ServiceNow Now Assist generative AI deployed across a 500-person retail IT help desk — reducing L1 ticket volume by 42% and cutting average resolution time in half in 6 weeks.
The client, a mid-market retail chain that had grown from 18 to 46 stores over three years, had scaled its store operations without proportionally scaling IT support. The IT help desk — a team of 8 — was fielding 340 tickets per week from 500 employees across store, warehouse, and head office. SLA compliance had dropped from 94% to 71% over 18 months. Agents were burning out, tickets were backing up, and store managers were calling the IT manager directly to bypass the queue.
An analysis of the 340 weekly tickets revealed that 58% were L1 issues: password resets, printer connectivity, POS software queries, and VPN access requests. These were resolvable from the knowledge base — if employees could find and understand the right articles. Most couldn't, so they raised a ticket instead.
ServiceNow was already in place as the ITSM platform. Now Assist was the logical tool. The question was whether it could be deployed meaningfully in 6 weeks, before the peak retail season arrived.
Now Assist's effectiveness is directly tied to the quality of the knowledge base it draws from. The client's knowledge base had 186 articles — but 94 of them hadn't been updated in over two years. Printer models referenced no longer existed. VPN instructions pointed to a deprecated client. Password reset procedures described a process that had changed after an identity provider migration.
We ran a two-week knowledge base sprint: auditing every article against current systems, rewriting 64 articles from scratch, archiving 38 obsolete ones, and creating 22 new articles for the most common issues that had no documentation. This was the unsexy work that made everything else possible.
Now Assist was configured for three use cases: the employee self-service portal (AI-powered search and guided resolution), agent assist (AI-drafted responses and resolution suggestions for agents), and ticket summarisation (automatic summaries of long ticket threads for incoming agents picking up mid-conversation).
Prompt engineering was the highest-effort part of the configuration. The retail environment has specific language: "POS" means point of sale, not point of service; "EOD" means end-of-day cash reconciliation, not end of day generally. We built a domain-specific prompt layer that contextualised queries before they reached the knowledge base, dramatically improving relevance of AI responses for store-specific issues.
The existing self-service portal was a list of categories with links to forms. It required users to know what type of issue they had before they could raise a ticket — which often they didn't. The redesigned portal put the AI search front and centre: employees describe their issue in plain language, Now Assist surfaces relevant knowledge articles and guided resolution steps, and only routes to a human agent if the AI cannot resolve it.
The portal was tested with 30 store employees in a controlled session before go-live. Usability feedback led to three interface changes — primarily around how Now Assist communicates uncertainty ("I'm not sure — let me connect you with an agent" versus confusing silence).
The 8-person IT team received two half-day training sessions focused on working with AI assistance rather than around it. The key cultural shift: agents were encouraged to review and refine AI-drafted responses rather than ignoring them and writing from scratch. Early resistance ("the AI doesn't know our systems") dissolved within two weeks as agents experienced the time saving firsthand.
In the first 30 days post-go-live, L1 ticket volume dropped by 42%. The 8-person team — unchanged in headcount — was processing the same total enquiry volume with significantly more time for complex issues. SLA compliance recovered from 71% to 96% within six weeks. The peak retail season arrived on schedule; the help desk handled it without additional headcount for the first time in three years.
The two weeks spent on knowledge base remediation before touching Now Assist configuration were the highest-leverage work of the project. Every hour invested in accurate, current, well-structured knowledge articles returned multiple hours of AI deflection downstream. Teams that skip this step and deploy Now Assist on a stale knowledge base get stale AI responses — and frustrated users.
The agent assist feature — AI drafting the first response for agents to review — had the most surprising adoption curve. Initially resisted, it became the team's favourite feature within two weeks. The time saving per ticket was modest (90 seconds on average), but multiplied across 200 agent-handled tickets per week, it freed up 5 hours of agent time per week — the equivalent of adding half an FTE at no cost.
"Black Friday used to be a nightmare for IT. This year, by the time I arrived at 7am, the AI had already resolved 23 self-service requests overnight. Nobody was waiting."
— IT Manager, Retail Client