Headless Ecommerce Agency London: 9 Questions Before You Sign

A headless ecommerce agency in London builds your shopfront separately from the commerce engine behind it, joined by APIs rather than baked into one platform. A full composable rebuild typically runs $275,000 to $1m and nine to eighteen months. The harder question is not which agency you pick. It is whether you needed the rebuild at all.

I get asked about this a lot, usually by a founder or an ecommerce director who has had three agencies in and come out with three wildly different numbers. One quoted £80k. One quoted £600k. One wanted £25k just to tell them what the project was. All three used the word "composable" more than they used the word "revenue".

So, look. I have sat on both sides of this. I have been the person scoping the migration and the person called in eighteen months later when the new stack is slower than the old one. Below is what a headless build actually costs, which of the numbers flying around are real, and the nine questions I would put to any agency before signing.

What a headless ecommerce agency in London actually does

Headless means the front end and the commerce back end are decoupled. Your product catalogue, pricing, checkout and order management sit in one system. Your storefront is a separate application, usually React or Next.js, talking to that system over APIs. Composable, or MACH, extends the idea: search, CMS, payments and personalisation all become swappable services rather than platform modules.

The direction of travel is real. Gartner's strategic planning assumption is that by 2027, at least 60% of new B2C and B2B digital commerce solutions built for the cloud will align with MACH architecture principles. The MACH Alliance's 2026 Enterprise Technology Report, a survey of 600 enterprise technology decision-makers across seven markets, found 78% of organisations with fully scaled composable architecture reporting clear ROI on AI investments, against 13% of those still in early planning.

Worth noting who is speaking there. The MACH Alliance is a vendor consortium whose members sell composable software. The finding is interesting, it is not neutral. Treat it as a signal about direction, not as proof that you specifically should replatform.

What a headless commerce build costs in 2026

The most useful published breakdown I have found is Elogic's Replatforming Cost Index 2026, which aggregates their own delivery data across 500+ projects with benchmarks from nine other agencies. Every figure is in 2026 USD and every figure is an agency estimate rather than independent research, which Elogic says plainly. I am using it because it is the most transparent source in a category that is mostly guesswork, not because it is gospel.

Three numbers from that index matter more than the headline bands.

First, the platform licence is only 20% to 40% of total cost. The other 60% to 80% is implementation, ERP integration, data migration and QA. If an agency leads with licence pricing, they are showing you the small half.

Second, going headless multiplies cost by 1.5x to 2.5x against an equivalent themed build, and adds eight to sixteen weeks. That is the premium you are paying for decoupling, and it is a real number you should be able to see itemised in a quote.

Third, Elogic puts the threshold at which composable is justified at roughly $50m to $200m GMV with 15 or more integrations and mature DevOps. If you are a £10m UK retailer with a Shopify store and two integrations, you are not the customer this architecture was designed for.

For a UK sanity check on rates: the same index puts Western European developers, UK included, at $100 to $175 per hour, against $45 to $85 in Eastern Europe. A London agency quoting you a blended rate below about £80 an hour is either subcontracting or juniors. Neither is automatically bad. Both are things you should know before you sign, not after.

Most replatforming statistics you will be quoted are marketing

This is the part I would most like ecommerce directors to internalise, because it changes how you read every proposal that lands on your desk.

The single most-quoted statistic in this space is that "83% of data migration projects fail or exceed budget", usually attributed to Gartner. Elogic went looking for the primary report and could not find one. Their own index says the figure "is attributed to Gartner across hundreds of pages but has no locatable primary report", and is just as often credited to Oracle instead. Then, a few sections later, the same page uses the 83% figure anyway. That is not a dig at Elogic specifically. It is how this entire category works.

Same story with the traffic-loss number. You will see "poorly executed migrations lose 20% to 40% of organic traffic" everywhere. It traces to an August 2026 Optimum7 study, and the study is worth reading, but read what it actually measured. Optimum7 put 1,000 replatforming questions to ChatGPT, Gemini, Claude and Grok and analysed the 4,000 answers. The 20% to 40% range is what the AI models say, not what Optimum7 measured across migrations. They have done 1,000+ real migrations and that experience clearly informs the work, but the headline figure is a survey of language models, not of shops.

Here is what I would actually trust, because it has traceable provenance and it is about data migration generally rather than ecommerce specifically:

• 64% of data migrations overrun their budget and 54% overrun their timeline (Forbes, 2021)
• Data migration cost overruns average 30%, time overruns average 41% (Bloor Group)
• Only 31% of IT projects finish on time, on budget and on scope (Standish Group CHAOS research)
• Forrester puts the average enterprise replatforming cost at $2m to $5m, cited by Fastr, who sell an alternative to replatforming and so have a dog in the fight

None of those say headless is a bad idea. They say a large data migration run by people who have not done many is likely to slip, and that your contingency should reflect that. Elogic recommends 25% to 40% contingency treated as expected allocation rather than cushion. That matches what I have seen.

Nine questions to ask a headless ecommerce agency before you sign

These are ordered roughly by how much money they save you. The first one saves the most.

1. What breaks if we do nothing for twelve months?

Make them argue against their own sale. A good answer is specific and dated: your Magento 1 instance is unpatched, your ERP connector is deprecated in March, your checkout fails on iOS 19. A bad answer is about agility, future-proofing or being ready for AI. If nobody can name the thing that breaks, you have a want, not a need, and you can sequence it properly instead of doing it now at a premium.

2. Show me the redirect plan, and what happens past 10,000 URLs

Redirects are where migrations quietly bleed revenue. The Optimum7 study found 301 redirect implementation was the first recommendation in 92% of AI answers on migration, ahead of Search Console verification at 88%. That consistency means your buyers, and your competitors, already know to ask.

The specific trap: Shopify Plus caps stores at 10,000 URL redirects. If you have a large catalogue with faceted navigation and paginated archives, your indexed URL inventory blows through that ceiling and you need a wildcard and consolidation strategy designed before you pick the destination platform, not after. Ask to see the mapping spreadsheet from their last migration. If it does not exist, that tells you something.

3. Who owns the front end in twenty-four months, and what does keeping it alive cost?

A custom headless storefront starts accruing technical debt on day one. Every framework major version, every dependency bump, every new integration is now your problem rather than your platform vendor's. Ask for the annual run-rate figure for frontend maintenance after handover, in writing. If they have not thought about it, you are buying a build, not a system.

4. What is your discovery deliverable, and what does it cost on its own?

Elogic prices discovery at $15k to $50k and puts scope creep at 30% to 50% when structured discovery is skipped. Pay for discovery separately, as its own contract, with a deliverable you own and can take to another agency. Any agency that will only do discovery as part of a full build is selling you a funnel, not a service.

5. Is this a big-bang cutover, or is there a parallel run?

Running both platforms for four to twelve weeks doubles your licence and hosting cost for that window. It is also the single cheapest insurance policy in the project. If the plan is a single weekend cutover for a multi-brand or multi-market estate, push back hard and ask what the rollback procedure is, specifically, hour by hour.

6. Which parts of this are you subcontracting, and to whom?

Nothing wrong with subcontracting. Plenty wrong with finding out during week nine. Ask which agency or which country, ask whether the people in the pitch will be on the project, and ask what percentage of delivery hours are in-house. Get the answer in the statement of work.

7. Itemise the ERP integration

ERP integration is the most reliable source of overrun in this category. Pre-built connectors run $5k to $30k. Heavily customised SAP or Oracle work reaches $250k to $600k in the first year. That is a 20x spread hiding inside one line item on a proposal. Make them break it out by data flow: products, inventory, pricing, orders, customers, returns. Each one is a separate piece of work and each one can fail separately.

8. Show me Core Web Vitals before and after, for your last three migrations

Headless does not automatically make a site faster. Building a fast storefront is a specialist skill involving server-side rendering, edge caching, bundle management and image pipelines, and plenty of teams doing their first headless build do not have it. Fastr describe an enterprise brand whose mobile Largest Contentful Paint went from 2.1 seconds to 3.8 seconds after a $3m migration. That is one vendor-reported example rather than a study, so weight it accordingly, but the failure mode is common enough that you should ask for real numbers from real projects. Any agency that cannot produce before-and-after data for three recent builds has not been measuring.

9. Under what circumstances would you tell us not to do this?

My favourite question, and the one that sorts the field fastest. An agency with an opinion will give you a real threshold: under this GMV, with fewer than this many integrations, without a frontend engineer on staff, don't. An agency without one will tell you every business is different. That is a sales answer.

The typical agency pitch against a CTO-led approach

I am obviously not neutral here. But the asymmetry is structural rather than moral: an agency that only sells builds will find a reason to build. If your technical advisor and your delivery vendor are the same company, you are missing a check. That holds whether you use us, a fractional CTO, or a friend who has done three of these.

When you should not replatform at all

Four conditions. If three or more apply, put the money somewhere else this year.

• Your current platform is supported, patched and not being deprecated in the next eighteen months
• Your slow pages are slow because of third-party scripts and unoptimised images, not because of your architecture, and nobody has tried fixing those first
• You have no frontend engineer on staff and no budget for one after launch
• The business case is written in terms of flexibility and future-proofing rather than a number you can measure ninety days after go-live

That fourth one catches most of them. Frontend control, campaign speed and marketing autonomy are all real problems, and none of them strictly requires an architectural overhaul. A page-builder layer, a caching and image-pipeline pass, or a re-theme will often get you most of the benefit for a fraction of the spend. Measure your current architecture's actual limits before assuming you have hit them.

If you do decide to go ahead, treat it like an investment decision with a technical reviewer on your side of the table. We do that as part of tech assessment and due diligence, and about a third of the time the honest answer is that the migration can wait a year. If it genuinely cannot wait, our CTO-led development team builds it, and the same person who told you to do it is accountable for delivering it. More on how we work with ecommerce businesses, and on the fixed-price versus hourly question that always comes up next.

So the short version. Three agencies gave that founder three numbers because they were scoping three different projects, and none of them had been asked the first question on this list. He asked it. The answer was that nothing broke for at least two years. He spent £40k on performance work instead of £300k on a rebuild, and his conversion rate went up. That is not an argument against headless. It is an argument for asking what breaks before you ask who builds it.

Frequently asked questions

How much does a headless ecommerce agency in London cost?

Per Elogic's 2026 agency cost index, a mid-market replatform runs $80,000 to $250,000 over four to eight months, and a full composable build $500,000 to $2m over twelve to twenty-four months. Going headless adds a 1.5x to 2.5x multiplier on top of an equivalent themed build. UK developer rates sit at $100 to $175 per hour.

Is headless commerce worth it for a mid-market retailer?

Usually not yet. The commonly cited threshold where composable pays for itself is around $50m to $200m GMV with fifteen or more integrations and a mature DevOps practice. Below that, a headless frontend costs 1.5x to 2.5x a themed build and hands you permanent frontend maintenance you probably are not staffed for.

How much organic traffic will I lose replatforming?

Widely quoted ranges of 20% to 40% come from an analysis of what AI assistants say rather than measured migrations, so treat them as directional. What is reliable is that incomplete 301 redirect mapping is the leading cause of loss. Complete URL mapping before cutover and post-launch crawl validation are the variables you control.

How long does a headless commerce migration take?

Four to eight months for mid-market, nine to eighteen months for monolithic to composable, and twelve to twenty-four months at full MACH scale. Going headless adds eight to sixteen weeks over an equivalent themed build. Budget at least six months of preparation before your intended go-live date.

Do I need a headless ecommerce agency or a fractional CTO?

Get the CTO first, then the agency. A fractional CTO decides whether the migration is justified, writes the specification and reviews the agency's work. The agency builds it. Combining both roles in one supplier removes the only independent check on a six-figure decision.