LLM strategy: Copilot, Claude or ChatGPT?
Which AI model for which purpose?
ISO 27001 · Microsoft Solution Partner
The question is no longer whether generative AI is entering the business - but which one, for whom and under what terms. Those who purchase licences indiscriminately today pay twice: first for unused subscriptions, then for consolidation. An LLM strategy answers this question systematically.
Which problem does an LLM strategy solve?
In many companies, AI usage is already a reality - just uncontrolled: employees use private ChatGPT accounts, specialist departments test tools without approval, and sensitive data ends up in unknown clouds. At the same time, potential is left untapped because no one authoritatively clarifies which tools are permitted, paid for and trained on.
An LLM strategy delivers both: clear guardrails for secure deployment and a well-founded selection of tools that suit your tasks.
The candidates at a glance
- Microsoft 365 Copilot: Strong wherever your data already resides - Outlook, Teams, Word, Excel, SharePoint. The prerequisite is a tidy Microsoft 365 environment with a clean permission concept; this is precisely what our Copilot Readiness Assessment.
- Claude (Anthropic): Distinct strengths with long documents, analysis and writing tasks, as well as handling complex instructions.
- ChatGPT (OpenAI): Broad ecosystem and high awareness among the workforce; relevant for businesses are the Team and Enterprise variants with contractually regulated data usage.
- Open-source models (e.g. Gemma, Llama): Full control over data and costs, operated in your environment - the path for sensitive data and regulated industries, see Self-Hosted AI.
The answer is rarely "one size fits all". What works best in practice is usually a portfolio: a standard tool for the majority, specialised models for individual use cases - and clear rules on what is allowed where.
Claude Code: AI in software development
Alongside assistants for the wider workforce, there are specialised tools for development. Claude Code works directly within the repository and terminal: it reads existing code, modifies files, executes tests and documents changes transparently. For teams with in-house software development or an extensive scripting and automation landscape, this is often the point with the fastest measurable impact.
The framework is crucial: repository access, approval processes, handling customer and production data, and the question of which models may run where. NMS establishes this framework - including connection to the existing Microsoft 365 identity and, where necessary, self-hosted models.
How NMS proceeds
Assessment: What tasks arise, which tools are in use (officially and unofficially), which data is affected?
Evaluation and selection: Models and licence models compared - evaluated by benefit, cost, data protection and integration effort, vendor-neutral.
Guardrails: Usage policy, data classification and technical protection, aligned with AI Governance for Microsoft 365.
Rollout with enablement: Rollout in waves, accompanied by training, so that the licences actually deliver real impact.
- Step 1 of 4AssessmentWhat tasks need to be performed, which tools are in use, and what data is involved?
- Step 2 of 4Evaluation and selectionModels and licensing models compared and assessed on a vendor-neutral basis.
- Step 3 of 4GuardrailsUsage policy, data classification and technical safeguards.
- Step 4 of 4Roll-out with enablementRoll-out in waves, accompanied by training.
First understand what is needed, then make a selection. And put guardrails in place before roll-out, not afterwards.
Who is this worth it for?
For every company looking to deploy more than a handful of AI licences - and for every business where AI is already being used without regulation today. The earlier the guardrails are established, the more cost-effective the journey becomes.
Copilot or ChatGPT - which is better for businesses?
That depends on where your work takes place. If your daily routine revolves around Outlook, Teams and Office, Microsoft 365 Copilot demonstrates its seamless integration. For open text and analysis work, Claude or ChatGPT are often the stronger choice. In practice, a combination with clear rules of use frequently proves best.
How much does Microsoft 365 Copilot cost?
Copilot is billed per user per month in addition to the Microsoft 365 licence; the terms change regularly. More important than the list price is the question of which user groups actually benefit - our assessment provides the decision-making foundation for this.
Are employees allowed to use private AI accounts for work?
Without regulations, this poses a significant data protection risk: inputs can be stored and processed by the provider. A usage policy plus approved corporate tools solves the problem - experience shows that bans alone do not work.
Do we need a separate AI tool for every department?
No. A standard tool for the majority plus targeted additions for special cases (e.g. development, legal, marketing) has proven effective. This keeps costs, training effort and governance manageable.
Solutions in a new dimension.
One conversation is enough to find out where IT, Microsoft Cloud and AI can take real weight off day-to-day business.
Your contact: Jan Emmerich, Managing Director