The Virtual Agent Deadline at Northwind Freight
Northwind Freight runs a 14-person service desk supporting 3,200 drivers and warehouse staff. First-contact resolution is 48%, and 62% of tickets are password resets and 'how do I' questions. The knowledge base was last reviewed 18 months ago; 40% of its 900 articles are marked obsolete. The CIO has approved budget for an AI virtual agent after a vendor demo and wants it live in 10 weeks. Budget is fixed at £120,000, and there is no separate content budget. The service desk manager wants to clean the knowledge base first. A union agreement requires consultation before any change that affects how staff request help. The vendor's reference customer had a cleaner catalogue and a two-year-old knowledge base. The programme manager says, 'We can buy the tool and fix content later.'
1. Which two ITIL guiding principles are most relevant to deciding whether to deploy the virtual agent in 10 weeks? Explain how each would change the decision.
Read the worked answer
Focus on value and start where you are are the strongest pair. Focus on value asks what outcome the desk is enabling - drivers getting back on the road and staff getting answers - not whether the tool can be bought. With 62% of tickets being simple resets and how-to questions, a virtual agent could create value, but only if answers are accurate. A knowledge base with 40% obsolete articles means automation would scale wrong answers and increase reopens. Start where you are does not mean keeping the legacy portal forever. It means using the 900 existing articles, the ticket taxonomy and 18 months of data as the starting asset, then improving them. The vendor's reference customer is not comparable because its catalogue and knowledge were cleaner. Together the principles support a narrow pilot on a few high-volume intents after content triage, not a 10-week full launch. Progress iteratively with feedback supports the same conclusion, but the decision hinges on value and on starting from current assets rather than a demo.
2. The programme manager says, 'Buy the tool and fix content later.' Evaluate this argument against the guiding principles and the constraints.
Read the worked answer
The argument treats acquisition as progress, but optimize and automate says automate only after simplifying and improving the work. Automating a 40% obsolete knowledge base would automate error. It also conflicts with focus on value: the vendor demo measures demo quality, not Northwind's outcomes. The 10-week deadline is a constraint, not a value statement. A short pilot can still meet it: clean the top intents, test with a small user group, and measure containment and reopen rates. The union agreement adds a people constraint; consultation is not a reason to delay indefinitely, but bypassing it would create resistance and risk. The fixed £120,000 budget also means a full content programme cannot be assumed. The programme manager is right that content can be improved iteratively, but not that it can be deferred until after launch; the first iteration must include content triage. Otherwise the tool will damage trust and make later adoption harder.
3. Recommend a first 10-week iteration and three measures that would show whether value is being created.
Read the worked answer
Weeks 1–2: analyse 12 months of ticket data, identify the top 10 intents, and mark 120 articles for rewrite or retirement. Weeks 3–4: consult the union on the new request channel and train 14 analysts on content ownership. Weeks 5–8: pilot the virtual agent with 200 users on password reset and five how-to intents only; keep a human escalation path. Weeks 9–10: review feedback and decide whether to expand, pause or change vendor.
Measures should be outcome-based: containment rate for piloted intents (resolved without human), reopen rate within 7 days (quality), and first-contact resolution or average handle time for the desk as a whole. Add user satisfaction for the pilot group as a guardrail. Do not measure 'number of articles loaded' as success; that is output, not value. This iteration respects start where you are, progress iteratively with feedback and keep it simple and practical.