Design - Effort Model (Internal)

Never client-facing. A living model, calibrated against five shipped MAJOR sites (see Calibration flag 1). Unlike Discovery's per-tier tables, this is a rate ladder: price is computed per project from the confirmed surface count and band.

Assumptions

• Conventions: website-pricing.md § Effort-Model Conventions. Expected mix: 25:75 P:M chapter-wide - System-led 20 · Expressive 25 · Premium 35. Grid prices remain as computed (validated against actuals); at the expected mix the chapter realises ~95% of rate card - the deliberate cost of principal-led design, with Canopy and the instruments as the levers that close it.
• Pricing basis: computed estimate at rate card. Design is priced per project (hours × rates, rounded to a clean number) - the rates carry the margin. Discovery's productised uplift does not apply: that uplift prices fixed SKUs sold before the site is known; Design is estimated from confirmed counts, and the fixed-price risk is carried by the variation mechanism (the surface inventory confirmed against the Cornerstone; growth is priced, never absorbed).
• Inputs come fixed from the Cornerstone: the surface inventory (2.1), behavioural complexity (2.4/2.2), ambition band (sold; confirmed at 2.3).
• Canopy (the Figma starter kit) partially underwrote the calibration sites, so the rates reflect partial-Canopy reality; as it matures, actuals should land below the rates - margin upside, not a pricing prerequisite.
• Excludes cross-cutting Management (PM/AM) and Build-side effort.

The Unit: Designed Surfaces

A surface is anything a designer had to design once: a template, a major variant of a template or section (a hero variant is a surface; a colour swap is not), or a set piece (an interactive tool, an animated brand illustration system) that no named integration covers. Where an integration covers it, its design hours come from the integration register in the Build effort model and it is not counted here as well: search, a store locator and a gated library are integrations, not surfaces. The surface inventory is produced in 2.1 alongside the IA and confirmed in the Cornerstone - variants are where footprints grow (the largest actual in the calibration set was a system-led site whose variant estate roughly tripled its template count), so the inventory names them up front.

Component counts do not price. Components are extracted from surface design; the library's formalisation is the separable design system line below.

The Ladder

Item	System-led	Expressive	Premium
Per designed surface (all-in)	5.5 hrs	7 hrs	10.5 hrs
Design system formalisation (when sold)	+20% of surface hours, minimum 16 hrs - one rule, scaling with size and band together	same rule	same rule
Testing (mandatory at L/XL - never waived; format scoped per project)	baseline: one moderated round at L, two at XL (P 6 · M 12 each) + recruitment at cost; scoped from ~£500 (unmoderated, key journeys) upward	same	same
Rich media direction (video / 3D / illustration commissioning)	-	-	scoped per set piece

Notes:

• Integrations contribute design hours from outside this ladder. The integration register in the Build effort model carries a design figure per integration (0 to 12 hrs). Those hours join this chapter's total before the P:M split but are not multiplied by the per-surface rate, because an integration's design does not scale with the ambition band the way a designed surface does.
• The per-surface rate is all-in: it carries the surface's share of tokens, layout, interaction logic, library extraction, and the two playbacks. The rates of 5.5 / 7 / 10.5 sit on the cleaned project-window actuals: System-led exactly on Tracsis (271.7h / 49 surfaces), Premium exactly on Into Games (~10.5/surface), Expressive between Concurrent's two readings (91.8h project-window; 120h including post-launch additions). The L/XL testing rounds price separately, not cushioned inside the rate. 2.3's concept multipliers (1× / 1.5× / 2×) are their own scale.
• The first surface still costs more than run-rate surfaces, even though the 2.3 concepts carry a good part of it; at real site sizes the per-surface average absorbs it.
• The design system line is separable, and buys the formal increment. The surface rates already carry each project's informal library hygiene (the calibration actuals included it); the line buys the formalisation beyond that - documented tokens, variant matrices, usage documentation, governance handover. It prices at 20% of surface hours (min 16 hrs): roughly £2,000 at S up to £17,000 at XL Premium, converging correctly toward the standalone Design system development product (touchstone §11) at the top - which remains its own engagement for cross-surface, multi-brand, governed systems.
• Inherited, usable client design system: the design system line halves.
• Role split: per the expected mix above (art direction, batch reviews, both playbacks); mid carries production.

Indicative Grid (Computed from the Ladder)

Representative surface counts per size: S 3 · M 10 · L 20 · XL 35 - the single representative set, shared with Build's effort model (which splits the same totals into templates, variants and tools) and both pricing grids. Each is its band's representative count, not its floor. L includes one testing round, XL two. Design system line excluded - add 20% (min 16 hrs) when sold: ~£2,000 (S) · £2,000–£4,000 (M) · £4,500–£9,500 (L) · £8,000–£16,000 (XL). Band boundaries validated against fifteen sized MAJOR sites (see surfaces.md).

Size (surfaces)	System-led	Expressive	Premium
S (≤5)	~18 hrs · £2,000	~22 hrs · £3,000	~33 hrs · £4,500
M (6–12)	~55 hrs · £6,500	~71 hrs · £9,000	~106 hrs · £13,500
L (13–25)	~110 hrs · £13,500	~140 hrs · £17,500	~211 hrs · £27,500
XL (26+)	from ~211 hrs · £25,500	from ~264 hrs · £33,000	from ~386 hrs · £50,000

All + VAT; XL always quoted bespoke from the actual inventory. Real projects price from their confirmed surface count through the ladder. Three cells moved £500 when the grid was recomputed with every P and M line ceiled to the whole hour, as the cost builder does; the ladder itself is unchanged. This grid prices the surface ladder only - named integrations add their own design hours from outside it, which is why the composed journeys in website-pricing.md show a higher design line than these cells.

Calibration Flags (the Important Part)

1. The model is calibrated on five shipped sites, including post-launch design additions:
   • CDSDS - system-led, ~16 surfaces, no design system: 95h actual → 5.9h/surface
   • Tracsis - system-led, large footprint (~49 surfaces once the variant estate is counted): 270h → ~5.5h/surface
   • Concurrent - expressive, ~15 surfaces (9 templates + 5 named hero variants + the search experience): 120h → 8h/surface
   • Source EV - expressive (high), ~20 surfaces (9 templates + calculator 2 + map 2 + 7 bespoke form flows; the animated flow system counts 0 as a system): 160h → 8h/surface
   • Into Games - premium, ~16 surfaces (11 templates + 5 named slice variations): 170h → ~10.6h/surface - the Premium observation, which the 10.5 rate sits on directly. A per-template plus per-component model was tested against the same five sites and rejected: it over-predicted three of the first four by ~2× and under-predicted the fourth, where the surface unit closed all five to within ~10%. This is why component counts do not price. The actuals are all-in design hours - client amend rounds and playbacks included, PM/account excluded - so the rates carry historical amend behaviour, and the capped loops mean projects run under this process should land at or under them. Capture future actuals on the same basis.
2. Counting is the named-variant test, revalidated. Count templates + named Section-level variants + bespoke form flows at every band; distinct interactive tools add surfaces. Illustration and animation systems do not - they are what the band rate carries (Concurrent's eight animated illustrations sat inside its Expressive rate; Into Games' ten choreographed mascots inside its Premium rate). A stricter structural-only variant test was trialled and rejected - it degraded the five-site fit (see surfaces.md § Calibration notes). Record the surface list explicitly on every estimate so the counting discipline calibrates with the rates.
3. The surface inventory is the load-bearing input. 2.1 must name templates, major variants, and set pieces explicitly - variant appetite is where a "10-template" site becomes a 49-surface one, and it is the difference between the grid's M and XL rows.
4. Bands are production modes, not polish levels. The calibration proved variant-rich, motion-polished sites can be system-led; what makes Expressive is brand-led bespoke composition, and what makes Premium is choreographed set pieces and rich media. Judge the band by how the work is produced, not how animated it looks.
5. The design system line is mandatory at L and XL, opt-in at S and M. At scale, an unformalised library is a liability the build and every later phase pays for; below it, the line is the client's explicit choice on the estimate. The 20% itself is research-derived, not observed (no calibration site bought the full formal increment; the rule rests on market benchmarks - starter libraries ~$150–250/component, deep systems 6–12h/element) - calibrate it on the first sale.
6. The principal share (13%) is assumed, not measured - capture the P/M split on the next chapter and refine.
7. The calibration sites were designed wireframe-first; the chapter runs system-first. The five sites the rates sit on all produced wireframes before hi-fi. This chapter produces run-rate surfaces directly as real layouts, so the rate is inherited rather than proved under the order it now describes. It should survive - the rate was always all-in, and the same work happens in a different sequence - but two forces pull opposite ways: dropping the wireframe-to-hi-fi translation takes hours out, while drawing every surface at full fidelity before the walkthrough can put hours back where feedback lands late. Capture the 3.1/3.2 split explicitly on the next project. 3.1 produces every surface and 3.2 only validates, so the ladder apportions roughly 90/10 toward the system - a judgement with no actuals behind it, and one that matters commercially, because the ambition band prices 3.1 while 3.2 scales with size and behavioural complexity instead.