A two-week, €2,500 audit of your support inbox. Founder-delivered, 200 to 500 tickets sampled, a written 90-day plan you keep whether you engage further or not. If we can't see a path to 30 percent deflection at acceptable CSAT, we tell you that too.
Ten business days from kickoff to deliverable. Credits against setup if you proceed within 30 days.
That chat bubble is a live Eltrus agent, the real product, not a contact form. Ask it anything about the audit.
Targets, not promises. The cost-side numbers sit inside the published deflection band for native helpdesk AI; the revenue-side ones are modelled on a brand ready for Tier 2. The audit produces your real numbers, on your shop.
The realistic published deflection band for native Gorgias AI is 26 to 56 percent. We target the upper half of that range, not above it.
Conservative total: €100 to 150k per year on a brand that size. Real numbers come out of the audit on your shop, not a generic spreadsheet.
We tested the public support surface of 52 UK, Irish, and EU Shopify brands exactly the way we would test yours: a logged-out shopper, a real browser, the three questions your customers actually ask. The same few gaps came up again and again, and each one quietly loses orders and goodwill. This is what we keep finding, and roughly what it costs.
What that costs you
A brand doing 4,000 orders a month is sitting on 150 to 250 "where is my order" messages a week, every one of them waiting. That is inbox hours, slower replies where a human is actually needed, and a quiet drag on repeat purchase. The audit puts your real number on it.
Every number above is counted from our own research, gathered May to June 2026 from public storefronts only, nothing accessed beyond what any shopper sees. The ticket data behind each storefront stays with the brand. The cost estimates here are industry norms and our modelling, not numbers from a client deployment. The audit replaces them with yours.
Tuned inside the helpdesk you already use
Six weekends of your ops lead's attention, then it drifts the moment they get busy. The retainer is someone whose only job is keeping it tuned.
Good at selling their own tool, gone once you are live. We tune what you already run, and stay long enough for it to compound.
We report verified-resolution and seven-day re-contact, the numbers that tell you whether deflection is real or the tickets just came back through another door. In writing.
Real findings, brand names withheld. These are not hypotheticals. Each was watched loading in a real browser and written down verbatim.
The full Gorgias chat scaffold sat in the theme. The widget never fired. The shopper met a contact form. We saw this exact shape in 15 of 37 brands: a paid subscription doing nothing the customer can see.
A skincare widget, offline evening and mid-morning both. "Is this safe in pregnancy?" got a form, not an answer. The contact page calls chat "the quickest way to reach us."
A gifting brand headed a section "Live Chat." Beneath it: "out of office hours we will get back to you the following working day." No widget ever loaded. On a deadline gift, that is a 24-hour wait.
We typed "Where is my order? It hasn't arrived." It replied "Have you placed an order with us, Guest?", took the question as the name, and forced a dropdown. Zero understanding, on the most time-sensitive question there is.
A capable AI agent listed "Do you ship worldwide?" as a quick reply. We asked about shipping. It said "I'll pass this to the right team" and escalated. Ten shipping articles in the help centre, all ignored.
We published a complete public-surface audit of OMNES, a UK womenswear brand (2.9 stars across 648 Trustpilot reviews at the time of the audit). Stack fingerprint, ticket-mix inference, deflection bands, a modelled cost of the gaps, every functional claim verified in a real browser. We have not been retained by OMNES and have never seen one of their tickets. It is the closest thing to "here is exactly what you get" that we can show before a first client.
Read the OMNES sample auditThree phases. You can stop after any of them.
A pseudonymised sample of 200 to 500 recent tickets, categorised by flow. You get a written report and a walkthrough call. If the inbox isn't worth deflecting, we tell you.
Knowledge base authored on your top five flows. The helpdesk's AI configured on those flows with confidence-based escalation. A small integration layer covers the Shopify-specific work the native AI doesn't handle (older order lookup, subscription mutations, 3PL reconciliation). The agent replies in your customer's language by default, auto-detected across EN, FR, DE, ES, IT, and others. First week monitored daily.
Weekly tuning on every escalated ticket. KB maintenance as your catalogue and policies change. Monthly review call. Quarterly macro audit. Three-month rolling term, then 30 days notice.
Every engagement starts with the audit. If you proceed within 30 days, the audit fee credits against setup.
A PDF and a spreadsheet of what's wrong in your current support setup and what to fix, in priority order. No code, no platform access beyond read-only.
We tune the helpdesk's AI on your real ticket mix, write the knowledge base, and run a small integration layer where the native tooling falls short on Shopify work. Weekly tuning after.
Everything in Tier 1, plus refund-to-exchange, subscription save, abandoned cart conversational, and pre-purchase product Q&A. For brands above €15M GMV with active subscription or returns volume.
Fees in euro, exclusive of VAT. Helpdesk subscription and AI API spend stay on your account; we invoice our fees only.
Most Shopify support inboxes resolve to three or four repeating shapes. We find yours, write the responses, configure the routing, and stay on it.
Order status, shipping delays, tracking lookups, address changes inside the cancellation window. Boring, repeatable, exactly what AI is for.
Return windows, refund timing, exchange offers. A well-placed exchange option recovers revenue that would otherwise leave the brand.
Pause, swap product, change cadence, update card. Routine subscription work, resolved before it reaches a human agent.
Typical ranges from published industry data and our 52-brand public-surface research. The audit produces your actual numbers.
Granting access to a support inbox is the highest-trust moment in this engagement, so we scope it tightly and say exactly what we do with what we read.
All of it sits under a GDPR Article 28 DPA with a named sub-processor list and a written AI Governance Schedule covering refund authority and policy gating.
Eltrus is an Irish Ltd run by one engineer. The reason a one-person shop can credibly ship this: a surgical-practice CRM live in production in Galway, a digital twin of the Irish power grid that reached EirGrid's CleanerGrid final, photonics research at Tyndall, and StudySmith, a full study platform built and launched solo from backend to billing. The hard part here is not the model. It is someone who will sample the tickets, write the knowledge base in your voice, and own the result.
You talk to the person who configures your AI and writes your KB articles, not an account manager. If you are unavailable, what happens to my support? Answered plainly under single-operator continuity on the trust page.
About EoghanBook the €2,500 diagnostic, or take a 30-minute call to check the fit.
