There’s an old story about a drunk looking for his keys under a streetlight. Someone stops to help and, after a while, asks, “Are you sure you dropped them here?” The drunk replies, “No, I dropped them over there. But the light’s better here.”

eCommerce marketing in 2025 can be like that. A lot of things can be.

For nearly a decade, the industry has looked for growth where it’s easiest to measure: Google Shopping, last-click attribution, and ROAS targets set by finance with total confidence. Everyone knows this approach is incomplete, but few actually try to fix it.

John Readman has focused his career on this challenge. He’s the CEO and founder of Modo25, which created ASKBOSCO, a platform that connects brands’ and retailers’ data to predict where their marketing budget should go.

He started in direct mail, moved into email, then search, and has worked in eCommerce since before it even had that name. In short, he’s seen every version of this mistake.

The love triangle nobody admits they’re in

John’s framing of the forecasting problem is worth repeating. Most brands have a revenue target. They hand it down. The CMO takes it. The digital team takes it. The agency takes it. Everyone pulls their respective levers.

The issue is that no one in this chain has clear incentives. Agencies have their own goals. Media platforms like Google, Meta, and TikTok are basically salespeople with great data.

“If you go for a meeting with Google,” John says, “the answer is normally to spend more money on Google.”

This isn’t being cynical. It’s simply how the system works.

So you end up with a brand trying to grow 10% because someone at the top needs 10% growth, while every external party involved is quietly nudging them toward the channel that suits them best. It’s not a media plan. It’s a negotiation with people whose incentives don’t align with yours.

Key takeaway: Media partners aren’t neutral advisors. Start your forecasts with your own data, then bring them into the process.

ROAS helped you get this far, but it won’t take you further

Performance marketing brought something genuinely useful, but then everyone became hooked on it.

When you could spend £1 and make £10 on Google Shopping, you kept spending pounds. Completely rational. Finance loved it. The feedback loop was immediate and legible. CFOs, who by nature distrust things they can’t put in a cell, became converts to paid search.

Then Google needed to fund Gemini, and the auction became more competitive. Everyone rushed to the bottom of the funnel, and the £1 to £10 returns disappeared.

But expectations stayed the same. Brands started spending more to get less, while also losing the ability to track results properly across browsers, devices, privacy changes, and platforms, yet still claimed credit for the same sales.

JohnJohn points out that brands got so used to measuring everything, they assumed it would always be possible. When tracking stopped working, they didn’t change their approach. They just argued over whose attribution was wrong.

Key takeaway: If your entire growth strategy depends on paid media efficiency, you’re not building a business. You’re renting one.

Where the money is really leaking

When John looks at eCommerce marketing budgets, the pattern is remarkably consistent.

Most brands have their paid versus organic traffic ratios the wrong way around. Best practice suggests aiming for 60% organic or direct traffic and 40% paid. In reality, it’s often the opposite, or even worse. Brands pay for traffic they could earn and neglect SEO to a risky degree.

This is even more important in 2026. The way people research and buy is changing. Large language models are now part of the process. Google and Shopify have announced the Universal Commerce Protocol.

Soon, a customer might ask ChatGPT for the best trainers in their size, get a shortlist, and buy directly within the LLM. No impressions, no clicks, no cost per acquisition, but only if your products are visible and well-represented there.

Brands that have ignored SEO for years are about to learn this lesson the hard way.

Key takeaway: Compare your top PPC keywords with your SEO rankings. If you’re paying for traffic you’re not earning organically, that’s where you need to invest in SEO.

Why letting Google take over is like putting an 8-year-old in charge of a sweet shop

Ad platforms claim they can handle budget allocation automatically. You set a target, and the algorithm decides. It sounds sophisticated, but John sees it another way.

“It’s a bit like putting the 8-year-old in charge of the sweet shop,” he says. “Because they’re going to eat all the sweets they can get away with eating.”

Google’s algorithm will optimize for what it values, based on the limits you set. It doesn’t know your stock levels, margin needs, planned promotions, or which products you want to clear out. It also doesn’t know that your best-selling black trainers will sell out in eight weeks, and that you should shift budget to slower-moving stock.

This is where predictive analytics comes in. It’s not about AI as a magic solution, but as a tool that can process huge amounts of data, marketing channels, stock, pricing, returns, margins, and create scenarios you can actually review. Where is there room to grow? If you need to hit a revenue target, what media mix will get you there? If you have a set budget, how should you allocate it to maximise profit rather than just revenue?

Key takeaway: Automation is helpful, but giving the platform full control of your budget isn’t automation. It’s giving up responsibility.

Scenario planning is the grown-up conversation

Most eCommerce budgeting starts from the top: someone wants 10% growth, so everyone else works out how to deliver it. John suggests starting from the opposite direction.

Start with the data. What’s the real opportunity? Which channels have room to grow? Which products have good margins? Which customer segments haven’t been fully explored? Instead of asking, “How do we hit this number?” ask, “What number can we realistically reach, and what would it take to get there?”s simple.

In practice, it requires connecting data that lives in separate systems, marketing platforms, inventory management, and finance, and having the scenario modelling capability to run multiple futures simultaneously.

John describes it as a roulette table: you can see where to move your chips to give you the best chance of the outcome you’ve defined. The outcome can be maximum revenue, maximum profit, best ROAS, or lowest cost per new customer. The constraints go in; the recommended allocation comes out.

A person still needs to enter the plan into the platforms. John is clear about this. Full automation isn’t available yet, and more importantly, people don’t yet trust it. It’s like driverless cars. They may be safer, but people are still uneasy. The data has to prove itself before companies will let it take over.

Key takeaway: Scenario planning isn’t a strategy document. It’s a modelling exercise. Run the numbers before you commit your budget, not after.

AI in forecasting (and it’s not just ChatGPT)

John makes a point that more people in eCommerce should hear. When most people hear “AI,” they think of large language models like ChatGPT or Gemini chat. Their data scientists, meanwhile, reportedly “come out in a cold sweat. They’ve been using it for years in different ways.”

AI in forecasting is different. Statistical machine learning models process years of historical data, demand signals, impression share, seasonal trends, and competition. The LLM layer helps you ask questions in plain language, like “What happened yesterday? What should we do tomorrow?” But the real work isn’t about generating text. It’s about analysing massive amounts of data faster than ever before.

This difference matters because brands are making AI investment decisions without really understanding the technology. The most valuable forecasting tools aren’t the chatbots; they’re the models behind the scenes.

Key takeaway: Make sure you understand what AI is actually doing in any tool you use. If a vendor can’t explain the model, be skeptical of the results.

Practical applications: things you can try this week

John’s practical tips covered a lot of ground.

If you’re in retail, Google’s AI video generation (Veo 3/Nano) has reached a new level. One retailer used it on a 60,000-SKU shoe catalog, running product images on white backgrounds through the model to create lifestyle shots. In a blind test against real photos, no one could tell the difference. If budget was your reason for not having contextual product imagery, that excuse doesn’t hold up anymore.

If you struggle with long documents (John does too), try NotebookLM. Upload a document, get a podcast-style summary, and ask questions about it in conversation. It’s great for due diligence, reports, or anything else you don’t have time to read in full.

If you use Shopify, ASKBOSCO is launching a native app that brings marketing, product, stock, returns, and inventory data together in one place. The goal is to finally give marketing and merchandising teams a shared view: which products to promote, which to hold back, what’s selling, and why. This shared view addresses a big reason for wasted spending.

Key takeaway: The gap between what marketing knows and what operations knows is costing you money. Close that gap.

This article was written with the assistance of AI.