3 ways AI and LLMs can accelerate your implementation

These technologies can accelerate every stage of programme delivery, from requirements through to testing, in a way that maintains control, safeguards quality, and unlocks early value.

These tools are not a silver bullet, and they don't replace the need for deep domain expertise, robust governance, or hands-on experience, but when used safely and effectively, can help experienced delivery teams move faster, improve accuracy, and reduce costs.

In this blog, I take a look at 3 ways they can help your system implementation, migration, or upgrade.

1. Speed up requirements gathering

Requirements gathering is often one of the most time-consuming phases of a system change. AI and LLMs can help you to speed up the following tasks, so that you can spend more time validating and challenging the requirements with stakeholders, and less time writing them from scratch:

• Extracting requirements from documentation at scale - rapidly scan and summarise legacy specs, meeting notes, and strategy documents.
• Categorising and structuring requirements - group by theme (e.g., strategies, classifications, interfaces, data) for easier prioritisation and traceability.
• Benchmarking against industry best practice - compare requirements against recognised standards (at arum for example, we have a catalogue of over 500 core requirements) and proven approaches used across similar collections implementations, helping to flag missing components and avoid common pitfalls early.
• Analysing legacy data sets - create data catalogue requirements based on existing data sets from legacy systems.
• Building better business cases - use generative tools to articulate functional requirements and link these directly to benefits and value outcomes.
• 2. Improve design

Design is at the heart of every successful system implementation. LLMs can help you with the following to reduce rework, improve maintainability, and protect long-term value:

• Validating rules engine syntax and prioritisation logic - reduce configuration errors before they hit testing.
• Generating or refactoring SQL, XML, and configuration code - accelerate solution design and data preparation.
• Supporting strategy design with guardrails - ensure collections logic aligns with agreed policy, regulatory expectations, and operational objectives.
• Enforcing design principles - validate consistency across process flows, user journeys, data models, and naming conventions, which helps ensure scalable, supportable solutions.
• Enhancing design documentation - auto-generate diagrams, data dictionaries, or flow summaries to support handover and training.
• 3. Automate testing

Testing is a major contributor to cost and time in any implementation. By combining human-led testing with AI-supported scripting, you can achieve speed while maintaining control, through:

• Generating test cases - rapidly generate test cases based on the business scenarios from requirements, strategy and rule definitions.
• Accelerating manual testing - help users understand expected outcomes, generate test inputs, and summarise outputs automatically. Agentic AI promises to be the next stage evolution of UAT.
• Enhancing test coverage - auto-suggest test edge cases and scenarios across integrations, data quality, and user journeys.
• Assisting with reconciliation testing - support financial and data reconciliation tools with AI-enhanced logic validation, reducing manual effort.
• Streamlining defect triage - summarise logs, suggest root causes, and help testers and developers speak a common language.