you tee up who you are, but also what Conative is and does — heading into this conversation about data quality, silos, and all these things you know a lot about.
Mike Le: Sure. Sean, thanks for having me. I'm Mike Le, founder of Conative AI. I came from Vietnam to New York in 2005 and started my first company in 2007 — a digital marketing agency that I scaled to over 100 people, working with a lot of ecommerce brands and enterprise clients over 18 years.
Working closely with so many brands, I kept seeing the same problem: brands underestimate the impact of stock issues on marketing performance because the inventory team and marketing team operate in silos. I also saw that most inventory teams were still using spreadsheets — very manual, very labor-intensive — and relying on simple year-over-year comparisons for planning. That leads to a lot of overstock and understock. So I created Conative AI, a predictive inventory platform to solve that problem. We have proprietary AI models built on deep learning to predict sales, and I've been working in this space for about three years now.
Sean McCarthy: Inventory and demand — it feels like one of those spaces where when things go wrong, they go very wrong. You're either missing out on huge opportunities or sitting on a mountain of dead stock. You mentioned that most companies don't have a data quantity problem — they have a data quality problem. What does that actually mean, and where does it show up when you start working with a new brand?
Mike Le: If you look at most fashion or beauty brands, they have a lot of data — ecommerce data, ERP data, Google Analytics, attribution platforms like Triple Whale. The data exists. The real problems are a few things.
The first is data consolidation — how you bring those sources together to make informed decisions. And that's where you immediately find discrepancies. I'll give you an example: we've worked with brands where the product category structure in their ERP doesn't match the category structure in Shopify. When you pull the data together, you have to choose one source as the master, and everything from the other source becomes confusing. What makes it more complicated is that both structures often go outdated over time — so they've actually built a third category structure in a spreadsheet and use that as the real source of truth. Because of that, many brands can't even fully trust their own data. And the quality of your decisions is only as good as the quality of your data.
The second issue is data tagging. If you work with a fashion brand, you're not just looking at style, color, and size. You want to know fabric, material, silhouette — and you want those attributes tagged properly. When you do, you can actually see what's winning and what's dragging you down. Data tagging is deeply underestimated, and it requires a real investment in auditing, cleaning, and mapping before your AI models can do accurate forecasting.
Sean McCarthy: That feels like a very parallel conversation to tech debt in website infrastructure — decisions piling up, discrepancies multiplying, and eventually you're unable to move quickly. In organizations you've worked with, what does data getting siloed actually look like? What are the most common silos?
Mike Le: There are many types of silos, but the one that hurts brands the most is the silo between inventory and marketing.
Brands don't fully appreciate how much the inventory situation affects marketing performance. I'll give a real example. A brand I know had sales drop 30% in a single month. Leadership questioned the marketing team — what's going wrong, what are you doing wrong? Then the next month, sales shot right back up. The reason? They restocked one SKU. That single SKU drove four times more revenue than the second-highest-revenue product in the entire portfolio.
That example is real, and it illustrates how large the impact can be. It wasn't targeting, it wasn't creative, it wasn't channel performance. It was something that happened on the inventory side.
The reason brands don't see this connection is because inventory teams and marketing teams use completely different tools and live in completely different analytical worlds. An inventory planner doesn't know what really matters to the marketing team, and vice versa. So insights never get shared effectively.
What we've seen work is as simple as having one inventory person sit in on the weekly marketing call — to hear about upcoming campaign direction, share shipment delays, flag stockout risk so marketing can plan around it. That one practice alone closes a lot of gaps. But you also need a platform that unifies both data sets so cross-team decision-making can happen in real time.
Sean McCarthy: You mentioned wanting to break these silos, but what barriers typically come up when brands actually try to do it?
Mike Le: The main cultural barrier is that when teams are siloed, they stay in their lane. When something goes wrong, the instinct is to say, "That's not my department's problem." We want to move away from that mindset entirely. As a brand, you grow when everyone works toward the same goal.
What I want to solve with Conative is a unified data view — where the marketing team and inventory team can look at the same platform together and make decisions in sync.
For example, you can look at a product that's trending and say: we're putting more ad spend behind this SKU, inventory needs to ensure we're stocked and those POs are prioritized. Or you can identify a product with low sales but a high conversion rate — meaning traffic is the constraint, not demand. If marketing pushes more traffic, inventory needs to back that up with stock.
The cultural shift is really about encouraging collaboration with the understanding that both teams are working toward the same revenue outcomes. When you sync the insight, the results on both sides get dramatically better.
Sean McCarthy: Most people's reality is still spreadsheets and rules. Conative AI brings deep learning into this picture. How does that actually change inventory and campaign planning?
Mike Le: Before I talk about AI, it's worth understanding why spreadsheets became the default. Brands are pulling data from multiple sources, pasting it into spreadsheets, and merging it manually. The only logic you can apply in that environment is rules — if/then logic, year-over-year comparisons. And it's deeply prone to human error.
Traditional enterprise planning tools moved to mathematical formulas and basic machine learning — an improvement, but still limited in what they can take as input.
Deep learning is fundamentally different. It's a multi-layer model that's already learned the patterns across your data and produces forecasts with significantly better accuracy. The power is in how it handles edge cases.For example: if a product sold a thousand units last year but was out of stock for 15 of those 30 days, you cannot use last year's sales as a reliable baseline for this year. Old methods can't account for that. Deep learning can — because you can feed the historical stockout data directly into the model and the model understands the constraint.
Same with trend reversals. If a product performed extremely well last year because it was on-trend, but demand has since shifted, relying on that historical data will get you badly overstocked. Deep learning can be calibrated to weight recent velocity more heavily and deprioritize stale patterns.
These outlier cases are where the biggest inventory mistakes happen. And they're exactly where deep learning outperforms rules-based approaches.
Sean McCarthy: Can AI also account for macro factors — economic trends, pricing sensitivity, things like that?
Mike Le: Yes, you can feed macro data into the model and let it learn the relationship between those factors and demand. But I want to be honest: that relationship should always be tested empirically, not assumed.
What we've found is that some factors have a very large impact on forecast accuracy — weather is a good example. Other factors that you'd expect to matter don't actually move the needle meaningfully. And some are simply black swans that no model should be expected to predict.
Different brands react differently to different inputs. So we evaluate each data feed by turning it on and off and measuring the effect on accuracy. If adding a macro variable improves the forecast, it stays. If it adds noise, it comes out. That's how you build a model you can actually trust.
Sean McCarthy: When teams make the shift from spreadsheet-based planning to agentic AI, what does that actually feel like operationally?
Mike Le: There are three feelings we consistently hear from planners using Conative.The first is feeling free. All the data is already collected and aggregated for you. The baseline forecast is ready without manual effort. All you need to do is apply your judgment — review the forecast, tweak where your expertise tells you to, and move forward. About 90% of the hard work is gone.
The second is feeling confident. The AI uses a method that is demonstrably more accurate than what came before. There are also safeguards built into the model — even if a planner blindly accepted every AI recommendation, the model has guardrails that prevent it from exposing the brand to excessive risk.
The third is feeling proactive. You know what's coming. You're not constantly catching up to a situation that already happened. You're ahead of it.
Sean McCarthy: What steps should a mid-market brand take today to prepare their data for an AI-driven future?
Mike Le: Start with the goal, not the data. Identify what decisions create the biggest positive or negative impact on your revenue. From there, trace back to what data is required to make those decisions well — and make sure that data is clean, current, and structured correctly.
A thorough data audit is critical. Data can be overwhelming, and you don't want to tackle everything at once. You want to focus on the data that actually matters for the decisions that move the needle.From there, the path is to find a platform that can aggregate that data in real time, clean it, and structure it consistently — so using it becomes easy rather than a project.
I want to be transparent: this first step takes real effort. It took me over a year and a half just to build the data infrastructure that could aggregate data from multiple brands, across multiple categories with completely different data structures, into a standardized system that could actually feed AI models accurately. That foundational work is why Conative can deliver results the way we do.
The good news is that we've absorbed that complexity so brands don't have to. When you come to Conative, that infrastructure is already built. You're not starting from scratch.
Sean McCarthy: What has you most excited as you look ahead to 2026?
Mike Le: The future of agentic AI in inventory.The first breakthrough was solving the data problem — building infrastructure that consolidates and cleans data across sources. The second breakthrough was the forecasting models themselves, the deep learning capability that produces accurate predictions at scale.
The next frontier is speed of value extraction. Right now, planners still have to look through a lot of data to understand what's happening and decide what to do. AI agents change that. Instead of searching through data, agents surface what actually matters — so you can zoom in on the right decision immediately and act faster.
In practice, that means: an agent that recommends which products to carry over versus discontinue next season. An agent that analyzes sales data to identify winning factors — what's working and why. An agent that flags unusual velocity changes so you can respond to a surprise winner or investigate a sudden drop. An agent that places POs automatically for consistent core products, eliminating the ongoing reorder task entirely.
All of that day-to-day work — gone. Teams get to focus on the big strategic decisions, the ones that actually move the brand significantly. That's the direction we're building toward, and honestly, the pace of AI capability is moving so fast that what's possible is changing almost month to month.
Sean McCarthy: Mike, I really appreciate you sharing this. It is a space that's completely changing how people think about their work — and how quickly it's changing is striking. Thanks for your time today.
Mike Le: Thank you, Sean.
