OMNES is shipping 648 reviews of refund silence.

Why this brand

OMNES is a UK sustainable womenswear brand selling at the affordable end of the contemporary occasion-wear bracket, with physical stores in multiple UK cities including London and Glasgow. The DTC storefront is on Shopify. Grips Intelligence reports an annual GMV in the region of €6.7M, and the team sits in roughly the 12 to 35 employee range based on LinkedIn signal. They are a small operator with a real product, a clean storefront, and a recognisable brand voice in their category.

They were picked for three reasons. First, they are a representative SMB Shopify brand in a non-US market: the audience for this site is mainly UK and EU founders, and the public-surface signal for OMNES is rich without being noisy. Second, the gap between their product reputation (which is largely positive) and their support reputation (which is currently 2.9 stars on Trustpilot across 648 reviews) is exactly the shape we are built to address. Third, the visible third-party stack on the storefront is minimal, which makes the inference layer cleaner and the recommendations more concrete. There are fewer competing explanations for what we see.

Two things this audit is not. It is not a roast. OMNES ships a real product, communicates their sustainability sourcing in detail, and has a polite, well-structured FAQ. The 1-star wall on Trustpilot is real, but so is the rest of the picture. It is also not a pitch for them specifically. If they read this, the honest message is the one in the section called What they could do without us: most of the gain available here does not need an outside vendor.

Methodology and what is missing

The disclosure above the fold names the headline constraint: no ticket data. This section names the rest.

The deflection band is not a projection for OMNES. It is the band that a brand on this stack, in this vertical, with this current automation surface, typically lands on when a Tier 0 engagement is run well. The lower edge of the band is the realistic floor, the upper edge is the realistic ceiling, and the actual outcome depends on the implementation, on continued FAQ investment, and on the brand's own appetite for changing the refund-cycle policy that is doing the most damage. A real Tier 0 would access 200 to 500 pseudonymised tickets under a data-processing addendum and would replace the inference layer with a measured ticket-mix. None of that has happened here.

Every screenshot on this page was captured fresh via playwright at 1440x900 on 27 May 2026. Cookie banners and the 15% off newsletter modal were dismissed in-browser before capture so the storefront's actual content is what you see, not the overlay layer.

The 90-day deflection target band

OMNES start from a public-surface deflection rate that is effectively zero: there is no helpdesk widget, no AI front door, no on-site tracker, and no FAQ search visible on the storefront. Whatever handling pattern sits behind the public mailto and contact form, there is no AI layer doing the work in front of it. The chart below is the band a brand on this stack typically lands on after a well-run engagement, given the category-share inferred above. It is not a projection of what OMNES will achieve.

The band sits below what a Gorgias native-AI case study would publish (their public range is 26 to 56 percent), and below the Help Scout AI Answers ceiling on FAQ-grounded deflection (their public claim is around 73 percent on FAQ-only intents). The reason is OMNES-specific: the dominant complaint category is refund delay, and AI does not deflect a refund delay until the underlying refund cycle is shortened or the expectation is set honestly upfront. That is a policy and operations change, not a deflection tool. If the refund cycle is fixed first and the AI sets honest expectations second, the upper edge of the band is reachable. If only the AI ships, the lower edge is more realistic.

What they could do without us

This is the section that matters. A useful audit is honest about which gains do not require an outside vendor. For OMNES, most of the immediate CX-experience gain is achievable in two to four weeks by the existing team plus one helpdesk install. None of the items below require AI.

1. Install a helpdesk. Any helpdesk.

There is no published chat, no Beacon, no Gorgias, no Intercom in the storefront source, on the public-surface evidence. Whether OMNES already run a back-of-house helpdesk that simply does not surface a widget, or whether they handle support from a shared inbox, is not visible from outside. Either way, a public-facing helpdesk plus a knowledge-base front door is the precondition for the rest of the work below. The two natural fits at this size are Help Scout (clean UX, strong knowledge-base, AI Answers built in) and Re:amaze (cheaper, Shopify-native, weaker AI tooling). Both ship a public chat widget, a tag-based queue, canned replies, and macros, and both produce a measurable drop in per-ticket handling time once macros are loaded.

2. Add a Klaviyo refund-status flow on the returnsportal.co webhook

returnsportal.co has webhook events for return received, return inspected, refund issued. Klaviyo can consume those. A three-stage email at day 0, day 5, day 10 names the cycle, sets the date, and gives the customer the path to chase only if the cycle is breached. This is the single highest-leverage change available on the current stack, and it is two to three days of work for one Klaviyo-literate person.

3. Add an order-tracking page on the storefront

Tested with playwright: /pages/order-tracking, /pages/track, and /pages/track-order all 404. Customers route to email for "where is my parcel". AfterShip, Wonderment, or even Shop App's own tracker block this category at the door. Half a day to install, two days to theme. Removes a steady share of inbound for as long as the tracker remains in place.

4. Publish a one-page Refunds policy with a stated cycle and a chase path

The 10-working-day post-warehouse cycle is already inside the Returns & Exchanges page, but it is buried in paragraph three of the General Information section. Pull it out, name it on the FAQ, and add a clear "if more than 14 working days have passed and you have not heard, email hello@omnes.com with your order number" path. This converts a complaint into a policy.

5. Add an SLA line on the Contact page

The Contact page states "Mon to Fri, 9am to 5pm" hours, but it does not commit to a reply SLA. Adding "we reply within 1 business day" (or whatever the real number is, even if it is two days) means a customer who emails on a Friday afternoon knows what to expect on Monday. The hypothesis worth testing here is that a meaningful share of the "ghosted" 1-star reviews would not have been written if the expectation had been set explicitly upfront. Public reviews show non-response; they do not prove the cause. Setting the SLA is cheap, and the test is whether the 1-star wall thins after it ships.

What an engagement would add on top

Everything in the section above is gettable without us. What an Eltrus engagement adds on top is the part that the brand cannot ship in a sprint. The integration layer, the eval discipline, and the brand-voice tuning that turns an AI front door from "a Help Scout AI Answers box" into something that actually sounds like OMNES.

• Phase B ticket diagnostic. 200 to 500 pseudonymised tickets over a 30-day retention window. Replaces the public-surface inference layer above with measured intent shares, measured refund-cycle median, measured first-response time, measured deflection-eligibility per category. The shape of the recommendations does not change, but the numbers do.
• The AI front door, brand-voiced. A retrieval-augmented agent grounded on the FAQs, the Delivery page, the Returns & Exchanges page, and the Refunds policy we ship in week one. Tuned on OMNES's own existing reply templates so it sounds like the team, not like a generic bot.
• The integration layer. Shopify order context into the agent. returnsportal.co status into the agent. Klaviyo flow triggers from agent events. The thin glue that makes "where is my refund" answerable in 30 seconds with the actual answer, not a templated apology.
• The eval set. A held-out set of 100 to 200 real customer messages, scored weekly on resolution, brand-voice fidelity, refund-leakage, and escalation rate. This is what stops the deflection rate from drifting backwards in month three when nobody is watching.
• The 90-day rollout discipline. Weekly checkpoints, a rollback plan per flow, an explicit handover doc at day 90 so the team can run it without us.

Impact range, with caveats

Ranges, not point estimates. The numbers below combine the Grips-reported GMV with industry norms for apparel ticket ratios, cost per ticket, and repeat-purchase economics in the UK SMB band. None of the CX-economics figures come from OMNES; they are typical-brand bands and they exist to bound the exercise. If any of the assumptions are wrong, the numbers move.

Volume side

At ~€6.7M GMV and an apparel AOV in the £80 to £120 range (typical for this category at this price point), OMNES are likely processing somewhere in the order of 4,500 to 7,000 orders per month. Apparel ticket-to-order ratios in this sub-category typically run 8 to 18 percent depending on returns volume and shipping complexity. That puts inferred monthly ticket volume in a 400 to 1,250 range. Wide because we do not have the data; tighter as a Phase B exercise would make it.

Cost side

At an internal cost-per-ticket of £4 to £9 (small team, no helpdesk yet, mostly time-cost) and a steady-state deflection band of 30 to 50 percent on the right stack, the annual saved-handling-cost range is roughly £6,000 to £67,500. This is the conservative cost-side number and excludes the helpdesk-install productivity win, which is separate.

Revenue side

The larger number lives on the revenue side, and it is harder to bound. Two channels.

First, preserved revenue from customers who currently return-from-frustration without buying again. UK apparel repeat-purchase rate at this band typically sits around 25 to 40 percent across published benchmarks. If a refund-status flow plus an AI front door recovers half the customers who would have given up after a poor refund experience, on an assumed base of 600 to 1,500 such customers per year, at a per-customer 12-month repeat value of £80 to £160, the preserved-revenue range is roughly £24,000 to £120,000 per year. All four inputs here are estimates; the band is wide because the chain of inference is long.

Second, recovered revenue from would-be returners who get a fit or sizing question answered pre-purchase and either buy with confidence or buy a different size. This one is harder to size; the conservative read is £15,000 to £60,000 per year at this revenue band.

Adding the cost and revenue sides, the defensible 12-month impact band lands somewhere in the £45,000 to £240,000 range. Treat the top of that as the optimistic case, the bottom as the floor, and the middle as a useful planning number. A Phase B audit would tighten both ends of the band materially.

Risks and constraints in this stack

The refund-cycle bottleneck is upstream of CX tooling

The single dominant complaint on Trustpilot is refund delay. AI deflection does not shorten a refund cycle. If the warehouse-to-refund cycle remains long, the AI front door buys the team time but does not solve the underlying complaint, and the deflection band lands at the lower edge. The honest version of the engagement says this upfront and pairs the AI work with a policy recommendation on the published cycle.

UK Consumer Rights Act 2015 boundaries

UK B2C returns are governed by the Consumer Rights Act 2015 and (for distance contracts) the Consumer Contracts Regulations. AI replies that promise outcomes outside the statutory regime, or that misrepresent the customer's rights, expose the brand to a different kind of risk than a generic CSAT issue. Any AI flow that touches refund, return, or faulty-goods language needs to be reviewed against the UK statutory baseline before it ships, and the eval set has to include statutory-compliance checks.

Small team, dual-channel

The team size means there is no dedicated CX manager to own the integration backlog full-time. Engagement scope needs to fit inside the existing operational headcount, with the helpdesk and tracker installs accounted for before the AI work begins. Public-surface signal also shows Instagram and TikTok presence, which means a real share of inbound is in social DMs rather than email; the audit above does not cover that channel and a real engagement would extend the diagnostic to social inbox.

Seasonal volume

Occasion-wear has wedding-season and Christmas peaks. Volume across the year is not flat, and the deflection band will look different in peak weeks than in trough weeks. A real rollout staggers the flows so that the highest-leverage one (refund-status) is live before the next return-peak window.

Single-vendor stack risk

The current stack is admirably minimal. Adding a helpdesk plus an AI layer adds two more vendors. Vendor-lock risk is mitigated by keeping the source of truth in the help center (Shopify-hosted pages, not vendor-locked KB) and by using a helpdesk that exports its data cleanly. Any AI training data we generate during the engagement should be portable, not trapped in a single vendor's internal-only fine-tune.

Brand voice drift

OMNES have a recognisable tone (warm, low-key, deliberately uncynical). Generic ecommerce AI sounds nothing like that. Brand-voice tuning is not optional on this brand; it is the difference between an AI front door that improves perception and one that flattens it.

Summary in one paragraph

OMNES are a real product on an admirably minimal Shopify stack with a real customer-experience problem that is mostly upstream of any AI tool. The fastest win is operational: install a helpdesk, publish the refund cycle, surface the returns portal, add the on-site tracker, set an SLA on the Contact page. None of that requires us. What an Eltrus engagement adds on top is the AI front door, the integration layer to make it answer specific questions with specific order context, the eval discipline to keep it accurate, and the brand-voice tuning that keeps the deflection band on the upper edge rather than the lower one. The defensible 12-month impact band is in the £45,000 to £240,000 range, conditional on the policy change shipping alongside the tooling. None of these numbers come from OMNES's tickets, because we have not seen them. Phase B is the version of this audit that does.