Context: The Sandton E-commerce Team Buying More Traffic Into a Leaking Funnel
A common pattern: a Sandton-based online retailer has a healthy GA4 landing-page session count, a reasonable add-to-cart rate, and then a checkout completion rate stuck in the low single digits on mobile. The marketing team responds the way most teams do - more budget, new creative, a fresh Promotion extension. Nothing moves, because the constraint is not in the media account. It is in the interface.
This is the entry point for a website UX audit in Gauteng. Not a redesign, not a mood board, but a diagnostic exercise that produces one thing: a prioritised, rand-ranked backlog of interface problems, each backed by evidence, with an owner and a way to prove the fix worked. This guide explains what such an audit contains, how the evidence is collected, and what the engagement should cost in ZAR terms.
A UX audit is a measurement instrument, not a design deliverable. If the output is a slide deck full of screenshots with arrows and no funnel data behind it, you have bought a sales document.
What a Website UX Audit Actually Is - and How It Differs From a Heuristic Review or a CRO Programme
Three things get conflated in procurement conversations. They are not the same purchase:
- Heuristic review - an expert walks the site against established usability principles and writes down violations. Cheap, fast, no behavioural data. Useful for catching obvious friction on a low-traffic site where analytics is too thin to be conclusive.
- Analytics-led UX audit - the same expert review, but anchored to GA4 funnel data, session replay and field performance. Issues are ranked by observed frequency and revenue impact, not opinion.
- Full audit plus CRO roadmap - the above, plus a sequenced testing and implementation plan, a measurement plan naming the events that will prove each fix, and a handover path into a continuous optimisation programme.
A CRO programme, by contrast, is ongoing: hypothesis, test, learn, repeat. The audit is the diagnostic that tells you which hypotheses are worth testing first. Prebo Digital runs audits as the front end of its conversion rate optimisation engagements precisely for this reason - an untested backlog is expensive guesswork.
The Five Evidence Layers a Credible Audit Collects
A rigorous audit triangulates. No single layer is sufficient, and a report built on any one of them alone should be treated with suspicion.
| Layer | What it reveals | Limitation alone |
|---|---|---|
| GA4 funnel and segments | Where drop-off concentrates, by device and channel | Says where, never why |
| Session replay and heatmaps | Rage clicks, dead clicks, form abandonment behaviour | Sampling bias, consent gaps |
| Heuristic evaluation | Violations of established usability principles | Expert opinion, not observed behaviour |
| WCAG 2.2 review | Contrast, focus order, labels, touch targets | Legal and inclusive-design lens, not conversion |
| Core Web Vitals field data | Real-user LCP, INP and CLS | No UX context, no intent signal |
Layer 1: GA4 Funnel and Segment Data as the Audit's Spine
Everything else hangs off the funnel. A GA4 funnel exploration should be built for the two or three paths that generate revenue: product view to add-to-cart to begin_checkout to purchase, or service page to form_start to form_submit. Segment by device category, by channel group, and by new versus returning. The output is a ranked list of steps where the loss is largest - that list sets the audit's priorities.
Layer 2: Session Replay, Heatmaps and Form-Field Analytics
Replay answers the question analytics cannot. Watch twenty sessions on the worst-performing step and count realities: repeated clicking on a non-clickable element, field-level drop-off on a form, third-party payment widget failures. Form analytics typically shows one or two fields absorbing most abandonment - often a phone number field validated too strictly, or a postal-code field that rejects valid Gauteng formats.
Layer 3: Heuristic Evaluation Against Nielsen's Ten Usability Heuristics
The ten heuristics remain the most portable framework for structured expert review. In practice the highest-yield checks for South African sites are visibility of system status (loading states, error messages), match between system and real world (terminology, currency, delivery language), error prevention and user control (can a buyer edit a cart item without restarting checkout?).
Layer 4: Accessibility Review Against WCAG 2.2 - Overlap You Cannot Ignore
Many WCAG failures are also conversion failures: unlabelled inputs, insufficient contrast on buttons, keyboard traps in modals, touch targets below the minimum size. Reviewing against WCAG 2.2 is therefore a dual-purpose layer, not a compliance afterthought.
Layer 5: Field Performance and Core Web Vitals From CrUX, Not Just Lab Scores
Lab scores from PageSpeed Insights describe a synthetic device on a synthetic connection. Core Web Vitals field data from the Chrome UX Report describes your actual visitors. On Gauteng e-commerce, the two often disagree by a wide margin, and the field data is what correlates with drop-off.
The Gauteng Reality Check: Mobile Data Cost, Payment Rails and Load-Shedding Resilience
A generic audit template written for a European market will miss the local constraints that matter most here:
- Data-cost sensitivity - an autoplay hero video on the homepage is a real barrier for prepaid-data users. Audit image weight and video autoplay against session duration and bounce on mobile.
- Payment-method mix - South African checkouts must handle card, EFT, instant EFT and buy-now-pay-later flows. Each adds a redirect, a loading state and a failure mode. An audit that only tests the card path has tested a fraction of the funnel.
- Connectivity resilience - load-shedding and unstable mobile connections mean forms that time out mid-submit. Does the form preserve entered data on retry? Does the checkout session survive a dropped connection?
- Language and literacy - for public-facing and financial services, reading level and plain-language labels materially affect completion.
If the audit brief does not mention mobile share of sessions, data weight, or the payment methods actually in use, the auditor is working from a foreign template.
POPIA and Consent: Auditing the Banner That Is Quietly Under-Reporting Your Conversions
South African sites subject to POPIA run consent banners that, when misconfigured, block analytics tags from firing for a meaningful share of visitors. The visible effect: GA4 and Meta both under-report conversions, funnel data is incomplete, and the audit is working from a distorted picture. Worse, platform-reported conversions can simultaneously overstate performance when modelled conversions are included. The audit must include a consent-mode check against the Information Regulator's guidance, plus a comparison of platform-reported versus blended, finance-sourced revenue.
Scoring the Backlog: Severity × Frequency × Revenue Impact
Every finding in the register gets three scores: severity (how badly does it break the task?), frequency (what share of sessions hit it?) and revenue impact (what is the estimated swing in rand if it is fixed?). Multiplying these, or applying an ICE or RICE variant, produces a defensible order. Without scoring, you get a 200-item list and a team paralysed by it.
Quick Wins vs Structural Fixes: Separating Copy and Form States From IA and Checkout Surgery
The backlog should be split explicitly. Quick wins - label copy, error messages, button contrast, field validation rules, loading states - can ship in days and often clear a meaningful share of the loss. Structural fixes - information architecture, checkout flow re-sequencing, platform migration - take weeks or a quarter. Treating them in one list means quick wins wait behind a redesign, which is the most expensive sequencing error in the discipline. Baymard's long-running checkout usability research is a useful reference for the structural category.
Three Audit Engagement Formats and Their Indicative ZAR Ranges
The following ranges are estimates for planning only. They vary with the number of unique templates, the complexity of the funnel, the depth of analytics configuration, and whether the output is a document or a document plus an implementation handover. Treat them as a budgeting conversation starter, not a quote.
| Engagement | Scope | Evidence depth | Timeline | Deliverables | Indicative range (ZAR, est.) |
|---|---|---|---|---|---|
| Heuristic-only review | 5-10 key templates, manual walkthrough | Heuristics plus WCAG spot-checks | 3-5 working days | Issue list, screenshots, prioritised recommendations | ~R15,000-R40,000 |
| Analytics-led audit | Funnel mapping, 10-25 templates, consent and tracking review | GA4 funnels, replay, heatmaps, form analytics, CWV field data | 2-4 weeks | Scored issue register, annotated evidence, measurement plan | ~R45,000-R110,000 |
| Full audit plus CRO roadmap | Whole funnel, checkout, multi-template, post-audit sequencing | All five layers plus stakeholder interviews and test prioritisation | 4-8 weeks | Issue register, roadmap, testing hypotheses, measurement plan, handover | ~R120,000-R300,000+ |
The single biggest cost driver is not site size - it is the number of unique funnel paths, including the non-card payment flows, that must be tested separately.
What the Issue Register Deliverable Should Look Like
Any engagement should end with a single artefact: a spreadsheet or database of findings where every row carries an annotated screenshot or replay link, the evidence layer it came from, severity, frequency estimate, effort estimate, expected impact, a recommended owner (dev, design, copy, media), and a status. If the deliverable is a PDF deck with no register, the team cannot track what was fixed and what remains open. Prebo Digital's custom reporting approach applies the same principle: findings live in a working system, not a presentation.
Defining the Measurement Plan Before Anyone Touches the Site
A fix without a measurement plan is an opinion. Before implementation, the audit should name the exact GA4 events that will prove each fix worked: form_start, form_submit, add_to_cart, each begin_checkout step, plus enhanced measurement signals for rage_click and dead_click where a custom event is needed.
Example: a form-label fix on a B2B quote form
Before: form_start = 1,200 / month, form_submit = 46 / month
After: form_start = 1,200 / month, form_submit = 71 / month
Control: segment by device - desktop should move little, mobile should move most
Proof: the fix worked only if mobile submit rate moved and desktop held steadyThe control logic matters. If everything moves at once, you probably changed traffic mix rather than the interface.
Common Pitfalls: Audits That Are Really Redesign Pitches, and Audits With No Data
Two failure modes recur. The first is the audit generated by a design studio whose revenue depends on rebuilds - every finding conveniently points toward a new site. The second is the audit with no analytics evidence at all: a heuristic walkthrough presented as if it were a behavioural study. Both cost money and delay the real fix. A third, subtler version: a register of 200 items with no prioritisation, which is technically thorough and practically useless.
Ask for the evidence chain. Every finding should point to a data source or a named heuristic, and the auditor should be willing to say what they cannot prove.
Desktop-Only Audits in a Mobile-Dominant Market: The Costliest Omission
South African session share on mobile regularly exceeds desktop for consumer e-commerce, and mobile conversion rates typically trail desktop by a wide margin. An audit that reviews only desktop - or reviews mobile as a shrink-wrapped version of desktop - will miss the majority of the loss. Mobile testing should include throttle conditions, not just a fast connection, and should test the payment-method flows most used by the actual audience.
Handover Paths: CRO Testing Programme, Design Sprint or Rebuild
The audit's conclusions determine the handover. Three paths:
- CRO testing programme - most common when the site is sound but converting below its potential. Findings become hypotheses; wins compound. This is where performance CRO engagements begin.
- Design sprint - when a template or flow needs reworking but the underlying platform is fine. A focus on web design and build scoped from the audit findings, not from a blank brief.
- Rebuild - only when structural issues (platform limits, information architecture, payment infrastructure) make incremental fixes uneconomical. If the audit recommends this, the report should quantify why, in rand.
Published case studies remain the best guide to which path fits a given site profile.
Which Engagement Suits Your Business? Mapping Three Buyer Profiles
Be decisive. Choose by profile, not by budget flexibility:
- A small B2B services site with low traffic and a stalled contact form - the heuristic-only review is sufficient. Analytics is too thin to support an analytics-led audit, and the likely issues (labels, error states, mobile layout) are visible without behavioural data.
- A Shopify store with meaningful ad spend and mobile checkout drop-off - the analytics-led audit is the right purchase. You need replay, funnel and consent-mode evidence before touching the store, and you need the issue register ranked by rand.
- A larger multi-template e-commerce or marketplace operation - the full audit plus CRO roadmap. Without the roadmap, the team will fix the first ten issues and then stall. The roadmap is what makes the programme continuous rather than a one-off project.
A short call is usually enough to determine which of the three applies. The audit itself is the diagnostic; the choice of format is a sizing decision.
