Agents Without the Hype: Where AI Fits in Your Ecommerce Team
with Henrik Hoffman, Co-founder & CMO, DEMA

“AI agents help brands plan and execute faster. Represent uses returns data to understand a product drop’s profitability in days—not weeks.”

You keep hearing that you need to “use AI agents.” What you rarely hear is where they go in your team, which work they should take on, and how they make your people better at their jobs instead of just busier. Henrik joins Paul and James to answer those questions, using real retailer examples.

Tl;dr: three reasons to listen:

1. See where agents fit in your team, and where your people still matter. You’ll get a practical picture of how ecommerce teams are being restructured. Some work moves to automations, workflows and agents. Other work still needs human judgement, and that’s where your team’s time and talent should go.
2. Get a rollout model you can actually copy. Hear how brands spread agents across the business by appointing an internal go-to expert who trains everyone else. You’ll also hear how they get buy-in for a change that is as much a shift in mindset as a line in the budget.
3. Hear how agents work in day-to-day operations. For example, Represent has embedded agents into its returns process and other teams have built DEMA into their weekly planning. Learn how they keep humans in charge and set a regular rhythm, so agents support the team’s priorities instead of working against them.

Also in this episode:

• Why the edge has moved from how you do things to how fast you get to an answer and act on it
• How to avoid choice paralysis and AI slop when you can suddenly do almost anything
• How this shift is changing DEMA’s product roadmap

Chapters:

[00:30] Why faster e-commerce decisions are becoming the edge

[03:15] Turning reporting into action, not just slides

[07:05] Avoiding AI slop and organisational rabbit holes

[09:05] Agents as colleagues and the risk of misalignment

[12:45] Bubble Room using separate agents for brand and performance

[19:40] Applying AI to current processes first

[23:25] Why teams want out of manual, repetitive work

[26:10] Starting with small wins and saving 15 minutes a day

[30:30] Competitor monitoring for peak trading periods

[31:35] Represent and real-time returns modelling

[35:20] Using historical data to validate models and assumptions

[37:20] Black Friday planning lessons from Baby Shop Group

[40:30] DEMA’s roadmap: measurement, agents, and business alignment

[46:30] Using agents to model impact before changing targets

In this episode, find out how to keep your attribution accurate as privacy rules tighten. Your ad management & optimisation is only as good as your tracking.

Listen if you want to spot where your analytics is leaking data and leave with clear first steps to fix it, no analytics background required.

Our guest expert is Littledata founder & CEO Edward Upton.

Key topics:

• Why server-side tracking has become essential, but isn’t a complete solution.
• How privacy changes, cookie consent and Shopify checkout updates affect attribution.
• Where Google Analytics fits alongside platform analytics and other reporting tools.
• Why richer behavioural and event-level data matters for analysis and AI.
• Common tracking gaps, including consent misconfiguration and lost links between sessions and orders.
• How to audit your setup by checking first-page tracking and consent behaviour.
• How to think about tracking budgets and the potential value of more reliable data.
• Who should own tracking and how ecommerce, marketing, CRM, and analytics teams should work together.
• Why first-party identifiers like email are increasingly important as customer journeys move across devices and apps.

Episode summary

Tracking ecommerce performance has never been harder. In this episode, we explain why and what brands can do about it.

Why tracking is getting harderServer-side tracking has gone from cutting-edge to essential, yet many brands still rely on free tools that can’t keep up. Privacy changes add to the pressure:

“Every year since iOS 14, Apple has tightened up its privacy restrictions, making it harder to track people from social media to Shopify stores.”

Why server-side tracking is now essential

Browser-based tracking relies on scripts and cookies that ad blockers, browser restrictions and iOS privacy changes increasingly cut off. Server-side tracking sends events from your server directly to platforms like Google Analytics, Meta and TikTok, so more of your data survives.

What it fixes

• Lost conversions: Fewer events are blocked, so ad platforms see more of your real sales and can optimise better.
• Broken attribution: Order data comes from your store’s backend, which is more reliable than a browser event that may never fire.
• Accelerated checkouts: Orders from wallets like Shop Pay and Apple Pay can be matched back to the original session using first-party data.

The most common two problems we see:

• Misconfigured consent banners: Poor setup means lost data. Make sure your first landing page records UTM parameters correctly.
• Orders that can’t be traced to a channel: Accelerated checkouts make it hard to link orders to the original marketing source. First-party data is the fix.

Where Google Analytics fits nowPlatform analytics have improved a lot, with better breakdowns and reporting. Google Analytics is no longer the primary tool, but it remains valuable as a secondary source for comparison and for feeding marketing dashboards.

“Google Analytics is still important as a secondary tool, even if it’s not the primary choice anymore.”

What to do next:

1. Audit your consent setup. Confirm tracking only fires after consent is given, and that you’re not losing data in the process.
2. Standardise your tracking. If you sell across multiple markets and systems, a dedicated tracking app keeps data consistent.

Episode timestamps

• [00:30] — Why ecommerce tracking is getting harder

• [03:00] — Server-side tracking becomes table stakes
• [05:15] — Shopify checkout changes and web pixels
• [07:45] — Google Analytics as a secondary tool
• [11:25] — Behavioral data and AI
• [14:05] — Common tracking problems
• [18:25] — Auditing consent and first-page tracking
• [21:00] — Budgeting for tracking
• [26:50] — Who should own tracking
• [36:05] — What server-side tracking can fix
• [40:10] — Attribution gaps that remain
• [43:00] — Agentic commerce and attribution

Product Discovery for Ecommerce: Context, Personalisation and AI

Reasons to listen:

1. Learn how product context and knowledge graphs can improve search relevance.
2. Understand which shopper behaviours signal genuine intent and which may be noise.
3. Get practical ideas for testing merchandising choices and deciding where AI needs human oversight.

How can ecommerce websites help shoppers find the right products when they don’t know exactly what to search for?

In this episode, James Gurd talks with Max Etheridge, Voyado’s Director for the UK and Ireland, about improving product discovery through richer product data, customer context and thoughtful use of AI.

They discuss why conversational search alone can’t fix weak product data, and how ontology and knowledge graphs connect products to categories, occasions and shopper needs.

That context can help a site understand a search like “something to wear to a summer wedding” and surface relevant products, even without an exact product description match.

The conversation also explores how ecommerce teams can distinguish meaningful purchase intent from passing interest. Repeated behaviours, such as refining searches, revisiting products or adding items to a wishlist, can be more informative than a single page view.

Combining those signals with product knowledge helps make personalisation more relevant to what a shopper wants now, rather than simply repeating what converted before.

James and Max examine practical ways to test product discovery, from mobile-versus-desktop strategies and filter layouts to content and product placement on category pages.

They also discuss who should control those decisions: algorithms can handle high-volume, lower-risk tasks, while people should retain oversight of brand-sensitive choices. Transparent reasoning and A/B testing help teams learn what works instead of relying on opinion.

Finally, they consider agentic search, the role of human merchandisers, and whether AI will change how teams use ecommerce platforms. One point runs through the discussion: new AI features are only useful when they solve real customer problems and have strong data and context underneath.

Chapters:

• [00:30] Introduction: product discovery, customer context and AI
• [03:45] Why context matters in ecommerce product discovery
• [07:00] Ontology and knowledge graphs explained
• [12:30] Reading shopper intent from behaviour
• [14:10] Personalisation beyond repeating past purchases
• [17:00] Who controls product discovery?
• [22:40] A/B testing merchandising strategies
• [25:20] Testing mobile experiences, filters and PLPs
• [30:30] Where AI should and shouldn’t make decisions
• [33:20] Agentic search and conversational discovery
• [37:05] Bringing content into product discovery
• [39:05] AI, user interfaces and the future of merchandising
• [46:40] Product roadmap and AI-first workflows
• [50:35] AI costs, guardrails and practical value