Chapter 4
Where to start with AI:
a practical roadmap
Strategy matters more than speed.
Rushing into AI investments runs the risk of spending budget on technology that fails and outcomes that aren’t achieved. But early AI projects have the potential for costs to grow by 5-10x of what was initially expected, according to Gartner. Planning (no matter the size of an AI project) gives you time to evaluate existing infrastructure, create a realistic list of what’s needed to really support a Trust’s AI workload, and assess technology partners with a full understanding of the task.
Crawl, walk, run
There’s a lot that network teams can do with AI, which makes it difficult to find a starting point. We’ve found that opting for a phased approach is a good option to prove the value, find the resource, and set expectations.
1. Crawl: Information readiness
2. Walk: Assisted wins
3. Run: Autonomous operations
(We show an example of this below)
Sticking to a methodical roadmap will help you maintain clinical resilience, address talent shortages, and mitigate risks before impact.
Crawl
Foundations and information readiness
Audit
Understand your team’s current state of operations. Where are your team spending most of their time?
Fix workflows
AI has to be applied to clean and tidy workflows. Are there any messy processes you need to get in order?
Digitise data
Make sure all your workflows and technical data (such as topology and documents) are in a digital system with open APIs so they can integrate with AI.
Establish a secure LLM
Provide a secure, private LLM environment to prevent sensitive information leaking into public domains.
Walk
Targeted walk and assisted operations
AI assistance
Use LLMs for contextual diagnostic assistance, such as interpreting cryptic log messages or finding known bugs in vendor documentation.
Automate admin
Target repetitive activities, such as daily environment checks, which usually involve multiple manual checks and recording values.
Alert correlation
Implement AIOps to consolidate correlating individual alerts into one incident, reducing alert fatigue for teams.
Human-in-the-loop governance
Introduce a decision-gate where AI suggests a technical change or resolution but a human engineer has final approval.
Run
Autonomous operations and clinical resilience
Agentic AI
Deploy autonomous agents to manage complex workflows, automate deployment, and add capacity.
Predictive maintenance
Move to a proactive approach, where AI monitoring detects degradation in network performance and alerts the team before impact.
Partly autonomous network
Reach a state where the network can self-heal from common faults, resolving issues in minutes rather than hours.
Advanced resource planning
Use AI to analyse timesheets and incident logs to identify gaps in service, capacity plan effectively, and support digital team members.
Is your Trust ready for AI?
Getting the basics right is important before jumping into AIOps. Cisco has released a useful benchmarking tool to help NHS Trusts pinpoint how ready they are to invest in AI workloads. The assessment is a quick survey and free to use.