AI frameworks and agent build
Customer-facing assistants that help, disclose, and hand over
Product-help assistants, configuration and troubleshooting guides, order and account support, conversational search: AI your customers talk to directly, built on twenty years of ecommerce platform work, with the transparency and guardrails the EU AI Act expects.
The customer is not inside your firewall
An internal assistant has a forgiving audience. A customer-facing one does not: it represents the brand, it can be probed and manipulated, it handles personal data, and under the EU AI Act it carries a transparency duty, so users must know they are interacting with AI. The build is the same grounded architecture as a knowledge assistant, with three things added: disclosure, prompt-injection and abuse guardrails, and a handover to a person that works.
The group has built customer-facing platforms for retailers, publishers and manufacturers since 2006, including the order portals, account areas and support flows these assistants now sit inside. That is why the integration is usually the easy part.
Four patterns
Product configuration and troubleshooting
For technical products: which option fits which requirement, what a fault code means, what to check first, drawn from manuals, parameters and known-fault histories, with escalation to support when the case needs it.
Conversational commerce and search
Finding the right product by describing the need rather than guessing the keyword, with answers grounded in live catalogue data and the store's own policies. Our Shopify and BigCommerce heritage applies directly.
Order, account and service support
Where is my order, change my delivery, what does my warranty cover: grounded in the systems of record, with the actions it may take defined and permissioned.
Multilingual support
The same assistant serving customers in their language, with translation grounded in approved terminology, for businesses selling across Europe and beyond.
Built in, not bolted on
Disclosure
Clear, early and repeated: the customer knows it is AI, knows what it can and cannot do, and knows how to reach a person.
Grounding and refusal
Answers only from approved sources and live data; questions outside scope are declined politely rather than improvised.
Prompt-injection and abuse defence
Inputs screened, instructions separated from content, tool actions limited and logged, and adversarial scenarios in the evaluation set from day one.
Handover
Confidence-based escalation to a person with the conversation and the assistant's reasoning attached, inside the helpdesk you already run.
Privacy and data protection
Personal data handled under your data-protection obligations, with retention, access and no-training terms documented in the scope.
Pilot on one journey, then widen
Two weeks of scoping (journeys, sources, actions, guardrails, success measures and the evaluation set, including adversarial cases), four to eight weeks of build and integration, then a pilot on one journey with monitoring and a person on every escalation. Widening to further journeys follows the evidence. Scoped and fixed before build;
About customer-facing AI
Will customers accept talking to AI?
They already do, when it is fast, accurate and honest about what it is, and when a person is one step away. Disclosure improves trust rather than reducing it.
What about the EU AI Act?
Customer-facing chatbots are limited risk, with a transparency duty. Where an assistant influences access to a service or handles a safety-relevant product, higher-risk obligations may apply; we classify each journey in scope.
Can it take actions, such as changing an order?
Yes, where you decide it should, through permissioned tools with thresholds and an audit trail. Many clients start read-only and add actions once the pilot has earned trust.
Which platforms do you integrate with?
Shopify, BigCommerce, Adobe Commerce and bespoke platforms on the commerce side; Zendesk, Freshdesk, HubSpot and others on the support side; ERPs and order systems by API.
Start with the journey that generates the most tickets
That is where a customer-facing assistant pays back fastest.