Exclusive interview with Gain about how agentic AI is changing the industry – part one

What is the solution that Gain has developed for the food and grocery industry, and its benefits?

Gain developed Natalie, an AI Employee purpose-built for procurement, merchandising, and commercial operations in retail and grocery. It’s easier to define her by what she is not: a dashboard or an AI “copilot” or agentic SaaS extension.

When we first launched Natalie, she was a category strategy expert capable of managing complex supplier negotiations end-to-end.

Powered by simulations, machine learning, and market intelligence, she handled the slow, challenging, monotonous grind of tail-spend procurement so human teams could focus on decisions that actually require human judgment. This includes sourcing strategy, negotiation execution, supplier communication, and contract implementation, etc.

But we've since promoted her.

Natalie's expanded role now spans procurement and merchandising in retail and grocery.

She performs the day-to-day commercial work that keeps retail and grocery businesses running: pricing optimisation, supplier negotiations, and trade spend management.

This is work that sits at the intersection of procurement and merchandising, the operational core that drives margin but rarely gets the analytical attention it deserves because teams are stretched across too many suppliers and too many SKUs to do it justice.

The benefits are straightforward. Natalie does the work your category managers and merchants wish they had time for. The research. The negotiation prep. The pricing analysis. The competitive benchmarking that everyone agrees is critical, but nobody has the bandwidth to do properly on a Tuesday afternoon when three other fires are burning.

She brings data-driven rigor to every interaction, operates continuously, and maintains institutional memory across all supplier relationships and negotiation cycles.

She compounds knowledge in a way that no rotating cast of analysts or consultants ever could. Your best category manager with six months in the role is starting from scratch on half their suppliers. Natalie never starts from scratch.

And to be clear about what she isn't: Natalie's not replacing humans. She's making them dramatically more effective. Your best category manager, armed with Natalie's intelligence and freed from the grind, becomes something closer to what the role was always supposed to be, strategic, focused on the suppliers and categories where human relationships and judgment genuinely move the needle. Natalie handles the long tail of work so humans can focus on the most vexing challenges and decision-making, not to mention communicating the “why” and “how” behind the gargantuan shift to AI-driven outcomes to other humans.

Where exactly does the AI sit in the negotiation workflow?

Natalie is not an agentic extension to a sourcing suite or a copilot that offers basic guidance. For example, Natalie's not just offering to design an RFI/RFX or offering to take supplier vetting and communication off the desk of a real human. She's actually doing all the work, starting with identifying the opportunity and progressing through to the implementation of a contract. And, of course, everything in between, including sourcing strategy identification (e.g., anchoring to commodity price changes, volume discounts, competitive pricing dynamics with other grocers/retailers, alternative supply options, etc.) and the negotiation itself. She is probabilistic by design which requires significant AI scaffolding that goes beyond stacking domain-trained LLMs alone.

To get specific about where the AI sits: everywhere in the workflow, not at one point in it.

Natalie ingests spend data, contract terms, market pricing intelligence, supplier performance history. She autonomously identifies savings opportunities that a human category manager would either miss or never get to because they're buried under 500 other SKUs. And once she flags an opportunity, she doesn't hand it off. She builds the sourcing strategy herself. Is the right lever a market-indexed price adjustment? A volume consolidation play? A competitive bid scenario? Some combination? She figures that out. Then she runs the negotiation. Communicates directly with suppliers, makes and evaluates counteroffers, applies tactics calibrated to the specific supplier relationship and category dynamics.

The architectural distinction that matters here is probabilistic decision-making with deterministic execution. At the strategy and tactics layer, Natalie is weighing likelihoods. What is the probability this supplier concedes on price if we anchor at X? What's the expected value of pushing for a longer term versus taking a smaller discount now? How does this supplier's margin structure actually constrain their flexibility? But at the execution layer, it's hard rules.

She won't exceed predefined negotiation boundaries. She won't commit to terms outside approved guardrails. She won't engage a strategic supplier without the right escalation protocols. Full stop.

How does this actually work under the hood?

It's a multi-layer system, not a monolithic model. Think of it as three interlocking tiers, though in practice the boundaries blur because the whole thing runs as a continuous loop.

First, the intelligence layer. Natalie continuously processes and synthesises structured and unstructured data: commodity price feeds, historical spend patterns, contract clause libraries, supplier financial health indicators, competitive market dynamics.

She's not retrieving data on demand like some glorified search engine. She maintains a living analytical model for each category and supplier relationship, updating it as new information arrives. When tomato paste futures move 12% in a week, Natalie doesn't wait for someone to notice. She's already recalculating implications across every affected contract and flagging where renegotiation windows have opened.

Second, the strategy engine. This is where the probabilistic reasoning lives. Given the intelligence layer's outputs, Natalie generates and evaluates negotiation strategies using a blend of game-theoretic modeling and learned patterns from thousands of prior negotiation outcomes.

She simulates supplier responses.

If we open with a 7% reduction ask anchored to the commodity decline, what's the likely counteroffer range? If we introduce an alternative supplier into the conversation, does that shift the probability distribution? She runs these scenarios continuously, adjusting as the negotiation unfolds and new signals come in. And to be clear, this isn't prompt engineering on top of a foundation model. It's a purpose-built reasoning architecture that uses LLMs as one component among several. Optimisation models, decision trees, Bayesian updating frameworks. The LLM handles language. The rest of the stack handles thinking.

Third, the execution and governance layer. This is the deterministic backbone, and it's what makes Natalie deployable in enterprise environments where a probabilistic system operating without hard constraints would be a nonstarter. Every action Natalie takes passes through a rules engine that enforces boundaries set by the customer. Maximum commitment thresholds. Approved negotiation tactics. Escalation triggers for high-value or strategically sensitive categories. Required human approval checkpoints at defined stages. This layer also handles audit trails, compliance documentation, and reporting so procurement leadership has full visibility into what Natalie did, why she did it, and what alternatives she considered before acting.

The piece that ties it together is the feedback loop. Negotiation outcomes feed back into the intelligence layer and refine Natalie's models. Supplier response patterns update the strategy engine's probability estimates. Governance exceptions and human overrides teach the system where its confidence was miscalibrated. She gets better at negotiating with a specific supplier every single time she engages them. She's not starting from zero each cycle the way a category manager who just rotated into the role might, or the way a consultant parachuting in for a sourcing event definitely would.

And this is why the AI Employee framing matters. Natalie isn't a tool you invoke when you need help with a negotiation. She's a persistent entity with accumulating knowledge, evolving strategies, and institutional memory that compounds over time. The AI doesn't sit in one box on a workflow diagram. It is the workflow.

What data does it require — and how clean does that data need to be?

Natalie needs what any good negotiator needs: spend data, contract terms, supplier information, and market pricing. The difference is she actually reads all of it!

The baseline is transactional spend data and supplier master data. From there, the more you feed her, the sharper she gets: contract terms, historical pricing, commodity indices, competitive/shelf benchmarks, supplier performance data. Each layer adds leverage. But you don't need all of it on day one. We start with what's available and expand the scope as more data becomes available to us. We can also “bring” data, such as commodity price data, if a customer does not have it.

On the cleanliness question, in short, data doesn't need to be perfect. Natalie is built to work with enterprise data as it actually exists, not as the IT team wishes it existed. She can handle inconsistencies, gaps, and the kind of messiness that accumulates in any ERP that's been running for a decade. That said, she's not magic: garbage in still limits what comes out.

The practical reality is that most clients have better data than they think. It's just scattered across systems, and nobody's had the time or tooling to pull it together into something actionable, let alone do some basic clean-up and classification on the fly (not to mention enrichment). Natalie does that synthesis work as part of her process. She'll also flag where data gaps are limiting her effectiveness, which turns out to be one of the more useful side effects. Clients end up with a clearer picture of their own data quality than they had before she showed up.

Need more insights on agentic AI and the future of shopping

Agentic AI is actively transforming the food and grocery industry from optimising supplier negotiations, streamlining procurement, and dynamically shaping merchandising and pricing, while redefining how consumers shop through AI-driven purchasing decisions.

To help industry understand the complexity, the direction of travel, and current case studies, we have created a comprehensive report called ‘Agentic AI and the future of shopping’ that subscribers can now access.

Share your story with the food & grocery industry

We welcome contributions from retailers, brands, and solution providers shaping the future of food and grocery. If you have a compelling case study or perspective on technology, operations, or shopper behaviour, we’d be keen to hear from you.

Please get in touch with [email protected]