Custom-Built eCommerce: When It Actually Makes Sense
By Diosh — Founder, AHAeCommerce | eCommerce decision intelligence for $50K–$5M GMV operators
A custom eCommerce build for a sub-$5M GMV operator typically costs $180,000–$450,000 in year one and $80,000–$160,000 per year in ongoing engineering, with a 14–22 month runway before the platform feature parity matches what Shopify Advanced provides out of the box. Operators who build custom anyway usually do so for one of three reasons: a structural feature dependency that no hosted platform supports, a margin profile where the platform fee math beats the engineering math, or a strategic differentiation thesis that requires control of the customer experience layer. Every other reason — "more flexibility", "we'll grow into it", "Shopify is too restrictive" — is a 24-month detour that ends in re-platforming.
This is the inverse of the typical narrative. The case for custom is not made by listing what hosted platforms cannot do. It is made by quantifying the cost of not building, against the operator's specific revenue model, growth profile, and engineering capacity. For most operators, that cost is negative — the build destroys more value than it creates. For a small minority, it is the correct decision. This piece defines the dividing line.
The Default Recommendation Is Hosted, Always
Per BigCommerce's 2024 industry report, the median time-to-launch on Shopify or BigCommerce is 6–14 weeks. The median time-to-launch on a custom Next.js + headless CMS + Stripe build with comparable feature parity is 9–18 months, validated across multiple agency-led projects in 2024 and 2025. The 8–12 month gap is pure opportunity cost — revenue not generated, marketing not iterated, customer feedback not gathered. At $1M GMV target run-rate, that opportunity cost is $670K–$1M in foregone first-year revenue alone.
The cost gap compounds in years two and three. Hosted platform costs at $2M GMV run roughly 5.5–6.5% of GMV (subscription + apps + processing + dev). Custom platform total cost at $2M GMV runs 8–12% of GMV in years one through three (engineering team + hosting + integration maintenance + DevOps + monitoring), settling to 4–6% only after the platform stabilizes — typically year three or later, if the operator survives the build cycle.
The hosted default is not a concession. It is the rational choice for the overwhelming majority of operators. The custom case must clear a high bar.
The Three Conditions Where Custom Is Correct
A custom build is justified when at least one of three conditions is durably true. "Durably" is load-bearing — temporary or aspirational conditions do not qualify.
Condition 1: Structural Feature Dependency
A core revenue mechanism cannot be implemented on hosted platforms or their app ecosystem. The test is not "is it harder on Shopify" — the test is "is it impossible or would the workaround compromise the model."
Examples that qualify:
- A subscription model with non-standard billing logic (e.g., dynamic pricing per cycle based on consumption metadata, partial-period proration with credit accruals, multi-product bundles with independent renewal schedules) where Recharge, Bold, and Loop's logic cannot represent the rules
- A B2B platform requiring per-customer pricing tiers, contract-bound minimum order quantities, ERP-driven inventory promising, and complex tax exemption certificates — capabilities Shopify Plus B2B handles partially but not fully for industries with regulatory documentation requirements
- A configurable product where the configuration logic is a pricing engine (e.g., custom apparel with 200+ variant dimensions priced via formula, not lookup) that breaks the variant model on hosted platforms
Examples that do not qualify, despite operators believing they do:
- "We need a custom checkout" — Shopify Plus checkout extensibility (released 2024) covers ~85% of customization requests
- "We have unique product attributes" — metafields and meta objects handle this on Shopify and BigCommerce
- "Our bundle logic is complex" — PickyStory, Bold, and native Shopify bundles handle 90% of bundle scenarios
The error pattern: operators classify "harder on hosted" as "impossible on hosted." The correct test is whether the workaround compromises the revenue mechanism, not whether the workaround is annoying.
Condition 2: Margin Profile Where Platform Fees Dominate
The hosted platform fee structure is regressive at very high volume. Above approximately $25M GMV, the dollars paid to Shopify Plus, transaction fees, and the app stack become a meaningful percentage of contribution margin. Below $5M GMV, those same costs are a rounding error against engineering team cost.
The break-even calculation: a custom build's engineering team runs $80K–$160K/year ongoing. To beat this on platform fee savings alone, the operator needs to be saving more than $80K/year on hosted platform costs that the custom build eliminates. At a hosted total cost of 6% of GMV, the saving is meaningful only when GMV is high enough that 1–2 percentage points of that 6% — the share that engineering can actually displace — exceeds team cost. That threshold is $8M–$15M GMV depending on team structure.
Below $8M GMV, the engineering team is more expensive than the platform fees being eliminated. Above $25M GMV, the calculation usually inverts and custom becomes defensible on cost alone — though even there, most operators prefer enterprise hosted (Shopify Plus enterprise tier, commercetools, Salesforce Commerce Cloud) because the cost differential narrows and the time risk does not.
This condition rarely holds for sub-$5M operators. When operators in this range claim it does, the model is usually wrong.
Condition 3: Strategic Differentiation Through Experience Control
The product or brand thesis depends on a customer experience that hosted platforms cannot enable, where the experience is the moat, not the operational backend. This is the rarest and most defensible case.
Examples that qualify:
- A content-commerce hybrid where the editorial layer and commerce layer are deeply intertwined (e.g., long-form storytelling with contextual product placement, where the storytelling format is bespoke and ranks on SEO based on the specific structure)
- A community-commerce model where the social/UGC layer is the primary acquisition channel and platform constraints on user-generated content limit the model
- A marketplace or multi-vendor platform where the platform itself is the product, not the storefront
Examples that do not qualify:
- "We want a unique design" — themes and headless-light implementations on Shopify handle this
- "We want a faster site" — hosted platforms with proper development hit Lighthouse 90+ regularly
- "We want better SEO control" — Shopify's SEO controls are sufficient for top-3 rankings in any commerce category
The differentiation must be in the experience itself, not in the operational details that hosted platforms also support.
The 20-Minute Decision Tree
Run this test before any custom build conversation continues.
| Question | Answer If Yes | Answer If No | |---|---|---| | Can the core revenue mechanism be implemented within hosted platform + app ecosystem with ≤15% efficiency loss vs. ideal? | Custom not justified on Condition 1 — proceed to Q2 | Custom may be justified on Condition 1 — quantify the workaround compromise | | Is current GMV above $8M and growing >40% YoY, with hosted platform fees + app stack > $400K/year? | Custom may be justified on Condition 2 — model engineering team cost | Custom not justified on Condition 2 — proceed to Q3 | | Is the customer experience itself the differentiation thesis (not the design, not the speed, not the SEO — the experience)? | Custom may be justified on Condition 3 — pressure-test the experience hypothesis | Custom not justified on Condition 3 | | If 0/3 conditions: do not build custom. The cost will exceed the value. | | | | If 1/3 conditions: build custom only if the case is unambiguous and engineering capacity exists. | | | | If 2-3/3 conditions: custom is the right call — proceed with full cost modeling. | | |
The vast majority of operators answer "no" to all three questions. They should not build custom. The hosted-platform discomfort that drove the conversation is real but addressable through better app stack governance, better dev resourcing, or a partial headless approach (Shopify Hydrogen, BigCommerce + Next.js storefront) that retains the hosted backend while customizing the front-end where it matters.
The Cost Reality at $2M GMV
A representative custom build cost model for an operator in this range, validated against agency-led projects in 2024–2025:
Year One
| Item | Cost (USD) | |---|---| | Initial build (Next.js, headless CMS, Stripe, custom backend) | $180,000–$450,000 | | Migration (data, SEO, customer accounts, content) | $25,000–$80,000 | | Hosting + infrastructure (Vercel, AWS, monitoring) | $18,000–$36,000 | | DevOps + observability setup | $20,000–$40,000 | | Testing and QA | $15,000–$30,000 | | Year 1 total | $258,000–$636,000 |
Years Two and Beyond
| Item | Annual Cost (USD) | |---|---| | Engineering (1.5–3 FTE: full-stack + DevOps fractional) | $180,000–$450,000 | | Hosting + infrastructure | $18,000–$48,000 | | Third-party services (auth, search, analytics, email, monitoring) | $24,000–$60,000 | | Maintenance + technical debt amortization | $30,000–$80,000 | | Year 2+ annual total | $252,000–$638,000 |
For the same operator, hosted platform total cost at $2M GMV is approximately $112,000/year (per the modeled stack including apps, processing, and dev). The custom delta — $140,000–$526,000/year — must be justified by either Condition 1 revenue impact, Condition 2 fee elimination at higher scale, or Condition 3 experience differentiation that produces measurable acquisition or retention lift.
If the differentiation thesis cannot show $200K+/year in incremental contribution margin at the operator's current scale, the build is uneconomic.
Worked Decision: A $1.8M GMV Operator with a "Configuration Engine" Argument
A scenario from advisory work in 2025:
- Operator: Custom outdoor gear, $1.8M GMV
- Argument for custom: "Our products have 12–18 configurable dimensions; Shopify variants are too restrictive"
- Proposed build: $320K initial, ~$220K/year ongoing
The model:
- Variant solution on Shopify: metaobjects + a third-party configurator app (Zakeke, Spiff, or custom Shopify Function) — modal cost $220–$580/month, full-fidelity custom configurator
- Workaround compromise: The configurator UX is ~85% of ideal — slightly slower, slightly less integrated with cart logic — but does not compromise the revenue mechanism. Conversion rate on configurator pages with Spiff was within 5% of conversion projected for a custom-built equivalent.
- Cost differential vs. hosted: $220K/year custom engineering vs. $4K/year configurator app = $216K/year delta
- Revenue lift required to justify: at 35% contribution margin, the delta requires $617K/year in incremental revenue — a 34% GMV lift attributable solely to configurator superiority
Verdict: Do not build custom. The configurator app captures 85% of the value at 2% of the cost. Reinvest the saved capital in marketing and product expansion. Revisit at $8M+ GMV.
This is the modal pattern. Operators identify a real but bounded constraint, mistake it for a Condition 1 dependency, and propose a $200K–$500K build to recover 10–15% efficiency. The math never works at sub-$5M GMV.
When the Math Does Work
The custom builds that succeed at sub-$5M GMV share a common profile:
- Founder is technical and either builds the platform or runs the engineering function in-house at low cost
- Revenue mechanism genuinely cannot be implemented on hosted (Condition 1) and the operator has validated the constraint with prototypes on multiple hosted platforms before committing
- Build is incremental — start with hosted platform, replace components as constraints actualize, end with a hybrid headless architecture rather than a ground-up rewrite
- The custom layer is narrow — typically the storefront and configuration/pricing logic — not the entire stack
The failure pattern is the inverse: non-technical founder, vague Condition 1 argument, ground-up rewrite, $300K+ initial spend, 14-month runway, missed launch by 6+ months, re-platform back to Shopify in year 3.
The Verdict
For operators below $5M GMV, the answer is hosted, with rare exceptions. The exceptions are not "I want flexibility." They are: a structural feature dependency the workaround compromises the model, a margin profile where engineering team cost is less than platform fee elimination, or a strategic differentiation thesis where the experience itself is the moat.
If none of those three conditions hold durably, the custom build will cost $250K–$600K in year one and $200K–$500K annually thereafter, with a 12–20 month opportunity cost on revenue. That capital reinvested in marketing, product development, and operational improvement will produce more lift than the build will ever pay back.
If one or more of the conditions hold, custom is correct — but the build should be incremental, not ground-up, and should preserve hosted-platform leverage wherever the differentiation thesis does not require ownership.
This Week
Run the 20-minute decision tree. For each of the three conditions, write out — with specific numbers, not adjectives — whether your business clears the bar. If you cannot quantify the answer in numbers, the answer is no. The exercise will save most operators reading this between $200K and $600K in misallocated capital, and the time it costs to do it is less than the time spent in the next custom-vs-hosted internal debate.
