Current & future

Overview

Workflow

Owner

Project overview

Read against the RFP, our conversations to date, and a full parse of both sitemap exports. Thomas is not buying a redesign. It is buying a change of operating model, and the only honest way to price that is to show where a developer sits in the loop today, and where they do not sit tomorrow. Every figure on this page is computed from the source documents, and the places where the brief and the live estate disagree are marked.

Target launch

Feb 2027, intent not constraint

Live estate

3,334 URLs, 1,012 items

Locales

11 live, 8–9 languages

Templates behind it

12–15, not 3,334 builds

Conversion today

≈1% against 3–4%

Vendor decision

Mid-September 2026

01 · Objectives

What this project is actually for

The line in the sand

This is not a redesign

“The objective is therefore not simply to redesign the website. It is to fundamentally change the operating model: marketing-owned rather than agency-owned; structured rather than accumulated; scalable rather than increasingly complex.”

RFP §2, restated in §42

Objective 01

No developer in the core loop

Create a page, pick a template, add a HubSpot form, localise it, make a variant, publish, redirect. Twelve steps in the practical demonstration, and the constraint under all of them is that none of it needs a developer. The cost of the alternative is on record: a menu rebuild through the current chain took roughly three months and shipped broken.

RFP §35

Objective 02

Total cost of ownership, not build cost

The RFP is explicit that low TCO beats low build cost, and asks for three-year and five-year totals. Current spend now itemises cleanly: £65k Drupal agency and Pantheon hosting, £40k SEO agency, £50k translation agency, £20k personalisation and CRO. Exactly £175k.

RFP §32 and §37 · figures confirmed with Thomas

Objective 03

Hold organic, gain answer engines

Traffic is already leaking to AI search, and Thomas is trying to create a category with no established search demand behind it. Answer-engine capability deserves to be judged on evidence rather than on claims, which is why section 02 measures it rather than asserting it.

RFP §12 and §13

Objective 04

Localisation stops being manual

The worst workflow in the current stack. The agency returns documents and eight to nine languages are rebuilt by hand, page by page. The consequence is a growth gap, not just a time cost: there is no SEO effort at all in the secondary-language markets, with France and the Netherlands named specifically, and those are the two largest non-English trees on the site.

Confirmed with Thomas

Objective 05

Conversion, roughly a fifth of where it should be

Conversion fluctuates between 0.1% and 1.3% against an industry target of 3–4%. Closing even part of that gap is worth more than the entire five-year cost saving, which is why this document does not lead with cost.

Confirmed with Thomas

02 · The flow

Seven steps, twice over

Today, in Drupal Target state Where the pain is, and where it is answered In-platform path Reuse rather than rebuild

03 · The estate

Three thousand URLs, twelve to fifteen templates

The number in the brief is wrong, and it matters. RFP §3 asks for “a scalable content architecture capable of supporting 500–1,000+ pages”. The live sitemap carries 3,334 URLs resolving to roughly 1,012 unique content items across eleven locales. Duncan’s own working estimate of “five years worth of resources and 60 core pages” understates the core count by about four times: it is 233 before localisation. Anyone pricing to the brief will be wrong, and anyone pricing to reality risks looking expensive next to those who did not. Say the number first, and price against it openly.

Template family 01

Resources, the bulk of the estate

1,803 URLs · 4 templates

/resources/type/{type}/{slug}

Blog and article item

The single largest content block on the site, and the reason the resource hub is unusable: 1,786 item-level URLs sit under a tree with exactly one category listing page above them.

Volume: 1,052 hr-blog plus 127 blogs, 94 blog and 79 articles. Four taxonomies for one thing.
Rationalisation: Four blog taxonomies collapse to one. Locale-specific type slugs stop existing.
Migration: Highest volume, lowest per-item risk. The right place for AI-assisted transform with sampled human QA.
Watch: 254 URLs untouched since before 2024. Prune rather than migrate, on traffic evidence.

Blott builds4 taxonomies → 1Bulk migration

/resources/type/whitepapers · hr-guides · gidsen · guias

Gated asset item

Whitepapers and guides, and the template that carries the hidden dependency: HubSpot forms are wired to documents in the file library, so a changed document URL breaks the form and any partner or internal link to it.

Volume: 116 whitepapers, 63 hr-whitepapers, 76 hr-guides, plus five localised type slugs
The dependency: The file library is not mentioned anywhere in the RFP. Duncan raised it unprompted and flagged it himself as missing.
Consequence: File-level URL preservation is a priced workstream, not a migration detail.

Blott buildsNot in the RFPHubSpot-coupled

/resources/type/thomas-news · hr-webinars

News and webinar item

Lower volume, but the webinar template carries registration and therefore the conversion path. Six webinar URLs live today against an RFP that lists webinars as a first-class content type.

Volume: 140 news across three locale taxonomies, 6 webinars
Note: Webinar and event registration is a stated requirement in §18 but barely exists on the live site. Build for the ambition, not the archive.

Blott buildsRegistration path

/resources · /resources/{topic}

Resource hub and listing

Effectively absent today. One listing page exists across 1,803 URLs, which means the entire library is discoverable only through search and internal links. The largest single organic opportunity in the rebuild.

Requirement: §15 asks for search and filtering by content type, topic, industry and use case
Relationship chain: Product → Industry → Use Case → Resource → Case Study → CTA, per §11

Blott buildsBiggest organic upside

Template family 02

Proof and people

316 URLs · 2 templates

/case-study/{slug}

Case study

112 unique case studies across 226 URLs. For a business selling into HR and talent, this is the proof layer, and it is the template most likely to sit behind a form and feed the relationship chain.

Volume: 112 unique · en 90, uk 88, nl 27, fr 8, then a thin tail
Design consequence: Needs to carry an economic result prominently, and link to industry, use case and product
Migration: High value per item. Human QA on all of them, not sampled.

Blott builds112 unique

/author/{name}

Author profile

22 unique authors across 90 URLs, a mix of internal people-science staff and external contributors. Cheap to build and disproportionately useful for entity clarity in AI answers.

Volume: 22 unique, including three doctorate-level in-house researchers
AEO angle: Author entities with credentials and schema are one of the few concrete answers to §13’s demand for specifics rather than “AI-ready” claims.

Blott buildsAEO evidence

Template family 03

Product, solution and market

~330 URLs · 5 templates

/solutions/{slug} · /use-case/{slug}

Solution and use case

47 unique solution paths and 26 use cases. These are the pages that have to carry the move from psychometrics into connection intelligence, a category with no established search demand behind it yet.

Volume: 104 solution URLs, 58 use-case URLs
Strategic weight: Category creation means these pages have to build demand, not just capture it. That is a content strategy question, not a template question.

Blott buildsCategory creation

/assessments · /assessment-platform · /connection-platform

Product and platform

Six overlapping product trees today: assessments, assessment-platform, connection-platform, recruitment-platform, employee-development-platform and thomas-cxi. Roughly 170 URLs describing a product set that should have one coherent structure.

Volume: 27 assessments, 33 assessment-platform, 28 connection-platform, plus four smaller trees
Rationalisation: One product template with a controlled hierarchy. This is where the IA work pays for itself.
Not specified anywhere: No requirement is stated for integration with the Thomas product itself: no SSO, trial provisioning or deep linking. Worth confirming.

Blott buildsSix trees → one

/industries/{slug} · /partners/{slug}

Industry and partner

Only 8 industries and 11 partners exist today, which is thin for a business that sells direct in a handful of countries and partner-led everywhere else. The partner layer is the top of most of the funnel.

Volume: 27 industry URLs, 19 partner URLs
Model: Enquiries route through HubSpot to a licensed partner. The site is the top of a distributed funnel, not a direct one.

Blott buildsPartner-led routing

/{locale} · /campaign/{slug} · /pricing · /tryfree

Market home, campaign and commercial

Eleven market homepages, eight campaign pages, plus live pricing, free-trial and payment-methods pages. The campaign template is the one marketing will touch most often, so it is the one the demonstration will be judged on.

Volume: 11 market homes, 12 campaign URLs, 14 legal and policy pages
Unstated: Live pricing, free-trial and payment pages exist, but no e-commerce, checkout or self-serve purchase requirement appears anywhere in the RFP. Confirm before pricing.

Blott buildsDemo template

Shared across every template

The component library

Not a template, and the reason this is not fifteen separate builds. It is also the entire answer to the governance question, and the thing that keeps a marketing team autonomous without letting the IA rot again.

Pieces: Hero and section blocks, cards, carousels, animation-capable modules, CTA and sticky CTA, HubSpot form embeds, breadcrumb and internal-link modules, proof and stat blocks, locale switcher
Why it matters: Build it once and fifteen templates become configuration. Build it per template and design and build roughly double.
Duncan’s ask: “We’d like to have some carousels with our animations and cards. Everything is static in our website.”

Blott buildsThe whole estimate rests here

Localisation

Eleven locales live, and the brief says five to ten

Uneven, and partly contractual

Coverage is deeply uneven, so a flat per-locale price misprices the work

Roughly a third of the estate is fully localised across eleven locales, and roughly a third is English with a UK twin and nothing else. Danish and Simplified Chinese carry under 100 URLs each. Some languages are contractually mandated rather than commercially justified, and Duncan would drop them if he could.

Locale	URLs	Note
en root, no prefix	967	The canonical set. 233 core pages, 622 resources, 90 case studies, 22 authors.
en-gb /uk	924	Near-complete twin of root. Duplicate-content and canonical decisions needed early.
nl	333	Direct-sell market. Has invented its own top-level slugs: oplossingen, bronnen, gidsen.
pt-br	262	Percent-encoded accented slugs live alongside unaccented duplicates.
fr	225	Direct-sell. Five distinct resource-type slugs of its own, including one typo variant.
sv	128	Thin coverage, own top-level vocabulary.
de	114	Thin coverage, own top-level vocabulary.
es-LA	108	Uppercase in the path, a case-sensitivity risk on cutover.
da	96	Under 100 URLs. A candidate for the contractual-only list.
zh-hans	91	Mixed pinyin and percent-encoded Chinese slugs.
en-au /au	86	Thinnest English variant.

Structural problem: Every locale invented its own top-level vocabulary, so there is no consistent cross-locale route to map onto. 464 distinct first-path segments, 191 of which occur exactly once.
hreflang: 539 URLs sit in a full eleven-locale cluster, 982 have only two alternates, and 122 alternates still point at raw /node/ paths. The graph is partly broken before migration even starts.
Open question: Which languages are contractual is not stated anywhere. Do not assume any locale can be dropped.

Blott builds122 broken alternatesContractual list needed

Data quality

What the sitemap gives away

Findings, not assumptions

Defects worth raising in the response

These came out of parsing the two exports Thomas supplied. Naming them is the cheapest possible proof that the audit is real rather than a promise.

Scheme: All 3,334 URLs are declared as http:// in the sitemap output, none as https://
Drupal leakage: 16 live URLs on raw /node/ paths, and 122 hreflang alternates pointing at them
Case and encoding: 207 URLs with uppercase path characters, 94 with percent-encoded non-ASCII slugs
Junk in the index: /test-page, /test, /test-text-component, /error-404-page, plus two Webeo personalisation asset containers sitting in the public sitemap
Duplicates: Live near-duplicate pairs such as team-interaction-optimisation-tio and team-interaction-optimization-tio
No prioritisation signal: 3,323 of 3,334 URLs at priority 0.5 with no changefreq, so migration prioritisation needs GA and Search Console data that has not been supplied

Audit findingAsk for GA and GSC access

Freshness, and what should not be migrated

Migration cost scales with volume, so the fastest way to reduce it is to migrate less. The lastmod distribution suggests a meaningful prune is defensible, but only against traffic data.

By year: 2022: 158 · 2023: 96 · 2024: 1,336 · 2025: 733 · 2026: 1,011
Reading: The 2024 spike is almost certainly a bulk platform-level touch rather than editorial activity, so it is not evidence of freshness
Stale: 254 URLs untouched since before 2024
Proposal: A prune decision taken jointly in discovery, on traffic and rankings, before a single page is moved. Every hundred URLs removed is real money.

Thomas decidesBlott recommendsCost lever

04 · The people

Whose working week this changes

The tension worth naming. The RFP asks for two things that usually pull against each other: a distinctive experience benchmarked against Greenhouse, and a governed component system that stops the page count sprawling again. Most suppliers will quietly trade one away. Our position is that the component library is what makes both possible at once, because the design ambition goes into the components rather than into one-off pages, and the governance then comes for free rather than as a restriction.

Who	What changes for them	What they need from the platform	What we will need from them
Anna BoultwoodVP Marketing	A cost position that can be defended line by line, and a funnel that can be changed without a development ticket.	Transparent recurring costs with the scaling driver named next to each one, and no mandatory retainer.	Confirmation of which languages are contractual, and the session volume behind the conversion rate. Without the second, the growth case stays a ratio rather than a number.
Duncan StimpsonWeb Content Manager	Stops being the whole pipeline. Pages built with a preview, shared as a draft, and localised without being retyped.	A component library with no free-form escape hatch, enforced SEO fields, and a localisation workflow that returns translation into the page.	Analytics and Search Console access early, the file-library structure, and the retain-or-cut decisions. All three change the plan rather than the polish.
Luke WarrenChief Growth Officer	A site that represents the business, and a measurable line from anonymous visit through to revenue.	Attribution that survives the journey into HubSpot, Salesforce and Dreamdata, without a tracking boundary in the middle of it.	Agreement on what success is measured against at launch, so the first quarterly readout is a comparison rather than a baseline.
Denice AngioliniPerformance Marketing Manager	Campaign variants and experiments without a ticket, against a conversion rate that currently sits near 1%.	Native experimentation and personalisation, configured rather than merely available.	The current campaign and conversion reporting, so experiments start from a known baseline.
Becky GriffinContent and SEO	Publishing that is one action rather than an assembly job, and organic effort that finally reaches the non-English markets.	Required SEO fields, controlled taxonomies, and internal linking that does not depend on remembering to do it.	A view on which markets matter most. France and the Netherlands are the largest non-English trees and currently carry no SEO effort at all.
KingaGraphic Designer	Design that ships without a developer, and a system rather than a set of one-off pages.	Components she can extend, and a design system that is the deliverable rather than a by-product.	Time in the component design phase. With no Figma library to inherit, the component library becomes the design artefact that outlives the project.

05 · Fixed

What is already settled

Worth resolving earlyFebruary is an intention, not a constraint

We had understood this as a hard date driven by the hosting contract, but the hosting agreement can be extended, which means the programme currently has no real forcing function. That matters more than it sounds. A date with nothing behind it slips, and a slipping date is how a transformation becomes a redesign. The candidates for a genuine forcing function are the agency contracts ending in July, the exec sponsor’s first full financial year, or a Q1 campaign calendar that needs the new funnel live. One of them should be chosen deliberately, early, and by Thomas rather than by us.

TimetableVendor decision mid-September, build from October

Firmer than the RFP timetable. Platform selection is targeted within two weeks of 24 August, vendor decision by mid-September, build starting October, launch before the end of February. That is roughly five months of delivery, and it puts design sign-off inside the first six weeks. Everything in this document is work already done, so that the September conversation can be about the plan rather than about capability.

Constraint · RFP §35No developer in the core marketing workflow

Tested live, in twelve steps, on a Thomas-specific scenario: new SEO landing page, approved template, approved components, metadata and structured data, internal links, HubSpot CTA, campaign variant, localisation, local CTA, approval and publish, redirect, performance review.

Constraint · 19 AugustSunk agency contracts, no break clause

Twelve-month development, support and hosting contracts were signed for a discount and run to roughly July. Exiting in February knowingly writes off three to four months of fees, and that is already accepted internally. It is also the most credible compelling event available: the saving becomes real on the day the last of those contracts ends, which argues for a July anchor rather than a February one if the date has to move.

Constraint · RFP §38Out of core scope, priced separately

Full brand redesign, full copywriting and content production, ongoing SEO services, ongoing AEO consultancy, and any mandatory development or agency retainer. A retainer can be offered but cannot be a condition.

Constraint · RFP §22 and §32WCAG 2.2 AA, and a five-year cost picture

Accessibility target is fixed. So is the commercial format: a twenty-line one-off cost table, a Year 1 to Year 5 recurring grid, explicit three-year and five-year totals, and every cost that scales with pages, languages, traffic, seats, forms, API calls or AI usage identified.

06 · Open

What has to be answered before this is a quote

Question	What it changes	Who answers
How much of the 3,334 URLs actually migrates?A named workstream, at Thomas’s request	Still the single largest variable in the estimate, but no longer something we have to sell. Thomas has asked for help deciding what to retain against what to cut, and has said thousands of legacy resources carry almost no traffic, some with one view in six months. It is a named workstream rather than a recommendation.	Duncan, on GA and Search Console data we have still not seen
How big is the file library, and what are its URL patterns?URL preservation is a hard requirement	No longer a risk we spotted; a requirement Thomas has stated. Whitepapers, on-demand documents and HubSpot forms all have to keep their URLs through migration. The volume still decides whether this is four days or fifteen, and it cannot be sized from outside.	Duncan Stimpson
Which languages are contractually mandated?	Whether the thin locales can be retired. Danish, Simplified Chinese and Australian English carry under 100 URLs each and cost the same to set up as the ones that matter.	Thomas legal and commercial
How much migration volume can Thomas absorb itself?A hybrid model, with the split set by available hours	Duncan has migrated Drupal, WordPress and Wix before and has already rebuilt pages in a target platform, so the question is throughput rather than expertise. How many hours are genuinely available sets the split between Thomas-led and supplier-led lines.	Duncan Stimpson
Is there any requirement to integrate with the Thomas product itself?	Whether SSO, trial provisioning or deep linking into the assessment platform is in scope. Not mentioned in any document, but live pricing, free-trial and payment-methods pages exist.	Duncan
How much traffic sits behind that conversion rate?Conversion is 0.1–1.3%, against a 3–4% target	Whether the conversion gap can be monetised. Knowing the rate without the volume gives us a ratio and not a number, and the exec sponsor will want the number.	Denice Angiolini
Is the enterprise SEO suite an existing cost or a proposed one?	If Thomas already pays for a dedicated enterprise SEO platform, choosing a CMS that includes answer-engine tooling is an avoided cost worth naming in the total. If not, it is not a saving and we will not present it as one.	Becky Griffin
Is Webeo being replaced, and when?	Whether CRO Level 3 personalisation is in the February scope or a phase two. Roughly £30–40k a year sits here, and displacing it has a relationship cost internally.	Anna
Who owns the component library after handover?No Figma library and no product designer today	Design goes straight into the component library rather than through a Figma handoff. That is settled. What is not settled is who extends it in year two, which decides whether autonomy holds or decays.	Anna Boultwood and Kinga
What is the compelling event, now that hosting can be extended?	Whether February holds at all. Without a forcing function the date drifts, and a drifting date is how the operating-model change quietly reverts to a redesign. Our recommendation is the July agency contract expiry.	Anna Boultwood and Luke Warren
Is closing the France and Netherlands SEO gap in scope?	There is no SEO effort in the secondary-language markets today, and those two are the largest non-English trees on the site. It is the clearest growth argument in the pack, and it is currently nobody’s line item.	Anna Boultwood
Will a prototype exist, and when?	§6 describes a platform-neutral prototype at length and lists ten supplier obligations against it. None exists today, so the obligations have no object and could arrive mid-project.	Anna

Current & future

We have not yet had analytics or Search Console access, so nothing here is dressed up as more certain than it is. Every chart carries its provenance. The measured ones come from parsing the sitemap exports supplied with the RFP and from reading the live site and live search results. Figures marked stated are Thomas’s own numbers rather than ours. The modelled ones have editable inputs, so anything we have read wrongly can be corrected on the spot and the chart redraws.

Measured

Computed from their sitemap exports, or read from live pages and live search results. Sample sizes are stated on the chart.

Stated

A figure Thomas or the RFP gave us. Where it is a reconstruction rather than a disclosure, the chart says so.

Modelled

Our assumption, with the inputs exposed. Change them and the chart follows. Nothing here is presented as a Thomas figure.

Assessed

Our judgement, scored transparently against criteria Thomas published, so the working is visible rather than asserted.

01 · The estate

What the sitemap actually contains

The brief is sized against the wrong number

Measured

RFP §3 asks for an architecture supporting “500–1,000+ pages”. The live estate is 3.3 times the upper bound of that range.

Unique items collapse hreflang clusters to one row per piece of content. Duncan’s own working estimate of 60 core pages is 233 in the data.

Seventeen resource taxonomies doing the work of five

Measured

1,787 item URLs spread across seventeen type values, four of which are blogs and four of which are whitepapers. The right-hand column is the target it collapses into.

Duplication is largely per-locale invention: gidsen, guias, guides and hr-guides are one template. One taxonomy, le-guide-complet-pour-manager-ses-talents, is a single article mis-filed as a content type.

Locale coverage is uneven enough to break flat pricing

Measured

Eleven locales live, three of them English variants. The top two carry 57% of the estate between them; the bottom three carry under 100 URLs each.

The RFP asks for five to ten languages. Eleven exist. Which of them are contractually mandated is not stated in any document, so none can be assumed droppable.

A third of the estate has not been touched since 2024

Measured

By last-modified year. The 2024 spike is almost certainly a bulk platform touch rather than editorial activity, so it is not evidence of freshness.

254 URLs predate 2024 entirely. Migration cost scales with volume, so this is the chart that turns a prune from a cut into a recommendation, once traffic data confirms it.

01b · Defects

Found in the supplied data and on the live site

02 · Discoverability

Whether anyone can find them

Thomas is invisible on the queries that define the category it is creating

Measured

Sixteen queries, four intent bands. Thomas appears where its own product names are the query, and disappears the moment the query is about a buyer’s problem. That is a product-page-only footprint, and it is the wrong shape for category creation.

Queries returning thomas.co

Caveat, stated plainly: the search tool available to us is US-only and returns titles and URLs with no ranking metadata, no AI Overview and no snippet data. Treat this as a directional US baseline for a UK-headquartered business, not a ranking audit. It cannot tell us whether these queries trigger AI answer boxes, and we should not claim otherwise. A proper answer-engine baseline needs direct prompting of ChatGPT, Perplexity, Gemini and Claude with citation capture, plus a UK-geo tool.

The category is already occupied, and one competitor owns the name

Measured

What comes back instead of Thomas, on the five category queries.

The hardest findingEdelman already ranks first for “employee connection index”

With an established, named asset: the Edelman Employee Engagement Connections Index. Same construct, near-identical name, incumbent authority. If Thomas intends to launch a Connection Index, it is walking onto held ground.

Name collisionTwo other claimants on the phrase itself

Connective Intelligence Inc. sells assessments under that banner, and an independent consultant ranks first for “connection intelligence assessment tool”. Separately, “connectional intelligence” is an established concept with its own body of content and more authority.

Semantic driftThe term collides with enterprise networking vocabulary

Cisco, Google Cloud Network Intelligence Center, Esri and a company literally called Connection occupy the generic queries. The phrase is not clean, which raises the cost of owning it.

Who holds the practical queriesThe engagement software category, wholesale

Vantage Circle, Workhuman, Simpplr, Paylocity, Achievers, Quantum Workplace, WorkTango, CultureMonkey, Gallup and ADP answer “how to measure employee connection” and its variants. These are content-marketing operations, and they are beatable, but only with a content architecture Thomas does not currently have.

Ceded groundThomas is absent from its own reputation and pricing queries

Capterra, Trustpilot, Glassdoor, AlternativeTo and test-prep sites own “Thomas International assessments review”. SourceForge, G2, GetApp and SoftwareSuggest answer “Thomas International pricing” without Thomas present. On the Hogan comparison, the narrative is written by four competitors positioning themselves as the alternative.

Authority inversionA third party outranks Thomas on Thomas’s own instrument

The British Psychological Society test review outranks Thomas’s own HPTI page. An LLM-facing encyclopaedia has also generated an HPTI entity page without Thomas as the source, which is precisely how a brand loses control of its own definitions in AI answers.

Domains counted across all sixteen queries. Genuine assessment competitors surfaced far less often than aggregators and review marketplaces, which matters strategically: intermediaries, not rivals, currently control how buyers encounter Thomas.

The content is on the page and invisible to machines

Measured

Breadcrumbs, FAQs, authors and dates all exist visually. None of them carry the markup that lets an answer engine use them. That gap is a configuration job, not a rebuild.

Rendered for people Marked up for machines

Sample and confidence: 16 URLs across every archetype. Head metadata read successfully on 17 of 17 fetches, so those counts are firm. Structured data could only be confirmed on 5 pages, because the fetch tool strips script tags and the raw-HTML route was rate limited. The absences shown are explicit on those five, and the remaining eleven are unverified rather than assumed empty. Worth re-running with proper crawl access before this goes in a proposal.

Head metadata: solid where it is automatic, sloppy where it is authored

Measured

Out of sixteen sampled pages. Canonicals are perfect, which is worth saying, because it means the plumbing works and the problem is governance.

Four non-standard og:type values are in use, none of them valid Open Graph types: “Thomas content pages”, “Resources”, “Case study” and “Resource Author”. The homepage also ships unreplaced template tokens, with og:image:width carrying its own property name as its value.

03 · The workflow

What the operating model is worth in hours

Hours per task, today against the target

Modelled

Editable. Current-state figures are our reading of “I have to build it one by one by one” and “a very time intensive thing to do”.

Today, in Drupal Target, marketing-owned

Localising one page set means the same page across the remaining ten locales, which today is retyped from a spreadsheet with one HTML tag per sentence. It is the largest single saving in the model and the one Duncan described most vividly.

Annual hours returned to the team

Modelled

Hours saved per task, multiplied by how often each one happens in a year.

Frequencies are our assumption: 120 blog posts, 24 landing pages, 12 localised page sets, 12 CTA tests and six new templates a year. This is capacity returned to a small team, not a case for a smaller one: the same people get to do work that is currently crowded out.

Capacity released, per year

—

Equivalent to — of one person, at 1,800 productive hours a year. Uncosted in the current run rate, because it is absorbed as manual rebuild.

The step that disappears

—

Of the twelve steps in the RFP’s own practical demonstration, this many currently involve a developer, an agency ticket or a blocked path. The target is zero, and that is the number §35 is really testing.

04 · Performance and money

What it earns, what it costs, and what it costs after

Conversion is running at roughly a fifth of the industry target

Stated

The most important number in this document. It moves the case from cost avoidance to growth, and it is the reason the cost comparison is not the headline.

Observed range today Industry target

Read carefully in both directions. At the midpoints it is a fivefold gap; on the most favourable reading, the top of the observed range against the bottom of the target, it is still 2.3 times. What we cannot do yet is turn either into money, because we have the rate and not the session volume behind it. One number from Thomas closes that gap.

Conversion today, midpoint of the observed range

0.7%

Fluctuating between 0.1% and 1.3% depending on what is running. Campaign spend currently buys traffic into a funnel that converts at a fraction of what it should.

Where it should be

3–4%

The industry target Thomas is measuring itself against. Closing even half the gap is worth more than the entire five-year cost saving, which is why cost is the floor under this case rather than the case itself.

04b · Cost

What the current model costs

Where £175k a year currently goes

Stated

Four supplier lines, given on 24 August. These reconcile to £175k exactly.

Two things worth drawing out. Translation at £50k is the second largest line, which moves localisation from a workflow complaint to a cost centre and is the strongest single argument for fixing that workflow first. And hosting sits inside the £65k agency line on Pantheon rather than being separately visible, which is exactly the kind of bundling that makes total cost of ownership hard to see. Internal time is uncosted on top of all four lines.

Recurring cost, five years out

Modelled

The new stack against today’s run rate, both escalated at the same rate so the comparison is fair.

Today’s run rate New stack

Driven live by the recurring grid and the escalation slider on the cost of ownership view. Excludes the optional retainer, so the platform comparison stays clean. Every underlying figure is indicative and needs verifying.

Cumulative position, including the one-off build

Modelled

The build is paid once; the saving recurs. This chart is where the two meet, and it is the one the CGO will repeat secondhand.

Deliberately conservative. It counts recurring savings only. No conversion improvement, no organic recovery, no value on the capacity released above, and no credit for the three to four months of agency fees already being written off. Any of those pulls the crossover earlier, which is the right direction for a number to be wrong in. And it does not always break even. On the revised numbers only the lean scope crosses zero inside five years, at roughly month 48. At the recommended and full scopes the recurring saving alone never repays the investment within the window, and the marker is simply absent rather than extrapolated. That is the honest read, and it is the reason the conversion gap in the charts above is the argument and the cost saving is only the floor.

05 · The platform decision

Scored on Thomas’s own criteria, at Thomas’s own weights

The seven things we believe this decision turns on

Assessed

Our reading of priority, drawn from the RFP and from every conversation so far. It is deliberately shorter than the twelve weighted criteria, because a document that scores everything does not say what matters most. If we have the order wrong, that is worth ten minutes of the next conversation.

#	Criterion	What it means concretely	What it demands of the response
1	Self-sufficiency	Remove reliance on developers and agencies altogether. The benchmark is a menu rebuild that took three months and shipped broken.	The twelve-step workflow demonstrated rather than described, run by Thomas rather than by us.
2	AI-search readiness	Judged on evidence rather than on platform claims, which is what §13 explicitly asks for.	Specifics, per §13. The schema audit and answer-engine baseline in section 02 are ours; we would expect the same standard of evidence from anyone answering this brief.
3	Content and design flexibility	Build a new component, a carousel for instance, without a two to three week development cycle.	The component library, shown being extended rather than shown finished.
4	Speed of deployment	AI-native publishing, so the team ships faster while governance holds. Our recommendation rather than a stated requirement.	A named and priced workstream. It is not in the RFP, and we think it should be.
5	Workable localisation	Multi-language publish, remote contributor publishing, and glossary support so terminology stays consistent.	The glossary and the contributor model. Neither is in the RFP text, and both are priced here.
6	HubSpot integration	Forms, CTAs and the routing that sends partner-market enquiries to a licensed partner.	Table stakes. Losing here is possible; winning here is not.
7	Personalisation	Replacing the outsourced tool that is currently fed by hand from HubSpot.	Native personalisation configured, not merely available, replacing a tool currently fed by hand.

Two of these, AI-native publishing and glossary-backed localisation, appear nowhere in the RFP. We have priced them anyway, because a brief written before the work starts cannot be expected to anticipate everything, and both will matter within a year of launch.

Why this is scored rather than asserted. The RFP is platform-agnostic, names five candidates, and asks suppliers to explain why they would not choose the alternatives. Scoring the candidates against the criteria Thomas published, at the weights Thomas set, makes the same argument without asking anyone to take our word for it. The scores below are our assessment and are open to challenge line by line, which is the intended effect. We have gone beyond the §28 list. Our understanding is that the primary list is Webflow, Sanity, WordPress, Framer and Storyblok, and those are grouped first below. The wider list follows for reference. If that understanding is wrong, or if there are candidates missing, tell us and we will rescore against the right set on the same basis.

Weighted score across all twelve criteria

Assessed

Each platform scored one to five per criterion, multiplied by the RFP’s published weight. A perfect score would be 100%. The five on the primary list are grouped first, because that is the comparison the decision actually turns on; the wider list follows for reference.

The spread matters more than the winner. Webflow leads on marketing autonomy, design ceiling and experimentation, which together carry 35% of the score. It does not lead on cost, and on migration it is level with the best of the field rather than ahead of it. The matrix below shows exactly where it does lose. Framer is the one worth a second look, and not a favourable one. It scores well on design and on speed of setup, and poorly on the things that only hurt later: multi-level content relationships, multi-stage approval, and a content graph at this scale. At 3,334 URLs across eleven locales those are year-three problems, and they are cheaper to avoid now than to fix then.

Score by criterion, so the working is visible

Assessed

Sorted by the weight Thomas assigned. Bar length is the score out of five.

The twelve demonstration steps, and who is needed for each

Assessed

The RFP’s own practical demonstration, step by step, in the current stack against the target. The constraint under all twelve is that none of them should need a developer.

Current-state assessments come from what Duncan described: no WYSIWYG or preview, no template library, translation rebuilt by hand, no approval workflow, redirects handled by the agency, and tracking that breaks at the Lovable iframe boundary. Each is worth confirming with him rather than asserting.

06 · Honesty

What we could not measure, and why

Not measuredCore Web Vitals and page speed

Duncan describes performance as “really poor”, and §21 asks for measurable targets, but we have no measured baseline. Network egress from our environment is restricted, so we could not run Lighthouse or pull CrUX field data. A PageSpeed Insights run across a sample of templates and locales would settle it, and we would rather quote a performance target against a measured baseline than against an adjective.

Partly answeredConversion known, traffic still not

Conversion is known, at 0.1–1.3% against a 3–4% target. What is missing is session volume, so the gap can be stated as a ratio but not as revenue, and the growth case stays weaker than it should be. Also missing: Search Console and ranking data, without which migration cannot be prioritised by value. Read access in the first discovery session unblocks both.

Not verifiedStructured data on eleven of sixteen pages

Confirmed on five. The remaining eleven are unverified rather than empty, and we have not rounded the finding up to the full sample. A crawl with proper access would settle it in an afternoon.

Now a requirementThe file library, still unquantified

A hard requirement: whitepapers, on-demand documents and HubSpot forms must all keep their URLs through migration. The volume is still unknown and still drives one of the more volatile lines in the estimate. It cannot be sized from outside.

Not observableWhether AI answer engines cite Thomas

Our search tool returns titles and URLs only, with no AI Overview, snippet or citation data, and it is US-only. The visibility chart is a directional organic baseline. A real answer-engine audit means prompting the assistants directly and capturing citations, in the right geography.

Not disclosedWhich languages are contractual

Some locales are understood to be contractually required, but not which. Three carry under 100 URLs each and cost the same to set up as the ones that matter. Until this is settled, localisation pricing has to assume all eleven survive.

Estimates

RFP §32 asks for a Year 1 to Year 5 recurring grid with three-year and five-year totals, and for every cost that scales with pages, languages, traffic, seats, forms or usage to be named next to the line it sits on. This is that model, made editable so Thomas can test it rather than take it. The itemised one-off cost table §31 requires sits in the written proposal, where it can carry the working alongside each figure.

Every recurring figure below is indicative and needs verifying. Platform pricing is list pricing and subject to negotiation. Several of these lines scale with traffic, seat counts and translation volume, none of which we have been given. We would rather show the model with its assumptions exposed and be corrected than present a single number and be believed.

Programme investment, one-off Optional retainer, per month Annual escalation 4% Thomas today, per year

01 · Recurring

The Year 1 to Year 5 grid

Recurring line	Year 1	Year 2	Year 3	Year 4	Year 5
Platform bundle: enterprise tier, ten locales, answer-engine tooling and experimentationIndicative and still being scoped with the platform vendor. Scales with CMS items, seats, locales and forms.		—	—	—	—
Hosting, CDN and environmentsIncluded in the bundle on a managed model. Today this sits inside the £65k agency line rather than being separately visible.		—	—	—	—
Translation£50k today, the second largest line in the current model. Modelled at £35k once a glossary and in-page translation remove the manual rebuild. That reduction is a hypothesis, not a commitment.		—	—	—	—
Internal searchScales with indexed items and query volume		—	—	—	—
Security, consent and monitoringConsent management, WAF, uptime and vulnerability monitoring, penetration testing cycle		—	—	—	—
Analytics and attribution toolingWebsite-side share only. Existing Salesforce and Dreamdata licences sit outside this scope.		—	—	—	—
Optional optimisation retainer§38 rules out any mandatory retainer, so this is offered and not assumed. Off by default.	—	—	—	—	—
Total recurring, per year	—	—	—	—	—
Thomas today, all in, escalated at the same rate	—	—	—	—	—

02 · The comparison

Five years, new against current

Five-year recurring, new stack

—

Three-year total —. Both figures are the ones §32 asks for explicitly, and both are computed from the grid above rather than quoted separately.

Five years at today’s run rate

—

Current spend as confirmed with Thomas, escalated at the same rate. Composed of:

Drupal agency and hosting, £65k
Translation agency, £50k
SEO agency, £40k
Personalisation and CRO, £20k
Plus internal time absorbed as manual rebuild, uncosted

Please correct any of these. The comparison is only as good as the baseline.

The comparisonRecurring saving over five years

— against the current run rate, before any conversion improvement and before any value is placed on the capacity returned to the team. RFP §37 sets low total cost of ownership above low build cost, and this is the line that test is decided on.

Said plainlyThe cost saving alone does not justify this

At the fuller scopes, recurring savings do not repay the one-off investment inside five years. We would rather say that than present a payback curve that only works by leaving things out. The case for this programme is the conversion gap in section 04, not the supplier consolidation, and cost is the floor under the decision rather than the reason for it.

What is not in hereFour things, deliberately

No conversion uplift, no organic recovery, no credit for the internal hours released, and no value on retiring the workarounds. Each of them is real and none of them is provable in advance, so none of them is counted.

03 · Assumptions

What this depends on

Twelve to fifteen templates, not a thousand pages

The model assumes the taxonomy rationalisation in section 03 is accepted. Meaningfully bespoke layouts per market or per product line change the build materially, and we would rather say so now than discover it in design.

The estate is pruned before it moves

Priced against 3,334 URLs as they stand. Deciding, on traffic evidence, what should not move at all is the single largest cost lever available and it belongs to Thomas. It needs analytics and Search Console access in the first discovery session.

Thomas supplies content, we supply structure

Migration moves existing copy rather than rewriting it. Rewriting core pages and producing new content are priced separately, in line with §38.

The file library can be preserved

HubSpot forms and partner links depend on those document URLs. We have assumed a mappable structure of a few hundred documents. It cannot be sized from outside, and it is the assumption most likely to move.

Integration access from day one

HubSpot, Salesforce, Dreamdata, GA4 and Search Console credentials available at build time. Waiting on any of these is the likeliest source of slippage against February.

Ten locales at launch

Eight to nine languages across eleven live locales. If none can be retired for contractual reasons, both the setup effort and the licence scale with the eleventh.

February needs a forcing function

We understand the hosting agreement can be extended, so the date currently has nothing behind it. We have planned to a five-month window regardless, but a programme this size benefits from a real deadline and it is worth choosing one deliberately.

Design lands in the component library

With no Figma library to inherit, the component library becomes the design artefact that outlives the project. That simplifies handover and it also means the library, not a set of page designs, is what we are really delivering.

Appointment in autumn, launch in February

Roughly five months. Design sign-off needs to land inside the first six weeks so that migration and build run in parallel rather than in sequence. Slipping the appointment does not slip the hosting expiry.

Platform commercials sit alongside, not inside

Agency services and platform licensing are quoted separately, so nothing in the programme investment double-counts a licence and the platform decision stays reversible on its own terms.