Q&A with Johan Hellman from nShift – How AI moves from promise to practice

 nShift connects retailers, brands, manufacturers, distributors, and logistics providers to more than 1,000 carriers worldwide and supports over one billion shipments annually across 190 countries.

That scale matters because AI in commerce only becomes useful when it is grounded in real delivery infrastructure, not generic outputs.

Johan Hellman is VP Product Management at nShift, where he is responsible for the company’s global carrier network, including pre-built connections to carriers and transport service providers worldwide. He has more than 15 years of experience across logistics, shipping, 3PL, TMS, supply chain, and carrier management.

As nShift rolls out nShift Companion, its AI-powered assistant for delivery operations, Johan is focused on a more practical question than most of the market is asking: not whether AI sounds impressive, but whether it tangibly helps teams make better delivery decisions inside live workflows. Companion is initially available in nShift Checkout, where it helps users explore setup, get recommendations, and act in plain language while remaining grounded in structured delivery context and platform safeguards.

Q: Can you tell us a bit about your background?

I’ve spent more than 15 years working in logistics and shipping, across 3PL, TMS, supply chain, and carrier management. What has stayed constant through all of it is the operational complexity. Delivery looks simple from the outside, but underneath it runs on complex carrier logic, different data structures, local market rules, and a lot of exceptions.

At nShift, I’m responsible for our overall platform direction, strategy, and implementation. That gives me and my team a front-row seat to one of the hardest realities in commerce. If the delivery layer is brittle, everything built on top of it becomes harder to trust.

Q: AI is everywhere right now. What problem are you actually trying to solve with it?

The mistake is to start with the technology. The useful starting point is the friction customers and internal teams already feel.

In delivery operations, that friction shows up when a business knows it should improve something but struggles to make the change quickly and safely. Maybe checkout options are underperforming. Maybe a team wants to adjust delivery logic for a market. Maybe the setup is there, but the knowledge of how to change it lives with a handful of specialists.

That is where AI becomes genuinely useful. Not as theatre, and not as a chatbot bolted onto the side, but as a way to reduce complexity inside real workflows and help people get from intent to action faster.

Q: Why start with Companion in Checkout?

Because checkout is where delivery stops being theory and starts becoming a promise.

This is the point where retailers decide what customers can choose, what they will pay, what timing they are shown, and whether that promise can actually be kept. If those decisions are wrong, the damage shows up fast in conversion, support volume, and downstream operational issues.

Companion starts in nShift Checkout because that is where better guidance can create immediate value. It helps teams refine delivery logic, pricing, and option mix in the part of the journey where conversion and cost meet. It is also where a bad AI answer would be most dangerous, so it is the right place to prove that AI can be useful without becoming reckless.

Q: What makes nShift Companion different from yet another AI chatbot?

A lot of AI tools sit outside the workflow and generate plausible-sounding language. That is not enough in delivery.

Companion is built to work inside delivery operations. It combines a natural language interface with nShift’s delivery platform so users can ask questions, get contextual guidance, receive recommendations, and move toward action within the actual operating environment. That matters because delivery decisions need to stay aligned with business rules, service logic, constraints, and controls.

Built on nShift’s agent-ready infrastructure, Companion can access customer configuration data through MCP (Model Context Protocol), a standard interface that allows AI systems to securely access structured system data.

So the difference is not cosmetic. The difference is that Companion is grounded. It is designed to help with live delivery decisions in a way that reflects how delivery really works, not how a generic model imagines it works.

Q: What kinds of day-to-day problems does it help teams solve?

The unglamorous ones – and those are usually the most valuable!

Teams often need help finding the right setting, understanding the consequence of a change, improving delivery choice, or reducing manual admin around configuration. In many businesses, even straightforward delivery improvements get slowed down by buried menus, fragmented documentation, or dependence on one internal expert.

Companion lets users start with the outcome they want in plain language. That could be lowering shipping costs for small parcels, improving delivery choice, or refining checkout performance. From there, it provides contextual explanations, recommendations, and guidance tied to the actual setup.

That is useful because most operational drag does not come from dramatic failures. It comes from small changes taking too long.

Q: AI is good at sounding confident. How do you stop it from putting delivery promises at risk?

That is exactly the right concern, and it is why the underlying platform matters more than the interface.

In live commerce, you cannot afford AI that improvises around delivery logic. The answer has to be grounded in structured data, delivery promises, validation rules, operational safeguards, and real workflow context. That is how you avoid a system making recommendations that look reasonable but break the experience later.

Companion is designed to work within those controls. It is not there to bypass constraints. It is there to help users move faster while keeping delivery promises intact. That is a big difference. We are not trying to make delivery feel magical. We are trying to make it more usable without making it less reliable.

Q: nShift talks a lot about delivery context. Why is that so important for AI?

Because context is the difference between something being fluent and something being useful.

Delivery is full of edge cases. Carrier services vary by market. Timing logic changes by region and cutoff. Some options look attractive at checkout but fall apart when you apply real operational constraints. If AI is not grounded in that context, it tends to flatten complexity and give you generic answers.

We use structured schemas, delivery promises, Model Context Protocol (MCP), and platform safeguards. That is what makes the output more trustworthy. You are not asking a model to invent an answer from scratch. You are asking it to reason within a system that already understands the delivery domain and has been proven in practice.

Q: Is this just about better UX, or does it change how delivery operations work?

It changes both, but the bigger story is operational.

A cleaner interface is helpful, but the real value is that Companion can reduce the dependence on specialist-only knowledge and speed up execution inside teams that need to act quickly. It helps people understand the setup they already have, see better options, and move with more confidence.

That can improve checkout performance, reduce manual work, and help teams make better decisions faster. Early feedback from customers suggests improvements in checkout conversion, productivity, and decision-making.

Those are meaningful signals because they suggest Companion is not just making the software feel smarter. It is helping businesses operate better.

Q: How does Companion fit into nShift’s bigger AI direction?

Companion is the visible product expression of a broader shift.

nShift started early to invest heavily in agentic AI infrastructure on top of its core platform so AI-driven capabilities can operate on stable, trusted delivery systems rather than fragile add-ons. Companion is one of the first tools that shows what that means in practice.

It also makes the strategy easier to understand. Instead of talking about AI in abstract terms, you can point to something concrete. A user inside Checkout asking for help, getting grounded recommendations, and making a real improvement without leaving the workflow. That is a much better proof point than another marketecture slide about transformation.

Q: What should retailers and ecommerce teams take away from this now?

They should get more demanding.

The market is full of AI claims, but the real question is whether the system underneath can support trusted execution. If it cannot, the interface does not matter. In delivery, weak execution shows up fast. It hurts conversion, creates exceptions, and erodes trust with customers.

Retailers should look for AI that is grounded in operational reality, not detached from it. They should ask whether the system can explain recommendations, work within real constraints, and support change safely in live environments.

That is the bar now. The winners will not be the companies using the most AI language. They will be the ones using AI to remove friction, improve decisions, and keep promises more consistently.