Frequently asked questions

Azetria designs and builds websites, business applications and SaaS platforms with the Laravel framework, implements automation based on artificial intelligence, integrates third-party APIs, and works on the visibility of those platforms on Google and in generative answer engines. The agency covers everything from initial discovery to production operations.

Eighteen people in house: backend and frontend developers, architects, AI specialists, SEO consultants, product designers and project leads. We do not subcontract development. On very large engagements we add freelancers who have worked with us for several years.

Both, provided the project has an economic reality. Our portfolio splits roughly between industrial and service SMEs, Series A or B startups, and business units of large groups. The criterion is not size but the presence of someone able to decide.

Yes. We regularly work with clients in Belgium, Switzerland, Morocco and Canada. Communication happens in French, English or Arabic depending on the teams. On-site visits are possible for discovery phases and major workshops.

Yes, and it is about a third of our work. We always start with a two-to-five-day audit: reading the code, reviewing the architecture, assessing debt and security risks. The audit is paid for, and its report is yours even if you then decide not to work with us.

A consultancy sells time; we sell a delivered outcome. In practice: fixed prices per work package, responsibility for architecture decisions, and contractual performance commitments. In exchange, we require a serious discovery phase before quoting.

Between €650 and €1,500 excluding VAT for a five-to-fifteen-page site including bespoke design, development, an editorial back office, the SEO foundation and go-live. Below €600 you are buying an adapted theme, which we do not offer — other providers do it perfectly well at that price.

The realistic floor is around €2,800 excluding VAT for a first scope usable in production. Most of our business applications land between €4,000 and €9,000 depending on the number of processes covered, the integrations required and security expectations.

A first commercialisable version starts around €4,500 excluding VAT and frequently exceeds €10,000 when multi-tenancy, recurring billing, a public API and an admin dashboard are needed. We recommend splitting: a narrow scope to market quickly, then packages funded by early revenue.

Fixed per work package in the vast majority of cases, which requires prior discovery. Time and materials is only offered for ongoing support or team reinforcement, at a day rate of €75 excluding VAT.

An overrun caused by our own mis-estimation is on us: that is what a fixed price means. An overrun caused by an out-of-scope request is covered by a priced amendment approved before any development. Over the last three years, the average gap between initial quote and final invoice is under 6 %.

Yes, between €350 and €900 excluding VAT depending on complexity. It produces a deliverable that belongs to you: detailed scope, data model, target architecture, schedule and a firm price. If you entrust us with the build, that amount is deducted from the development fee.

Between 12 % and 20 % of the initial development cost per year. That covers security updates, defect fixes, monitoring, backups and a pool of hours for improvements. An unmaintained application becomes vulnerable within eighteen months.

From one week after design approval, content supplied. A simple site fits in that week; as soon as the scope grows — multilingual, client area, integrations — allow ten to fifteen days. The timeline is settled during discovery and stated before we start, never discovered along the way. Copywriting is the most common cause of delay: allow three to five extra days if you entrust it to us.

One to two weeks for a first scope in production, discovery included. We deliver in daily increments: a first module is usually usable by day four, which lets you start training users while the rest advances.

Yes, by putting more people in parallel, with a 35 % surcharge. Compression has a physical limit: beyond roughly 25 % of gain, adding developers slows the project down because of coordination cost.

Usually three to five working days after signature. We refuse to start without the full team available: on a one-to-three-week cycle, beginning with half a team makes the promise unkeepable.

We deliver in daily increments, within a fixed-price framework. On a one-to-three-week project a two-week sprint would make no sense: the staging environment is updated on every merge and the progress check is daily. The overall scope is set during discovery; what remains adjustable is the order of delivery.

A decision-maker genuinely reachable for the duration of the project, access to your existing systems if an integration is planned, and a clear idea of the outcome you want. A detailed specification is not required: discovery exists precisely to produce it. On a one-to-three-week cycle, an approval that slips by three days shifts everything by the same amount.

Around one hour a day for the project lead, plus half a day for business users at acceptance. The day-one discovery workshop needs two hours of your time.

A single project lead for coordination and the architect responsible for your platform for technical matters. Both attend the first meeting and stay the same throughout the engagement.

Yes. Any new request is priced in days and euros, and you decide: add to the budget, replace a planned feature, or defer to a later phase. Nothing is built without written approval.

The contract includes an exit clause at the end of each work package: you keep the code produced, the documentation and the access, and you pay only for delivered packages. We prefer a clean separation to an engagement that drags on.

Three measurable reasons: a mature ecosystem covering queues, cache, authentication and testing without exotic dependencies; a large community, so you can hire for your own team; and a predictable release cadence with multi-year security support. Developer comfort matters, but it would not be enough on its own.

Only to take over an existing site the client wants to keep. We do not start new projects on WordPress: the security debt from third-party plugins and the difficulty of evolving the data model beyond a blog make the total cost higher in the medium term.

For type robustness, the quality of the JSON implementation (indexable jsonb), partial and functional indexes, and extensions such as pgvector that enable semantic search without an extra database. On a long-lived business application, those differences translate into saved development hours.

Mandatory code review before any merge, PHPStan static analysis at level 6 minimum, automatic formatting via Laravel Pint, a test suite run on every commit and a performance budget checked in continuous integration. None of these checks is optional.

Yes, targeting 80 % coverage on business logic and 100 % on critical paths such as payment or authentication. Tests are not a separately billed option: they are part of development.

We target WCAG 2.2 level AA and run an audit before launch: full keyboard navigation, verified contrast, semantic structure, alternative text, focus management and screen reader compatibility. The compliance report is handed over to you.

Yes by default: our environments are hosted in France or Germany with providers subject to European law. Hosting in another region is possible if your business requires it, with the corresponding contractual clauses.

Laravel is a web development framework written in PHP, released in 2011 and now the most widely used in its ecosystem. It provides structure and ready-made components — routing, database, authentication, queues, cache, testing — that remove the need to rewrite an application's foundations each time.

Yes. Limits encountered in production almost always come from database queries, caching and architecture rather than the framework. With well-indexed PostgreSQL, Redis for cache and sessions, and Octane where needed, a Laravel application handles several thousand requests per second on modest infrastructure.

Concretely: business logic isolated in services and actions, data access through interchangeable repositories, typed data transfer objects instead of associative arrays, and controllers that only orchestrate. The practical result: changing the data source or the rendering layer does not touch business logic.

Through Laravel queues supervised with Horizon: sending emails, generating documents, calling external APIs, AI processing. An HTTP request should never exceed a few hundred milliseconds; everything else runs in the background with explicit retry rules.

Yes, it is a significant part of our work. The approach depends on volume: progressive rewriting with both systems coexisting for critical applications, or a full rebuild when the existing codebase is too degraded to serve as a reference.

An AI agent is a program that receives an objective, decides for itself which steps to take, calls tools — your API, a database, an external service — and checks its result before handing back. The difference from a simple language model call is autonomy across several steps and access to real actions.

Those where a repetitive task consumes human time on text or documents: qualifying inbound requests, extracting data from invoices or contracts, drafting first versions, answering internal questions from your documentation, classification and routing. Return is calculated in hours saved, not in demo effect.

Through three cumulative measures: grounding answers in your documents with source citation, format constraints verified by code before display, and mandatory human validation on any irreversible action. A system with no source to answer from must say it does not know — that is a design requirement, not an option.

It depends on the architecture, and you decide. With enterprise offerings from OpenAI and Anthropic, data is not used to train models and retention is contractually limited. If your sector requires it, we deploy open models on European infrastructure, at a higher running cost.

With an evaluation set built from real cases in your business, run on every prompt or model change. We track correct-answer rate, abstention rate, latency and cost per request. Without that, every change is a gamble.

Replace it, no; absorb repetitive requests, yes. The deployments we observe handle 40 % to 60 % of first-level requests automatically, with immediate handover to a human as soon as confidence is insufficient. Promising more would be dishonest.

The technical foundation always is: URL structure, semantic markup, structured data, performance, redirect plan, sitemaps and robots file. Content strategy and link building are a separate engagement, because they require continuous work over several months.

Technical fixes take effect within three to eight weeks. A content strategy takes four to eight months to become visible on competitive queries, and twelve to eighteen months to reach full potential. Any promise of results within a month should be treated as a warning sign.

No, and nobody seriously can: ranking depends on competitive factors outside our control. We do commit to deliverables and verifiable intermediate indicators: indexed pages, technical coverage, performance, and progression on a tracked basket of queries.

Three metrics Google measures on real visits: LCP, the time to render the largest visible element, to keep under 2.5 seconds; INP, responsiveness to interactions, under 200 milliseconds; and CLS, visual stability, under 0.1. They influence ranking and, more importantly, bounce rate.

A poorly prepared one does: the industry average loss exceeds 30 %. It is avoidable with an exhaustive inventory of existing URLs, a 301 redirect plan verified one by one, retention of high-performing content and heightened monitoring for six weeks after cutover.

No. We build authority through citable content, specialist press relations, partnerships and legitimate professional directories. Buying links risks a manual penalty that often takes more than a year to recover from.

GEO, or Generative Engine Optimization, covers the techniques used to get a brand or a piece of content cited in answers produced by generative engines: ChatGPT, Gemini, Claude, Perplexity or Google's AI overviews. The goal is no longer to earn a click on a link, but to be the source a system uses to build its answer.

SEO optimises a page's ranking in a list of results; GEO optimises the probability that a specific passage is extracted and cited. The levers differ: self-contained answers at the top of sections, figures attributed to a source, explicit definitions, comparison tables, identifiable authors and entity consistency across the web.

No. Generative engines largely rely on the classic search index to select their sources: content invisible on Google has little chance of being cited by ChatGPT. GEO layers on top of SEO, it does not replace it.

By regularly querying the main assistants with a basket of questions representative of your market, and recording how often your brand is cited, in what context, and which competitors appear. We complement this with analysis of referral traffic from those assistants' domains in your server logs.

For a site that publishes content in order to be found, no: blocking GPTBot or ClaudeBot means excluding yourself from generated answers. The question is different for a publisher whose content is the product being sold. We document this choice in each project's robots.txt.

A Markdown file placed at the site root that presents the entity, its offering and its reference content in a form directly usable by a language model, without parsing HTML. The convention is not an official standard, but its setup cost is negligible against the potential clarity gain.

Security updates for the framework and dependencies, defect fixes, uptime and error monitoring, tested daily backups, a monthly activity report and a pool of improvement hours that can roll over for three months.

No, but it conditions the warranty beyond the initial six months. An unmaintained application accumulates known vulnerabilities: after eighteen months without updates, catching up usually costs more than two years of contract.

On dedicated servers or European cloud depending on needs: Scaleway, OVHcloud, Hetzner or AWS Paris region. Cloudflare always sits in front for protection and content delivery. You can remain the owner of the accounts if you prefer.

99.9 % on standard contracts, with monitoring every minute and immediate on-call alerting. A redundant architecture allowing 99.99 % is possible; it multiplies infrastructure cost by two to three, which is not justified for every use case.

Yes, with no obstacle from us. The code, architecture documentation, access and deployment procedures are handed over. We provide two weeks of knowledge transfer to the incoming team free of charge.

You do, entirely, once the balance is paid. The transfer covers all economic rights and is stated explicitly in the contract. The Git repository is hosted in your organisation from day one, not transferred at the end.

No. We only integrate open source libraries under permissive licences, listed in the contract. No feature of your application depends on an Azetria component you would have to pay to use.

Only with your written agreement, and you approve the text before publication. Some clients accept a mention without figures, others a full case study, others nothing at all. Declining has no effect on the relationship.

We act as a processor under the GDPR, with a data processing agreement setting out purposes, retention periods, security measures and sub-processors. We support you with your record of processing activities and impact assessment where required.

Six months on any defect within the delivered scope, free of charge. Enhancements and new requests are not covered by the warranty and are priced separately.