Questions we are asked before signing anything
Direct answers on cost, ownership, AI, automation and what happens after launch. If yours is not here, email info@nexlavision.com and we will answer it properly.
Working together
A focused internal tool is typically six to twelve weeks. A platform replacing a core business system is usually four to nine months, delivered in releases rather than one launch. Discovery gives you a defensible range before you commit to the full build.
Cost tracks scope, integration count and how well-defined the requirements are. Rather than quote a number blind, we run a short paid discovery that produces an architecture outline and a cost range you can plan against. That output is yours whether or not you build with us.
With a free scoping call, then a paid discovery of one to three weeks. Discovery produces an architecture outline, a delivery sequence, identified risks and a cost range. If you already have specifications, we review those instead.
Yes. Intellectual property transfers to you on completion of payment. Code lives in your repositories from day one, together with infrastructure definitions, tests and documentation.
Often. We start with a code and architecture review that tells you honestly what is salvageable, what should be rewritten and what each option costs before anyone commits to continuing.
The two that actually bite are dependency and opacity: a system only the supplier understands, and progress you cannot verify. Both are addressable. Insist on your own repositories, documentation and infrastructure-as-code, and on reviewing working software rather than status reports. We build that way by default, and you should require it from anyone you engage.
AI, chatbots and agents
Retrieval-augmented generation grounds a language model in your own documents rather than its training data. You need it whenever answers must come from your content — policies, contracts, manuals, tickets — and must be traceable to a source. It is the right approach when accuracy and citation matter more than open-ended generation.
No. We use providers and configurations where customer data is excluded from training, and for sensitive workloads we deploy models in your own cloud tenancy or on self-hosted infrastructure. Data handling terms are agreed in writing before any build begins.
Running cost is driven by traffic, model choice and how much context each request carries. We measure cost per request during the prototype and design against a target — retrieval strategy, prompt size and caching all move the number materially. An unexamined architecture decision becomes a recurring bill.
By constraining what it can answer from, and by measuring. We ground responses in retrieved sources, build an evaluation set from real examples, and test every change against it. Where a wrong answer is costly, we add human approval before the action rather than after.
A model that can call tools and take multiple steps toward a goal rather than answering a single question. Useful when a task needs several lookups or actions in sequence. In production the important design work is the boundary: which tools it may call, what it may modify, spending limits, and where a human has to approve before something irreversible happens.
Workflows and n8n
A workflow you can see is a workflow your team can change. n8n provides a visual pipeline with run history, error handling and retries built in, and it can be self-hosted so data stays on your infrastructure. Where a step genuinely needs custom logic, n8n runs code nodes — so it is not a choice between the two.
Ones that are repetitive, rule-based, span more than one system and carry a real cost when they go wrong. If a process has none of those traits, automation adds a system to maintain without removing much work — and we will tell you so.
It should fail loudly and recover safely. Every workflow we build has explicit failure branches, retry with backoff, dead-letter handling for anything unrecoverable, and alerting to a person. Silent failure is the one outcome that is genuinely worse than no automation at all.
ERP, cloud and life after launch
Buy, in most cases. Established products encode decades of edge cases you would otherwise rediscover. Building makes sense when your operating model is genuinely unusual, when licensing at your headcount is disproportionate, or when you need one or two modules rather than a whole suite.
That is the normal case. Most projects involve integrating with an ERP, CRM, accounting package or legacy database that is staying put. We treat replacement as a last resort rather than a default.
Usually the one your organisation already has skills and agreements for. We work across Azure and AWS, and the difference matters far less than how the application is built. We design so the workload is portable rather than dependent on a single provider's proprietary services without good reason.
Role-based access, audit logging, managed secrets, dependency scanning and encryption in transit and at rest are baseline, not extras. For regulated work we agree the security model in writing during discovery and design to it, rather than retrofitting before a review.
Support is arranged to match how critical the system is — from a maintenance retainer with defined response times through to a continuing development team. Either way, monitoring and alerting are configured before go-live rather than after the first incident.
Still have a question?
Ask it directly. You will get an answer from someone who builds the systems, not a sales response.