The 50% shakeout: why half of ecommerce agencies may not survive AI

Half of us won't be here

At the EAG agency retreat in Gran Canaria, straight after the Seville summit, Rachel Jacobs, founder of Ecommerce Agency Growth, opened with a prediction. Around half of today's ecommerce agencies won't exist in three to five years.

Rachel has spent a decade consulting with thousands of agencies, and she's watched plenty of them die already, mostly of ordinary causes like a lost anchor client or a founder who got bored.

Her point was that AI is about to make those ordinary causes arrive much faster.

I've co-hosted these events with Rachel and Tiara Tragas all year, from an agency retreat in Colombia in January through Austin, Australia, Edinburgh and Seville. Looking back through my notes, Gran Canaria wasn't where the split started. It has been widening in front of us since the spring.

How the numbers break down

Her split was roughly 20% of agencies investing in AI with real intent, another 40% who'll eventually copy whatever the leaders do, and a remaining half drifting towards the exit. Of the survivors, she thinks only 5 to 10% will actually lead.

She called it a hypothesis, and it is one, but the gap between the top and the middle is already measurable.

One enterprise agency in the room shared its numbers. It spends $30,000 to $50,000 a month on AI, has made early-stage R&D work 20 to 50% more efficient and repeatable processes 30 to 100% more efficient, and an internal survey found 63% adoption across its 147 staff.

Most agencies I talk to have a ChatGPT Team licence and a Slack channel called #ai-ideas that nobody has posted in since March.

We've been watching it happen all year

My notes start in Austin in April. Since then, the same question has come up at every event: what happens to an agency's services once AI can do the work?

A month later in Australia, Klaviyo's partnerships team ran a quick show of hands. Most agencies in the room used AI to run their own operations, but far fewer had packaged it into something a client could buy. Klaviyo's view was that using AI internally is now the minimum and the growth is in selling it to clients. Agencies there were also starting to be asked how AI efficiencies would show up in their pricing.

By Edinburgh in July, only three or four agencies in a room full of agency leaders were offering AI as a service to clients. Everyone else was using it to make their existing services slightly cheaper to deliver.

That's a problem, because if AI only makes you faster at what you already sell, the client eventually pockets the difference.

Also in Edinburgh, Grant from Grantbot gave the room a test I keep coming back to: if your best AI user quits tomorrow, does their work stay? In most agencies it doesn't, because it lives in one enthusiast's folder of prompts and none of it belongs to the business.

The cost of sitting it out

On a panel in Seville, I said that doing nothing now is worse than doing nothing last year. Another founder put it better: the baseline for quality is rising ridiculously fast.

The problem is compounding. An agency that started properly twelve months ago has a year of documented processes and working prompts, and you can't buy that back in a quarter. Mac King, co-founder of Domaine, made the same point on stage in Australia, warning that agencies who start late may end up having to buy their way back in.

Pricing is usually where this shows up first. An ops lead at the workshop said tasks are now finishing in a fraction of their estimated time, which means an agency selling hours gets paid less the better its AI works. Effort-based KPIs stop meaning much when the effort disappears, so the agencies ahead of this are moving to outcome-based pricing and metrics like retention and MRR.

Button-pushers are first in line

Rachel was clear about who goes first. Specialist agencies can still win as full-service shops or by being excellent at one thing, but button-pushers are in trouble.

A button-pusher runs the platform and builds the flows without owning any insight the client couldn't get somewhere else. AI happens to be extremely good at pushing buttons, and it doesn't take holidays.

The alternative she described is a data intelligence layer around your ideal client. Every audit and sales call is a data point about your niche, and if you pull them together you know things about your corner of ecommerce that nobody else does. Brands can't get that from a prompt, which is why they'll keep paying for it.

Brenden Rawson at Andzen gave the clearest version of this in Australia. Every agency has access to the same AI, from ChatGPT and Claude through to the tools built into Shopify and Klaviyo. What differs is the proprietary layer on top. His line was that AI is the engine and IP is the fuel, and he defined IP as decisions multiplied by repetition over time.

Andzen has turned ten years of audits and quotes into its own tools. Its audit assistant, trained on a decade of problem and solution pairs, diagnosed the cause of a returning client's KPI drop in about 15 minutes. Its quoting calculator prices work from ten years of real timesheets, so sales stops promising things delivery can't do.

I asked Rachel what she'd say to the founders who are sure they're in the safe half.

What the top 10% are doing

None of the agencies ahead of this had a secret tool, and most of what they described sounded unglamorous.

Adrian from Tante-e walked through his agency's "Engine Room", a Notion-based brain fed automatically from Trello, Slack and call transcripts, so anyone can ask how the agency handles something and get a real answer. His main lesson was that automating a messy process just gets you a faster mess, so fix the process and give the brain an owner before you automate anything.

The founders getting results also set the rules from the top. Plenty of agencies have a few self-taught AI enthusiasts each doing their own thing, which produces islands of clever work nobody else can use. One speaker put AI adoption at 70% people and 30% technology, with the founder defining a framework for each department and internal champions working inside it.

They also check that speed isn't costing them quality. The enterprise agency tracks AI's impact across the whole project lifecycle, has built a 13-stage agent to handle low-priority bugs and now treats scoping as structured data. In Australia, Angus Cowan showed an AI command centre that watches benchmarks across every account and reads sentiment across Slack, email and calls, so leadership can spot a churn risk early.

And they sell AI outcomes to clients. Domaine moved Marc Fisher's six brands onto a single Klaviyo setup in four weeks, under public pressure, and the results shared in Australia included 46% year-on-year growth in SMS revenue and a 20% cut in CRM execution time. Domaine now leads with an AI readiness audit. Elsewhere, agencies are packaging the tools they built for themselves to sell to other agencies and brands, and one is projecting a 31x return across 800 products for Alpinestars from answer engine optimisation work.

The view from Domaine

Mac King helped launch the Shopify Plus Partner Program before co-founding Domaine. I asked him where the agency's job goes next.

The view from Andzen

Brenden's demos at the EAS events have been some of the most practical sessions of the year. I asked him what building Andzen's own tools has changed.

Where to start on Monday

You don't need a $50k monthly AI budget to get on the right side of this, but you do need to start now.

Start with one process your team runs every week, write it down properly and automate it. Then work out what you know about your niche that nobody else does, and if the answer is nothing, building that knowledge is your real AI strategy. It's also worth asking whether leads would still come in if you disappeared for three months.

Rachel's 50% figure might turn out to be too high or too low, but the gap between the agencies acting on this and the ones waiting has been obvious at every event this year.