Lors de notre premier CPG Data Summit, qui s'est tenu le13 mai 2026, nous avons accueilli dans nos locaux de New York des professionnels des opérations commerciales, du marketing commercial, de l'analyse de données et des données provenant de tout le pays. S'appuyant sur le succès de nos sommets consacrés aux déductions et au RGM, cet événement a mis l'accent sur les données elles-mêmes — cette couche essentielle qui sous-tend tout, de la planification commerciale au RGM.
Au cours d'une journée entière de sessions et de tables rondes, des professionnels et des experts ont abordé les défis concrets liés à la qualité des données, à leur intégration, à leur gouvernance, ainsi qu'au rôle émergent de l'IA dans les opérations commerciales du secteur des biens de grande consommation.
We gathered some top takeaways from the conversations, case studies, and debates that took place over the course of the event.
Across every session, one theme returned with striking consistency: most data problems in CPG commercial operations are not technical failures — they’re structural ones. Bad master data, unclear ownership, and the absence of governance frameworks were cited as the root cause of everything from ROI analysis gaps to deduction disputes to forecasting disconnects.
Don Baker of The Partnering Group was direct about what he sees across his clients:
Don walked the room through the three pillars of master data readiness — the Pricing, Product, and Customer dimensions — and the specific failure modes that compound over time when governance is absent. These included: EDLP allowances hidden inside ERP pricing rather than planned and identified as customer spend, PPG structures that don’t follow TPM entry rules, check digit items shipped without billing, and customer hierarchies that don’t match external syndicated definitions.
The implication: organizations that skip governance frameworks aren’t just leaving insights on the table — they’re actively generating misleading ones.
One of the most provocative reframes of the day came from Don Baker in the opening session: the scope of RGM is wider than most organizations are treating it. With retailers now functioning as media companies, fines and fees reaching 2–3% of gross sales at some companies, and digital trade investment increasingly hard to separate from traditional trade spend, the boundary between “above” and “below” the net sales line is dissolving.
Earnings Before Interest & Taxes (EBIT) is a company’s profit from core operations before accounting for interest and tax expenses. This framing matters for data strategy: if you’re only capturing promotional spend and scan allowances in your trade system, you’re missing the full picture of what you’re investing in each customer relationship. Don also raised a pointed challenge around everyday pricing: at many of his clients, 70–80% of trade spend is going to everyday allowances — and almost nobody is running ROI analysis on it.
The room recognized this gap. As one participant noted, getting customer profitability right requires solving for the data first: “You’ve got salespeople making deals and not really knowing what they’re giving away. At the end of the day, customer profitability looks very different.”
Marine Bibal, co-founder of EasyPicky, introduced a concept that resonated strongly with attendees: the freshness gap. CPG brands are meticulous about product freshness, yet routinely monitor promotions on stale data — sometimes receiving execution information weeks after the promotion window has already closed.
Marine shared concrete examples of how execution reality diverges from plan reality: a beautifully designed snack brand Christmas display positioned at the back of the store where no shoppers could see it; a soft drink promo display placed directly next to its biggest competitor. In both cases, IPOS systems reported the display as live, and post-promo reports showed underperformance with no clear explanation.
Her prescription: make every store visit a data event. With image recognition technology working through video and full offline capability, teams can now access KPIs in real time while the field rep is still in the store — allowing discrepancies to be corrected before leaving, not just reported after the fact. She introduced four questions that her customers now embed into every promo brief:
1. Was the display actually built, and where in the store?
2. Was the promo price tag correct from day one?
3. Was the product in stock throughout the promo window?
4. What was the promo right next to mine?
Hands went up when asked who in the room had experience with in-store image recognition. Several attendees cited use cases in planogram compliance and competitive shelf auditing. Promotion compliance remains an emerging capability, but one with clear upstream implications for trade planning and ROI integrity.
UpClear’s Kurt Kaiser brought a candid vendor perspective to the day: Gartner survey data puts TPM satisfaction at troublingly low levels, and a significant contributor is the gap between what platforms can do and what they’re actually being used to do. For many organizations, the end state is a well-maintained trade system — and an equally well-maintained Excel file created to manage work-arounds.
The culprit, in many cases, is incomplete data integration. Upstream data loads (customers, products, prices, shipments) tend to be well managed. Where things fall apart is in the downstream: distributor depletion data and retail sales. When asked how many people have retail sales integrated into their trade planning, only a small minority of the room raised their hands.
Andrew Dentinger, Product Management leader at UpClear, walked through why this is hard — and how AI is starting to help. The process of taking raw retail sales data through cleansing, labeling, augmentation, and harmonization is labor-intensive, error-prone, and full of CPG-specific nuances (interpacks, combo SKUs, check digits, mismatched hierarchies) that off-the-shelf tools don’t understand.
UpClear’s approach uses layered AI techniques — from exact matching to forward/backward pass filtering to confidence scoring — to accelerate product harmonization and data preparation, while preserving a human override at every step. The modeling itself, Andrew noted, is increasingly commoditized. Copilot and similar Large Language Model (LMM) tools can run a price elasticity analysis on a clean spreadsheet in seconds. Getting to that clean spreadsheet remains the hard part.
Throughout the day, the conversation about AI kept returning to the same guardrail. Across the opening discussion, Andrew’s data readiness presentation, and Don’s closing governance session, speakers converged on a shared thesis: AI amplifies what’s already there — good or bad. It will not fix a broken process. It will not flag dirty data. It will give you a confident answer based on whatever it’s been handed.
Don drew a direct line from this to organizational behavior: when companies realize AI-driven productivity savings, they often take it as a one-time headcount reduction rather than reinvesting in the critical thinking roles that can actually capture value from it. His recommendation: reapply those savings to data and analytical functions that can put AI to work properly.
Andrew was similarly measured about the technology itself: for most CPG use cases, simple regression and statistical models that are decades old will outperform cutting-edge LLMs — because the data environments are small and bounded. “You have 104 weeks of Nielsen data to work with. You aren’t looking at sentiment analysis across billions of observations in a social media feed.”
The practical upshot, endorsed across sessions: start with governance. Get master data right. Define what you’re trying to measure and why. Then consider whether AI is the right tool for a given step in the process — and always keep a human in the loop.
Thierry Soudee — UpClear, fondateur et PDG
Thierry a ouvert la journée par une analyse fondée sur les données de l’évolution du rôle du CDO dans le secteur des biens de grande consommation (CPG) — passant de la gestion des systèmes à la gouvernance, à la conception de produits basés sur les données et à la supervision de l’IA. Il a mis en perspective l’ampleur de l’explosion des données à laquelle sont confrontées les marques de taille moyenne, a souligné l’émergence du RGM en tant que fonction incontournable et a posé les bases des discussions de la journée autour de la qualité des données, de la confiance et de la rapidité de la prise de décision.
Don Baker — The Partnering Group
Don a apporté à ses deux sessions ses nombreuses années d'expérience en conseil en gestion des données de référence (RGM) ainsi que la franchise propre à un professionnel de terrain. Au cours de la première, il a passé en revue les domaines clés de la préparation des données de référence : tarification, produit et client. Il a présenté des études de cas, énoncé les choses à faire et à ne pas faire, et lancé un défi constant à l'auditoire : « Définissez d'abord le langage des enjeux métier avant celui de la finance, sinon votre système commercial ne remplira jamais sa fonction. »
Marine Bibal — EasyPicky, cofondatrice
Marine a présenté le concept de « décalage de fraîcheur » dans les données relatives à la mise en œuvre des promotions — c'est-à-dire le délai entre le moment où une promotion est lancée et celui où les marques obtiennent une visibilité sur la manière dont elle a réellement été mise en œuvre en rayon. Elle a donné des exemples concrets de plans promotionnels qui semblaient parfaits sur le papier mais qui ont échoué en magasin, et a présenté un cadre permettant d'intégrer des données qualitatives sur la mise en œuvre dans le processus de planification des promotions.
Andrew Dentinger — UpClear, directeur produit
Andrew a présenté une vue d'ensemble concrète de ce que signifie réellement la « préparation des données », en abordant la détection des anomalies, la tokenisation, les défis liés au mappage des produits et le rôle des données synthétiques. Il a également livré un point de vue sans détours sur l'IA : dans le secteur des biens de grande consommation (CPG), la valeur réside moins dans les modèles génératifs avancés que dans l'utilisation de l'IA pour nettoyer, étiqueter et préparer les données en vue de la modélisation — une étape que la plupart des entreprises effectuent encore manuellement.
Don Baker — The Partnering Group
Au cours de sa deuxième session, Don a établi un lien entre la stratégie en matière de données, la conception organisationnelle, les structures de gouvernance et les modalités pratiques de mise en œuvre, même pour les petites équipes. Il a présenté son propre dispositif de gouvernance des données de référence mis en place chez Sara Lee, a expliqué en détail comment la gouvernance basée sur des formulaires de saisie pouvait servir de point de départ, et a conclu en donnant des conseils sur la stratégie de tarification dans le contexte actuel d'incertitude liée aux droits de douane et au coût du carburant.
