AI Implementation and strategy
AI only works once the foundation is in place.
First foundation and infrastructure, then processes. Lived processes produce structured data, which comes together in one curated place. From there, AI can work across the whole company.
01 The order
Foundation first. Then AI.
Each stage rests on the one before. Skip one, and AI works on data that nobody can trust.
- 01
Foundation and infrastructure
Identities, devices, network, security, cloud. Stable and documented.
- 02
Processes
Core processes described and actually lived. Each one as a playbook: who does what, when, with which system.
- 03
Structured data
Only emerges from lived processes: consistent fields, clear owners, known quality.
- 04
Curated data location
Data from the systems brought together, cleaned, with access rights. One reliable source.
- 05
AI across the company
Assistants, agents and reports work on this curated location. Traceable, with a rights concept.
Buy the tool, sort out the data later. This shortcut saves weeks at the start. Afterwards, every answer of the AI has to be checked by hand, because nobody knows which data it is based on.
How far a company has come on the way to stage 05 is measured by NextEra Copilot: its scale runs from level 1, in someone’s head, to level 5, AI-autonomous.
02 The difference
Automation or AI?
Automation follows a rule. AI weighs things up.
Rule
Automation
- Fixed rule: if A, then left, otherwise right.
- Same input, same output.
- Fast, cheap, auditable.
- Breaks on exceptions that nobody foresaw.
Judgement
AI
- Weighs many options: scores from 1 to 10, alternatives from A to Z.
- Evaluates the context, then decides or proposes.
- Handles unstructured input such as emails, documents and free text, and catches exceptions.
- Needs clean data, guardrails, a human for consequential decisions, and a log.
The best solutions combine both: automation carries the process, AI decides where judgement is needed.
03 Examples
Six examples from everyday work.
Generic examples, not case studies. Each shows which stage it builds on.
- Automation
Onboarding new employees
- What happens
- Accounts, licences and devices are set up by fixed rules as soon as the contract is signed.
- What it needs
- Stages 01 and 02: clean identities and a described joiner process.
- Combination
Incoming invoices with approval
- What happens
- AI reads invoices and assigns them, the automation routes them for approval by amount and cost centre.
- What it needs
- Stages 02 and 03: an approval process and consistent master data for suppliers and cost centres.
- AI
Pre-sorting requests and tickets
- What happens
- AI reads incoming requests, recognises topic and urgency and proposes priority and owner.
- What it needs
- Stage 03: categories and responsibilities that are lived in the process.
- AI
Supplier and contract review
- What happens
- AI compares contracts with your own requirements and flags deviations. A human decides.
- What it needs
- Stage 04: requirements and model contracts in a curated location.
- Automation
Monthly reporting from several systems
- What happens
- Key figures are pulled from the curated data location every month and reported in the same format.
- What it needs
- Stage 04: a data location where the systems already come together.
- AI
Internal knowledge assistant
- What happens
- Employees ask about policies and procedures and get answers with a source.
- What it needs
- Stages 04 and 05: curated knowledge with access rights, so everyone sees only what they may see.
04 Guardrails
Guardrails from the start.
AI in a company needs rules before it touches decisions.
- Data protection Personal data in AI applications falls under the Swiss FADP and, where the EU is involved, the GDPR.
- EU AI Act The EU AI Act classifies AI systems by risk. The higher the risk, the more obligations.
- Human oversight Consequential decisions are made or confirmed by a human.
- Logging and traceability Every AI decision is logged with input, result and source.
The overview of all rules is under Regulation.
FAQ Answers
Questions about AI implementation.
Where do you start with AI?
With the foundation, not with the tool. First, identities, devices, network, security and cloud must be stable and documented, then the core processes. Only processes that are actually lived produce structured data that AI can work with reliably.
Why not start directly with an AI tool?
An AI tool is only as good as the data it receives. Without described processes and consistent data, it delivers answers that nobody can verify. A short test makes sense, but it is not enough as a basis for use across the whole company.
What is the difference between automation and AI?
An automation follows a fixed rule: if A, then left, otherwise right. Same input, same result. AI weighs up: it scores several options, takes the context into account and decides or proposes. Good solutions combine both.
What is a curated data location?
A place where data from the various systems is brought together, cleaned and given access rights. It is the one reliable source on which reports, assistants and agents work. Curated means: someone is responsible for content and quality.
Does every company need AI?
No. Many tasks are solved faster, cheaper and more verifiably by a good automation. AI pays off where judgement is needed: with unstructured input, many options and exceptions. The question is not whether to use AI, but where it does something better than a rule.
What does the EU AI Act require?
The EU AI Act classifies AI systems by risk: prohibited, high, limited, minimal. The higher the risk, the more obligations, from transparency to documentation and human oversight. It also applies to Swiss companies if their AI systems are placed on the market in the EU or their output is used there. The obligations for high-risk systems apply from December 2027.
Which stage is your company on?
Thirty minutes, no presentation.
