The Connective Layer Retailers Need Before Another AI Tool
Alexandra Franco, Founder at Empact
RH-Hub: You say inconsistent results are rarely a tactics problem — they stem from a missing orchestration layer. Retailers own dozens of systems that don't talk to each other. How do they know if they need another solution or the connective layer underneath?
Alexandra: Before the tool question, it's important to ground the next decisions on: what are you actually solving for, and what does a win look like this quarter and two years out?
From there, map what you have. Some systems connect easily, others are custom builds or older tech that fight you every step. Once it's laid out, the next move is usually - build the connective layer first, before you buy another tool. Often the capability you think you're missing is already there, sitting in a system that can't reach the rest. That layer gets your systems talking and gives you one accurate picture across them, so the next tool you add plugs into that context on day one instead of starting blind. For example: Inventory - what a customer sees online doesn't always match what's on the shelf, or the store ten minutes away. Get that into one view and AI has something to work from. Customers in your store, the item's nearby, can you get it to them today?
Every retailer wants the same thing: sell it, and bring the customer back. The connective layer is what lets AI help.
From there, map what you have. Some systems connect easily, others are custom builds or older tech that fight you every step. Once it's laid out, the next move is usually - build the connective layer first, before you buy another tool. Often the capability you think you're missing is already there, sitting in a system that can't reach the rest. That layer gets your systems talking and gives you one accurate picture across them, so the next tool you add plugs into that context on day one instead of starting blind. For example: Inventory - what a customer sees online doesn't always match what's on the shelf, or the store ten minutes away. Get that into one view and AI has something to work from. Customers in your store, the item's nearby, can you get it to them today?
Every retailer wants the same thing: sell it, and bring the customer back. The connective layer is what lets AI help.
RH-Hub: You've written that AI defaults to average unless it's grounded in your own data. What does that mean for retailers using AI in customer-facing work?
Alexandra: It means sounding like everyone else. The AI usually isn't wrong, it's handing you the same right answer it's handing your competitors.
Look at the World Cup, a lot of players are in neon pink boots, across Nike, Adidas, Puma, all of them; nobody coordinated it. Every brand ran the same research and reached the same conclusion: pink is the highest-contrast color against grass. Each did the optimal thing, and now the pitch is full of identical boots, none standing out.
For a retailer, that sameness costs more, because a brand is built on consistency. Show up the same way enough times and customers start to recognize you. Recognition becomes familiarity, familiarity becomes trust, and trust is what gets them to buy. Let AI run customer interactions without your brand and data grounding it, and that chain breaks. The voice drifts off-brand, and a gap opens between who customers think you are and what they're now getting. Worst case, you go past generic and start sounding like a completely different company. What holds it together is the part only you have: your data, your customers' history, your point of view (including what you're not). Ground the AI in that, and it sounds like you instead of the category.
Look at the World Cup, a lot of players are in neon pink boots, across Nike, Adidas, Puma, all of them; nobody coordinated it. Every brand ran the same research and reached the same conclusion: pink is the highest-contrast color against grass. Each did the optimal thing, and now the pitch is full of identical boots, none standing out.
For a retailer, that sameness costs more, because a brand is built on consistency. Show up the same way enough times and customers start to recognize you. Recognition becomes familiarity, familiarity becomes trust, and trust is what gets them to buy. Let AI run customer interactions without your brand and data grounding it, and that chain breaks. The voice drifts off-brand, and a gap opens between who customers think you are and what they're now getting. Worst case, you go past generic and start sounding like a completely different company. What holds it together is the part only you have: your data, your customers' history, your point of view (including what you're not). Ground the AI in that, and it sounds like you instead of the category.
RH-Hub: Retail teams are lean and stretched. What AI fluency do they actually need in 2026 — and what can they stop feeling guilty about not knowing?
Alexandra: Less than they think, but not nothing. They don't need to build any of it, but they do need a working sense of what's happening under the hood: where their data goes, what the AI does with it, and how to steer the output with the right context.
There's a framework I like for this, the four Ds, from professors Rick Dakan and Joseph Feller. None of them are technical.
AI also reinvents a process - what took three steps by hand might take six, each one narrower. Part of the skill is seeing that new shape and knowing where you step in to review and provide feedback, which takes real time.
What they can stop feeling guilty about: which model is best this month, which platform to standardize on, whether they could build it themselves, and whether they're burning enough or not enough tokens. When you want to go deeper, ask the AI to explain what it's doing in plain words. You learn AI by working with it, not just studying.
There's a framework I like for this, the four Ds, from professors Rick Dakan and Joseph Feller. None of them are technical.
- Delegation: know what to hand to the AI, not everything needs to go through it.
- Description: tell it clearly what you want.
- Discernment: judge whether what comes back holds up, and flag what's off.
- Diligence: own what actually ships.
AI also reinvents a process - what took three steps by hand might take six, each one narrower. Part of the skill is seeing that new shape and knowing where you step in to review and provide feedback, which takes real time.
What they can stop feeling guilty about: which model is best this month, which platform to standardize on, whether they could build it themselves, and whether they're burning enough or not enough tokens. When you want to go deeper, ask the AI to explain what it's doing in plain words. You learn AI by working with it, not just studying.
RH-Hub: Every vendor pitch now says "AI-powered." What one question in a demo separates real capability from hype?
Alexandra: One question does most of the work: "Can I see it run on my data?"
Every demo is staged on clean, curated data chosen to make the tool look good, but your business isn't clean. Test it on a real snapshot of your data as it is today, not the ideal version: the gaps, the duplicates, the records that don't fit the template. Do it in a sandbox with backups, so you're testing reality without risking it. Then watch, does it hold up on the mess, or only on the tidy example they brought? Two follow-ups tell you the rest. What happens when it's wrong, and how does a person get brought in to catch it? "It doesn't get things wrong" is the hype answer. And is it really built for you, or a generic model with a thin layer of settings? Most "AI-powered" tools are built for the average company, that works right up until your business is the exception. So, does it adapt to you, or do you have to become average to fit it?
Every demo is staged on clean, curated data chosen to make the tool look good, but your business isn't clean. Test it on a real snapshot of your data as it is today, not the ideal version: the gaps, the duplicates, the records that don't fit the template. Do it in a sandbox with backups, so you're testing reality without risking it. Then watch, does it hold up on the mess, or only on the tidy example they brought? Two follow-ups tell you the rest. What happens when it's wrong, and how does a person get brought in to catch it? "It doesn't get things wrong" is the hype answer. And is it really built for you, or a generic model with a thin layer of settings? Most "AI-powered" tools are built for the average company, that works right up until your business is the exception. So, does it adapt to you, or do you have to become average to fit it?
RH-Hub: You insist a human stays in the loop. In retail, where should that line sit — and where do companies typically get it wrong?
Alexandra: Keep the person on anything that goes out into the world: your brand, your money, a customer relationship on the line, that's a human's call. AI can draft the reply to the upset customer, model the markdown, write the campaign, but someone with judgment and taste decides what happens next, what ships.
It usually breaks one of two ways. First, the human becomes a formality, approving twenty things a day and not really reading any of them (this is not oversight). The second way is subtler: the output review happens, but it's a dead end. The person fixes the same thing every time and never feeds it back, so the AI keeps making the same miss. This is where feedback makes a real difference - the human catches what's wrong and teaches the system so the next version is better. Over time the AI needs you less on the small stuff, and your attention goes where it counts.
Keep the human on the high-stakes calls and make the review intentionally teach the system.
It usually breaks one of two ways. First, the human becomes a formality, approving twenty things a day and not really reading any of them (this is not oversight). The second way is subtler: the output review happens, but it's a dead end. The person fixes the same thing every time and never feeds it back, so the AI keeps making the same miss. This is where feedback makes a real difference - the human catches what's wrong and teaches the system so the next version is better. Over time the AI needs you less on the small stuff, and your attention goes where it counts.
Keep the human on the high-stakes calls and make the review intentionally teach the system.
Alexandra Franco is the founder of Empact, where she builds the marketing systems and AI-supported workflows that lean or stretched B2B GTM and marketing teams are missing. She's a marketer first, with 16 years in B2B marketing that includes retail media and composable commerce, and an AI practitioner second. Her approach is practical: mapping where AI actually fits in the marketing function, where it makes sense and where it doesn't, and pulling real manual effort out of the day so teams get time back for what matters most.
