Questions buyers ask before hiring a WordPress engineer.

JSON-LD + llms.txt + speakable + 50-word answer blocks. With code.

Why this matters in 2026

AI search assistants (ChatGPT, Claude, Perplexity, Gemini) are now a non-trivial source of B2B referral traffic. Sites optimised for traditional SEO can show up in their answers, but the citation rate is 3 — 5× higher for sites with structured AEO setup.

Four-part setup

1. JSON-LD per content type. FAQPage for FAQ entries, Article for insights, HowTo for tutorials, Service for service pages. speakable property on the 50-word answer block.
2. 50-word answer blocks. Every long-form post starts with a 50-word answer block that completely answers the question. Crawlers and AI agents lift this as the citation.
3. llms.txt at the root. A markdown index of your highest-quality content with summaries. Some agents read it; the rest will soon.
4. Refresh-date metadata. Visible in the UI and in JSON-LD. AI agents weight freshness heavily for cost / regulation / pricing queries.

What I don’t recommend

“AEO plugins” that promise instant citation. They generate boilerplate JSON-LD with no semantic structure. Hand-write the schema; it’s not that much code.

Full essay with the schema templates ↗

2 — 4 weeks for a real fix vs cache-only fixes.

The 4-day cache “fix”

Install WP Rocket, enable Cloudflare APO, claim victory. LCP improves by 30 — 50% but the underlying problems remain. The next plugin update or content change regresses it. This isn’t a fix.

The 2-week real fix

Plugin teardown (remove 5 — 12 plugins), inline above-the-fold CSS, defer the rest, async fonts, image pipeline rebuild (WebP/AVIF + responsive srcset), edge caching configured per page type. Result: sub-1 s LCP that holds through content updates.

The 4-week thorough fix

Above plus theme refactor to remove render-blocking JS, third-party tag audit (most marketing sites ship 8 — 14 unnecessary tags), and a Lighthouse CI guardrail so regressions get caught in the PR not in production.

The fix that holds

The fastest LCP I’ve shipped is 0.43 s on mobile p75 for a regulated-finance site. It took three weeks. It’s been holding for 18 months because the team kept the Lighthouse CI guardrail enforced.

Scope · exclusions · milestones · IP · subcontracting · indemnity. Sample SOW linked at the bottom.

The six required sections

1. Scope. Every page, every template, every CPT, every form, every integration, listed by name. No “etc.”
2. Exclusions. Explicit out-of-scope list: copywriting, photography, branding, paid ads, hosting fees, third-party plugin licences, content migration beyond N pages. This prevents 80% of scope creep.
3. Milestones. Day-one deposit, midpoint deliverable, handover. 40 / 40 / 20 is standard.
4. IP transfer. “All deliverables transfer to Buyer on final payment. No perpetual licence retained.” One sentence.
5. Subcontracting. “Vendor will not subcontract any portion of the work without naming the subcontractor and price in a written addendum.” Prevents the bait-and-switch.
6. Indemnity. Both directions, capped at total fees paid. Standard.

What shouldn’t be in the SOW

Hourly rates (you’re paying for outcomes), “unlimited revisions” clauses (they’re either a lie or a trap), and rights-grant language that lets the vendor reuse your code.

Download sample SOW ↗ · Drop-in template, fill in scope, ship in an hour.

40 / 40 / 20. Why 50 / 50 is worse than it looks.

The 40 / 40 / 20 case

• 40% on SOW signature. Covers discovery, design decisions, initial scaffold. The vendor is committed; the buyer is committed; nobody’s negotiating in week two.
• 40% at midpoint. Delivered against a written milestone: e.g., “all templates rendering on staging with placeholder content; admin UX functional.” If the milestone slips, the payment slips.
• 20% on handover. Loom walkthrough delivered, runbook delivered, repo transferred, warranty period begins.

Why not 50 / 50

Half the budget is leverage. If half is held to the end, the buyer’s leverage is concentrated in a single moment — handover. That moment is also when the vendor is most tired, the scope ambiguity is most evident, and the buyer has the least context on what’s actually been delivered. 50/50 looks fair; in practice it creates the worst handover meetings.

Why not 33 / 33 / 33

The middle milestone is too small to motivate. By month two of a four-month build, the vendor is operating on momentum, not invoice schedule. 40% at midpoint matches the risk profile.

The retrofit playbook. Manual checklist + axe-core + the five plugins that won’t pass NVDA.

Five-step retrofit

1. Run axe-core on every template. Fix the automated issues first (40% of the work). Don’t waste manual SR time on issues a script can find.
2. NVDA + keyboard pass on five key flows. Homepage, primary CTA, contact form, search, an inner content page. Document by template, not by issue.
3. Replace the five plugins that don’t pass NVDA: Elementor’s pop-ups, Slider Revolution, Smush’s lazy-load fallbacks, WPForms’ Smart Tags, and any “accessibility plugin” claiming instant compliance. List of theme-code replacements available on request.
4. Lock down the editor. Every block needs an alt-text field that’s required, not optional. Every link gets a “where does this lead?” prompt for screen-reader users.
5. Document the conformance. Statement page, evidence trail (axe-core CI logs, manual test dates, tools used), known-issues log with target remediation dates.

What you don’t have to do

Tear out the theme. AA is achievable on any well-written theme. The decision is whether the theme is well-written — which is a separate audit.

Three hosts benchmarked. Honest about why “best” is the wrong question.

The benchmarks

All three tested with the same site (this one, actually), same plugin set, same CDN config. LCP measured at mobile p75 over a week, location US + EU + Africa.

Why “best” is the wrong question

The biggest LCP gains come from your site, not the host. A poorly-built site on the fastest host is slower than a well-built site on a cheap host. The right question is “what’s the slowest host my site can tolerate?” — because anything past that is wasted spend.

What I actually recommend

Kinsta for B2B service sites that need editorial confidence and live support. Rocket.net if budget is tighter. Cloudflare Pages + headless if you’re willing to operate a Next.js front-end.

Blocks are enough for 90% of B2B marketing sites. Specific exceptions below.

The case for blocks-only in 2026

Gutenberg + ACF + theme.json now covers: column layouts, conditional visibility, custom CSS per block, reusable patterns, query loops, content sync between blocks, locale-specific variants, and full design-token integration. Five years ago, page builders won on every one of these.

When you still need a page builder

• Marketing teams that ship 4+ landing pages per month with novel layouts. The page-builder workflow trades long-term technical debt for short-term velocity. Sometimes that’s the right trade.
• Sites with no dev team. If the editor can’t open a block.json file, the page builder is the only sustainable answer.
• Sites where the brand changes every quarter. Page builders let the editor reshape blocks without dev support; custom blocks lock the layout to dev cycles.

The path most teams should take

Custom theme on Gutenberg + ACF, with two or three “flex blocks” that the editor can compose freely. This is the best of both worlds for the 90% case. Classic vs FSE essay ↗ goes deeper on the tradeoffs.

12 active for a B2B marketing site. 20+ is debt. 38 is what happens when nobody owns it.

The shape of the 12

From real B2B WP installs that pass audits: 1 form plugin (CF7 or Gravity), 1 SEO plugin (Yoast or Rank Math), 1 backup plugin (UpdraftPlus or BlogVault), 1 security plugin (Wordfence or iThemes), 1 caching plugin (WP Rocket), 1 image plugin (Imagify or none if your host handles it), 1 ACF Pro (or Free + native fallback), 2 — 4 integration plugins (HubSpot, Mailchimp, Stripe, etc.), and 1 — 2 editorial workflow plugins (Stream, Pretty Links).

What 20+ plugins means

Every plugin is a dependency, an attack surface, an update obligation, and a render-time tax. Twenty plugins means twenty release cadences your team has to track. It also means twenty admin UIs no editor has been trained on.

What 38 plugins looks like

Three competing slider libraries because nobody removed the previous one. Two analytics endpoints (one broken). A “performance” plugin actively fighting the caching plugin. An accessibility plugin claiming instant compliance. This was the Cyberhive site in October. Read the rebuild ↗.

The plugin-diet rule

Before adding a plugin, name the one you’ll remove. If you can’t, the plugin doesn’t get added.

Cloudflare + WP Rocket + object cache. The order matters.

The stack, top-down

1. Cloudflare (edge). APO enabled. Page Rules for cache-by-cookie on logged-out traffic. HTTP/3, Brotli, Polish for image optimization. £4/mo for Pro tier if you need real APO.
2. WP Rocket (page cache). Page caching with cache preload, mobile-specific cache, lazy-loading off (Cloudflare does it better), critical CSS inlining on.
3. Redis object cache (server-side). Most managed hosts ship this now. Persistent object cache is the biggest single TTFB win on dynamic pages.
4. OPcache (PHP). On by default in PHP 8+; verify opcache.jit_buffer_size is set.

What kills the stack

Two page-cache plugins fighting. A “performance” plugin that minifies CSS in a different order than Rocket. A WP-Admin tool that bypasses cache for “fresh” logged-out previews — and shows it to everyone. Every fight between layers is a regression waiting to happen.

The cheap version

Cloudflare free tier + WP Super Cache + your host’s object cache. Sub-1 s LCP achievable. The £30/mo for Rocket + Cloudflare Pro is worth it if you’re paying yourself anything per hour for the maintenance.

4 — 12 weeks. The variability is editorial parity, not engineering.

The 4-week migration

Single-locale marketing site, 4 — 6 templates, no e-commerce, no logged-in surfaces. Next.js front-end, ISR, image CDN. Possible if (and only if) editorial parity is established in week one — every existing component documented and reproducible in the new front-end.

The 8-week migration

Above plus two locales, editorial workflow with role-based access, third-party form integration, full structured-data round-trip. Two weeks of the eight are spent on editorial parity QA.

The 12-week migration

Above plus WooCommerce checkout (this is the variability), regulated-content workflows, multiple author roles, complex editorial states. The checkout flow is the long pole.

What blows the timeline

Discovering on week six that the existing site has a block the new front-end can’t reproduce. The fix is week one: catalogue every state, every conditional, every editor pattern. If you skip this, you’ll be re-quoting in week four.

Honest answer for each. The vendor risk is now the dominant risk.

WP Bakery — no

The vendor effectively abandoned active maintenance in 2024. Updates are security patches only; no new feature work. Every WP minor release breaks something. If you have a WP Bakery site, plan the migration; don’t plan another year on it.

Elementor — yes, but with conditions

Active vendor, real release cadence, large community. But: the rendered HTML is heavy, the editor is slow on sites with 50+ pages, and the “Pro” tier creates a real licensing dependency. Elementor is fine for marketing sites that ship campaigns; it’s painful for marketing sites that ship considered content.

Divi — no, for B2B service sites

Elegant Themes is a single vendor with a single-pricing model and a track record of opinionated breaking changes. If they ship a “Divi 6” that requires a rebuild, your site rebuilds with it. For an agency that ships 30 marketing sites a year, that’s fine. For a B2B service business with one site, it’s risk concentration without upside.

The general rule

If your site is going to live longer than three years, the page-builder vendor risk compounds against you. Custom theme is the de-risking option. 5-year TCO comparison ↗.

Three specific signals. The migration cost most teams underestimate.

Signal one: your editorial team is two people who never edit

If marketing ships content once a quarter and dev ships site changes weekly, the WP CMS is the wrong tool. Astro or Next.js with MDX is a better fit. Migration cost: 4 — 8 weeks.

Signal two: you’re spending 30%+ of dev time on WP overhead

Plugin updates, theme conflicts, hosting issues, security patches. If this is a meaningful slice of your team’s calendar, the operational tax has exceeded the editorial benefit. The right answer depends on your editorial team: Webflow for non-technical editors, Next.js + Sanity for technical ones.

Signal three: regulatory load is forcing edge-deployment guarantees

Some regulated industries (finance, healthcare, public sector) require edge-deployment with strict data-residency. WP’s traditional hosting model isn’t a great fit. Headless WP or fully off-WP is the right move.

The cost most teams miss

Editorial parity. The CMS that replaces WordPress has to support every editorial state your existing content depends on. For a five-year-old marketing site, that can mean reproducing 40+ block types, 12 page templates, and the editorial workflow your team has trained on. Budget 30 — 50% of the migration cost for this — most teams budget 10%.

Twelve questions, the right answers, and what AI portfolios can’t fake.

The twelve questions

1. “Show me the last three sites you shipped, ordered by recency.” — Real sites, not screenshots. Public URLs.
2. “What’s the Lighthouse score on the homepage of each?” — They should know without checking.
3. “What’s the LCP at mobile p75?” — Same. Public PageSpeed Insights URLs are stronger than any portfolio claim.
4. “How many plugins on each?” — Single digits is the right answer for a B2B marketing site.
5. “Do those sites pass WCAG 2.2 AA?” — Specific answer, with evidence.
6. “Show me a SOW you’ve used.” — If they don’t have one, hard pass.
7. “How do you handle ‘while you’re in there’ scope creep?” — Looking for a process, not just willingness.
8. “What’s your milestone schedule?” — 40/40/20 is healthy. 50/50 is risk-loaded to the end.
9. “Who owns the code at the end?” — “You do, on final payment” is the only acceptable answer.
10. “What’s your communication cadence?” — Friday written status is the gold standard.
11. “Can I talk to your last client?” — If the answer is no, hard pass.
12. “What’s your AI policy?” — Looking for transparency, not absolutism. AI as draft is fine; AI as final review is not.

What AI portfolios can’t fake

LinkedIn-verifiable client testimonials. Public PageSpeed Insights URLs from sites you can independently visit. A SOW with an actual signature. A GitHub history that goes back more than 12 months. References who answer.

Eight tells. The “perpetual licence” clause is at the top.

The eight

1. “Perpetual licence” clause in the SOW. The vendor wants to reuse your code, your design, or your content in their template library. Decline; this is a one-way value transfer.
2. No fixed milestones. “We’ll invoice as work progresses.” Translation: scope is negotiable forever, and your leverage decays.
3. “Unlimited revisions” promise. Either a lie or a trap. Real engineering has a finite revision count documented per milestone.
4. No SOW at all. “I work on trust.” Walk.
5. “My hosting” or “my staging.” The vendor owns the infrastructure the site lives on. You’ll be hostage at handover.
6. Generic LinkedIn / website. No client testimonials with real names. No portfolio with public URLs. AI portfolios are good at this now; reference-checking is the antidote.
7. “Email me with any questions, but don’t expect a reply for a few days.” Translates to a 5 — 10 business day SLA. Make sure the SOW makes that explicit, then decide if it works for you.
8. No exit clause. If you can’t fire them mid-engagement and walk away with what’s already shipped, you’re locked in. Decline.

What the green-flag engagement looks like

Fixed SOW, 40/40/20 milestones, written Friday status, public client portfolio with real URLs, IP transfers on final payment, named indemnity, the vendor’s hosting recommendation is on your account.

Concrete contract clauses. The “Friday update” pattern.

The five contract clauses

1. 1-business-day inbox SLA. “Vendor will respond to any communication within one business day. Vendor will respond to a site-down emergency within 4 hours.” With teeth: missed SLA = the next milestone is free.
2. Weekly written status. “Vendor will deliver a written status update by 17:00 UTC every Friday. Five sections: shipped, open, risk, decision needed, next week.” Missed week = same penalty.
3. Concurrent-cap declaration. “Vendor confirms that at the time of SOW signature, Vendor is engaged on N concurrent projects, and will not exceed N+1 during the engagement.” Public count on the vendor’s site is the gold standard.
4. Code repo access from day one. The buyer has read-access to the repo from the first commit. You can verify activity even if the vendor goes quiet for a day.
5. Exit clause. If the vendor goes quiet for more than two consecutive business days without notice, the buyer may terminate and retain all work-to-date with no further payment owed.

The Friday update pattern

I send a five-section written status every Friday by 17:00 UTC. Five sections, same template, every week. The discipline keeps clients informed even when I’m heads-down on a build week. It’s the single best contract clause for preventing the “what’s happening?” panic.

Professional indemnity, public liability, IP transfer clauses. Numbers.

Professional indemnity (PI) insurance

The standard for UK-based freelance WP devs is £1m cover via Hiscox or Markel. Annual premium £450 — £900. Required by most enterprise procurement; useful for everyone else. Ask for the policy number on request — you should be able to verify it independently.

Public liability insurance

£2m typical via the same underwriters. £150 — £300/year. Not always required for purely digital work, but useful if the engagement involves any on-site work.

IP transfer clauses

“All deliverables transfer to Buyer on final payment. No perpetual licence retained by Vendor over Buyer’s site, brand, or content.” One sentence. If the vendor’s SOW says anything weaker than this — like “Vendor retains rights to component code for portfolio use” — negotiate.

Subcontracting clauses

“Vendor will not subcontract any portion of the work without naming the subcontractor and their price in a written addendum approved by Buyer.” Prevents the work being silently handed to a less-skilled subcontractor at the original price.

Indemnity

Both parties indemnify each other against losses arising from the other party’s breach, capped at total fees paid under the SOW. Standard, balanced, and protects both sides.

A buyer’s framework for asking the right questions.

The four roles AI can play

1. Scaffold generator. AI writes the initial structure; human edits every line. This is the safest, most common pattern in 2026.
2. Refactor pair. AI suggests refactors on existing code; human reviews and accepts each suggestion. Slightly riskier — the model can introduce subtle behavioural changes.
3. Test generator. AI writes test cases for human-written code. Useful, but watch for tests that just memorise the implementation.
4. Autonomous agent. AI writes and merges code without per-line human review. Should not be happening on your project in 2026.

The questions to ask

“Which of these four does your AI use look like?” — You want roles 1 and 2 mostly, 3 occasionally, 4 never. “Show me an example commit where AI scaffolded and you edited.” — Real commit, real diff. “What’s your AI policy on customer data?” — Customer code should never go to a third-party model without explicit consent. “Do you use a local model for sensitive work?” — Bonus points if yes.

The red flags

“AI does everything; I just review.” If the human review is the bottleneck (and it should be), the vendor can ship faster than human review allows. That’s autonomous-agent mode in disguise. The code looks fine and breaks in production.

My AI policy ↗ documents what AI touches on every project and what it does not.

Six tells. LinkedIn-verified testimonials are the floor.

What AI portfolios get right

Great-looking screenshots. Plausible client names. Generic-but-correct descriptions. Visual consistency across the portfolio. If you stop here, you cannot distinguish.

The six tells

1. Public URLs you can visit. Real client work is at a real domain. Visit each one. AI portfolios use mockup URLs or expired domains.
2. LinkedIn-verifiable testimonials. Real quote, real name, real job title, real linked profile. The testifier should be findable independently and willing to respond to a DM.
3. Public PageSpeed / Lighthouse reports. Run a fresh PageSpeed Insights against each portfolio URL. The scores in the portfolio should match within a small margin.
4. GitHub or BitBucket history. Open-source contributions, your-username/repos. AI portfolios can’t fake 18 months of consistent commits at variable hours.
5. References who answer. Ask for two references. Email both. Both should respond within 2 — 3 business days.
6. Live conversation. A 30-minute call. AI portfolios are run by humans, but those humans can’t fluently answer “tell me about the hardest decision on the Cyberhive build.” Specific stories with timestamps and names are the unfakeable evidence.

The minimum bar

Two LinkedIn-verifiable testimonials with public domain URLs you can independently audit. Anything less is a guess.

£4,500 — £14,000. Gift Aid + accessibility eat the budget in unexpected ways. Honest version below.

The £4,500 floor

• Single-locale marketing site, four to six template types.
• Stripe + GoCardless donation flow with Gift Aid declaration capture.
• WCAG 2.2 AA conformance documented (most public-sector funders now ask for the audit report).
• Editor training (60 min Loom) so volunteers can run it.

What pushes you toward £14,000

A volunteer portal (login-gated content with role-based access), a grant-application form with file uploads and conditional logic, and EAA compliance documentation if you serve EU beneficiaries.

Where nonprofits over-spend

On a CMS too complex for the volunteer team. The right answer is usually fewer custom blocks, not more. The donation flow gets 60% of the technical attention; the rest of the site should be templates the volunteer team can ship without dev support.

I take one nonprofit engagement per year at cost. Email me@pixypuala.com if your timeline overlaps.

No.

Yes — every Sprint Catalogue engagement is fixed-price. Audit Sprint is fixed-price. Build engagements have a fixed-price target with an explicit change-order process for scope additions.

I read your site, the brief, and any existing design files. I send a written scope within two business days. The scope lists every page, every component, every CPT, every form, and every integration. You approve or amend in writing. No verbal scopes.

Anything not in the written scope. I’m generous on small additions (a copy change, a CTA tweak); I bill on anything that requires new template work, new data shapes, or new integrations.

Yes — 50% on signature, 50% on Loom delivery. Build engagements can split into three or four milestones.

Concrete brackets, updated Q2 2026, for UK B2B service businesses with one to three locales and a clear ICP. Numbers below are fixed-bid, not “starting at” theatre.

The brackets

• £7,500 — £12,000 · Marketing site, one locale, custom theme on Gutenberg, four to six template types, conformance to WCAG 2.2 AA on launch. 4 — 6 week build window.
• £12,000 — £18,000 · Same as above plus a second locale, ACF-driven editorial workflow, headless analytics endpoint, and a published runbook for the in-house team.
• £18,000 — £24,000 · Three locales, regulated-copy editor lock-down (role-based block permissions), Salesforce/HubSpot integration, EAA conformance documentation pack.

What changes the price

Editorial parity (how many existing pages, how strict the layout fidelity), accessibility evidence (axe alone is £0; manual NVDA on five locales is £2,000), and integration count. Brand work is not in scope — that’s a separate engagement with a separate vendor.

What doesn’t change the price

The plugin license cost (you pay direct), hosting (you pay direct, I recommend Kinsta or WP Engine), and content production (your team writes).

Read the Sprint Catalogue for fixed-price scope on the rebuild itself, and Care Plan for the post-launch retainer.

£1,200 — £6,000 depending on locale count, plugin spread, and how much manual screen-reader work the buyer wants in the deliverable.

What’s in the £1,200 floor

• axe-core + Lighthouse automated pass on five key pages.
• Manual NVDA pass on the homepage, the primary CTA flow, and one form.
• Written report ranked by user impact, with the WCAG 2.2 success criteria referenced.
• Loom walkthrough of the report so the dev team understands the why behind each finding.

What adds cost

Extra locales (each adds ~£300 of manual SR work — automated tools miss diacritic and direction issues), plugin sprawl (each form plugin is its own a11y surface), and a sample remediation patch (£800 — £1,400 depending on theme complexity).

What I don’t sell

“Accessibility plugins” that promise instant compliance. They don’t work. The audit report tells you which plugins to replace and why; the implementation is a separate engagement.

The Audit Sprint is the fixed-price entry point for this kind of work. Five days, three deliverables, no follow-up sales call.

$12,000 — $48,000. The variance is almost entirely editorial parity — the number of editorial states the existing site supports that the new front-end has to reproduce — not the framework choice.

Cost breakdown for a typical $24,000 migration

• $3,000 · Discovery + editorial inventory. Every page template, every block, every editorial state catalogued. Two weeks. If the buyer skips this, the migration overruns.
• $8,000 · Next.js front-end (or your framework of choice). Type-safe data layer, ISR, image pipeline, structured data.
• $5,000 · Editorial parity. Reproducing every existing component in the new front-end, including the ones nobody documented.
• $4,000 · Hosting + edge config. Vercel/Cloudflare/Netlify setup, custom domains, preview branches.
• $4,000 · Cutover + warranty. URL parity, 301 map, search re-index, a 30-day post-launch fix window.

What makes it more expensive

WooCommerce (add 60 — 90% to the budget; checkout is the hard part), Polylang/WPML with locale-specific block variants, and any editorial workflow that depends on plugin-rendered HTML you can’t reproduce.

If you’re under $40k of total budget, headless is rarely the right answer. The honest essay on when headless wins ↗.

Care Plan starts at $800/month. Audit Sprint is $1,500 single site. Sprint Catalogue is $3,500. Fractional Lead is $4,500/month with a 3-month minimum. Build engagements start at $18,000 (Real-time), $22,000 (Web App), $28,000 (Mobile App).

The first is a template-flip. The second is a custom theme. Specific column-by-column comparison below.

The cost difference compounds. The $2,000 site costs $4,000/year in remediation by year three. The $20,000 site costs $4,800/year on a Care Plan retainer that delivers measured improvements every month. Same five-year TCO, very different posture.

The decision isn’t “which is better” — it’s “which mistake can your team afford to make”. If you’re stretched thin, the $2,000 site postpones the real conversation. The $20,000 site forces you to have it now.

$200 — $800 per month. What “maintenance” should actually include if it’s worth the spend.

What’s not maintenance (but gets sold as it)

• Running plugin updates with no regression test. Worth ~$30/mo if you outsource it; worthless if you have a junior dev.
• Generic “monthly health check” reports that say nothing.
• “Unlimited small changes” with no scope definition. You’ll fight about scope every month.

What you should pay for

Staged updates with automated regression tests, off-site encrypted backups verified monthly, a Lighthouse delta report you can forward to your board, a security scan that names CVEs by package, and a fixed inbox SLA. Two hours of “while you’re in there” changes per month — copy, content, small CSS — capped so the retainer stays focused on stewardship not feature work.

What I charge

$400/mo flat. Six slots, three of which are currently filled. Open the Care Plan terms ↗.

$2,500 — $9,000. What’s in scope, what’s not, and why I bracket fixed-price.

$2,500 floor

A 14-day Performance Rescue Sprint. Target sub-1 s LCP on mobile p75 for the homepage and one inner template. Includes plugin teardown, CSS triage, image pipeline rebuild, edge caching configured, and a public PageSpeed Insights report on delivery. Fixed price, fixed window.

$5,000 — $9,000 band

Same scope but for sites with 40 — 80 plugins, multiple locales, or a page builder that’s tangled into the editorial workflow. The variance is almost entirely plugin replacement work — every plugin you remove has to leave behind theme code that the editor can still use.

Why fixed-price

Hourly billing on perf work is broken. The buyer pays for my learning curve. Fixed-price forces me to bring a methodology, not a stopwatch. If I can’t deliver, I refund and walk; that’s happened twice in three years.

The Sprint Catalogue lists the fixed Performance Rescue terms.

Five-year TCO comparison from real engagements. The honest answer.

The premise

Compare the same site shape — six templates, two locales, a B2B service business — built three ways. Numbers come from twelve real engagements where I either built the replacement or audited the existing.

What the table hides

The $30,600 column is a slow-motion crisis. Three perf rescues, one retrofit, one panic when the page builder ships a breaking change. The $42,000 column is a flat monthly line; everything is in the budget from day one.

When the page builder wins

One-off campaign sites. Internal tools. Sites you’d actually be happy to rebuild in two years. For long-haul B2B marketing sites, custom theme wins on every dimension except day-one cash.

£1,800 — £8,000 depending on plugin sprawl. Where the cost lives.

Cost drivers, ranked

1. Plugin sprawl. Each form plugin, each slider, each pop-up is its own a11y surface. A site with 8 user-facing plugins costs 3× a site with 2.
2. Locale count. Translated content needs separate manual passes for SR pronunciation, direction, and language attribute correctness.
3. Custom blocks. Each custom block needs keyboard testing, focus management, and SR labelling.
4. Brand contrast debt. If the brand palette doesn’t pass AA, you’re remediating brand decisions, not code.

What’s in the £1,800 floor

Audit + sample remediation patch on three high-traffic templates. axe-core CI workflow set up. Sample accessibility statement page. The buyer’s team implements the rest from the report.

What’s in the £8,000 ceiling

Above plus the full remediation, refactoring the brand palette to AA, replacing two to four user-facing plugins with theme code, and full NVDA + VoiceOver + keyboard documentation per template.

The Accessibility Sprint (WCAG / EAA) covers the £4,500 mid-band fixed-price option.

£3,500 — £18,000 in 2026, depending on how much of the EU traffic actually triggers EAA scope. Q1 2026 enforcement examples cited below.

The scope question

EAA applies to products and services offered to EU consumers. A UK B2B SaaS selling exclusively to EU enterprises is in scope. A UK marketing site that also ships brochures to EU prospects is in scope on those pages. A UK marketing site with no EU CTA isn’t in scope at all. The first £500 of any engagement is establishing the scope correctly.

Cost breakdown

• £3,500 · Marketing site, one locale, scope-narrow conformance documentation, axe-core CI, two-page accessibility statement.
• £8,500 · Same plus two additional EU locales, NVDA + VoiceOver pass per locale, content remediation for two templates.
• £18,000 · Above plus product-page remediation, regulatory liaison support, and a one-year monitoring retainer for evidence trail.

What’s already happened in 2026

Three Q1 2026 enforcement cases I tracked: an Irish B2B SaaS fined €11,000, a Dutch retailer fined €34,000, a French insurance broker fined €92,000. Each was scope-resolvable; each had less than six months of warning. Don’t be them.

2 — 12 weeks. The “3 — 6 day” promises are a tell.

The brackets

• 2 — 4 weeks · Rebuild on an existing brand. The brand is locked, the content is frozen, the buyer has decision authority. Three to four templates. WCAG-AA on launch.
• 4 — 8 weeks · Full marketing site rebuild. Five to seven templates, two locales, ACF-driven editorial workflow, real performance budget enforcement.
• 8 — 12 weeks · Above plus regulated copy locks, headless analytics, custom integrations. The variance is integration count and decision speed, not engineering.

What kills the timeline

Marketing decision authority that takes two weeks to confirm a button label. Brand teams that hold copy hostage until the dev work is “done.” Multiple internal approvers without a single decision owner. None of these are engineering problems.

What the “3 — 6 day” sites actually do

They reskin a premium theme. The “build” is theme installation and content paste. That’s a fine deliverable for some buyers. It’s not a custom site, and it’s a different conversation.

7 — 21 days depending on locale count and plugin spread.

The 7-day audit (£1,200)

Five key pages, one locale, automated tools (axe-core + Lighthouse) plus manual NVDA on the homepage and one form. Written report, 30-minute Loom walkthrough, one round of clarifying questions. Five business days end-to-end.

The 14-day audit (£3,500)

Above plus two additional locales, manual SR pass on the primary CTA flow per locale, sample remediation patch on three templates, axe-core CI workflow installed on the buyer’s repo. Two weeks calendar time.

The 21-day audit (£6,000)

Above plus VoiceOver pass per locale, full keyboard navigation test, focus-management review for every interactive component, and a written accessibility statement page draft. Three weeks calendar time.

Why not faster

Automated tools find 30 — 50% of real WCAG issues. The rest needs manual SR work, which is slow on purpose. A 24-hour “audit” is an automated scan with a fresh PDF cover; it will pass any reasonable smell test and miss every issue that gets you fined.

One. Me. I sub-contract design or copy work to named, vetted collaborators when an engagement needs it, with the client’s knowledge.

Yes. Mutual NDAs are standard. I prefer plain-English short-form. I’ll review yours, or send mine.

You can. I bill for work delivered to the cut-off date. I hand over everything (code, docs, Loom walkthrough of in-flight work). No reactivation lock-in.

Yes — every testimonial on the Testimonials page links to the testifier’s LinkedIn. You can DM them directly.

Yes — professional indemnity insurance with worldwide coverage. Certificate available on request.

I invoice from outside the EU. I work remotely with clients worldwide — EU, UK, US, Canada, Australia, and beyond. Contracts are governed by your jurisdiction unless we agree otherwise.

WordPress (Classic themes, Gutenberg-compat). Vite. Tailwind v4. Swiper. Bootstrap Icons + Remixicon. ACF Free, CF7. Next.js + Drizzle for Web Apps. Expo + React Native for Mobile Apps. Liveblocks / WebSockets / CRDT for Real-time.

No. No Elementor, Divi, WPBakery, Beaver, Bricks. The theme owns the design; the database owns the content; the editor edits content, not layout.

I support both. The theme works end-to-end with ACF Free + a native repeater fallback. ACF Pro adds nicer admin UX but is not required.

I don’t host. I will recommend hosts (Kinsta, WP Engine, Rocket.net, Pressable) and set up the WordPress side. The contract is between you and the host.

The theme is translation-ready and ships an RTL stylesheet. For multilingual content I integrate with Polylang or WPML — pick one before scope.

I’ll integrate WooCommerce if the engagement requires a shop, with HPOS compatibility and Block-Editor checkout. I do not retrofit the existing Pixy Puala Classic theme for shop UI in v1.

I do headless cutovers as one of the Sprint Catalogue offerings. WP REST + Next.js + WP-Image-CDN. Editorial stays in WordPress; rendering moves to Next.js.

Yes — for post-content composition. The theme’s design system is not Block Editor-driven; it’s Customizer + template hierarchy + meta. Editors compose body content with core blocks.

Sometimes. For content-heavy sites where editors compose pages, FSE is the right answer. For marketing sites with a designed point of view, Classic is. Decision is per project.

Your GitHub or GitLab. I push to a branch, you review the PR, I merge after approval. I do not keep private copies of client code.

I focus on WordPress. For headless I’ll work with Sanity, Contentful, or Payload if the project requires it. Not Drupal, not Joomla, not Craft.

Slots open at the top of every page (currently 2 / 3, next start 09 Jun). I commit to a start date only after the scope is signed and the first payment clears.

I respond to every email, Slack DM, or Cal.com follow-up within one business day. If I miss it, the next milestone is free.

UTC+1. I overlap with US Eastern in the morning and US Pacific in the early evening. European clients see same-day responses on most days.

One 30-min kickoff. One 15-min weekly check-in. One 30-min Loom walkthrough at delivery. Anything else is async unless we agree otherwise.

A 15–25 min recorded video at the end of every engagement that walks through what shipped, where the code lives, and how to maintain it. Yours forever.

30 days of warranty on the work. Anything in scope that breaks I fix at no charge. After 30 days you can move to the Care Plan retainer or pay hourly for changes.