An agent picks up a ticket. Before they can do anything useful they read the whole thread, guess at a category and a priority, hunt for the last time somebody fixed something similar, and judge whether the customer is calm or about to escalate. Do that a hundred times a week and it is the quietest cost on a service desk.
Most AI coverage in service management is about the chat bot on the portal. Halo AI is a native module that also puts machine assistance into the fulfiller's ticket view, configured at Configuration > AI.
For the agent it provides AI Insights, covering sentiment, tonality, suggested priority and category, resolution type and thread summarisation. Alongside it sit AI Ticket Matching built on vector embeddings, AI Suggestions that propose field values from matched historical tickets, article suggestions and creation, AI search, response improvement and thank you detection. The 2026 second release, 2.236, added ticket clustering. Setting names and prerequisites are in the Halo documentation.
Triage and routing in Halo were rule based, so anything the rules did not anticipate landed in a general queue. Knowledge articles were rarely written, because that is the first thing a busy agent drops.
The ticket now arrives partly read, and the human still decides. Suggestions surface in the ticket and in the problem and resolution finder, each with an Apply suggestion button beside them.
The agent opens a ticket to a summary of the thread, a read on sentiment and tonality, matched prior tickets and a suggested article, plus one click suggestions for category, priority, urgency, impact and assignee. Response improvement cuts rewriting on replies, and thank you detection stops tickets that are resolved in all but name sitting open.
AI Suggestions rules can propose the assigned agent and the category, which improves first time routing and means fewer reassignments. Ticket clustering groups similar tickets, generates a cluster description and resolution, and exposes a Cluster Matched Tickets tab, which is where knowledge gaps become visible without a manual trend analysis.
AI report analysis can be pulled into scheduled and composite PDF reports through the $REPORTAIANALYSIS and $REPORTAIANALYSISDATE variables. Separately, SLA breach criteria are now visible directly within ticket lists, so risk is easier to spot without a dedicated report. Clients who will not permit AI processing can be excluded individually, at customer profile > settings > miscellaneous > Exclude from AI functionality.
Switching it on is about half a day: pick the connection type, enable Create Embedding Scores for Tickets, set matching to Built in functionality, choose a vector database, nominate an AI Embedding Field and enable embeddings and insights on the ticket types you want. Getting real value out of it takes two to four weeks, spent on the vector database, a considered embedding field and insights context, historical indexing left to run before you trust a suggestion, and tuning of the minimum vector match score. Nothing needs migrating, but suggestion quality depends on how clean your historical categories, priorities and resolutions are.
Halo's HaloITSM pricing page states there are no tiers, editions or add ons, and that all capabilities including AI are included as standard for every agent, at £66 per agent per month billed annually in UK pricing. The AI guide states no edition requirement for individual AI features. The real cost lands elsewhere: your model consumption and your Azure AI Search instance are billed separately by the provider, and Halo notes that a free OpenAI account is not sufficient for production. No figure is published at small service desk volume, so budget it as a variable.
Halo also describes a Fulfiller Operations Agent inside the ticket, and process agents covering Dispatch, 1st Line Support, Major Incident, Knowledge and Change Management. No configuration guide, prerequisites or availability date is published for those, so treat them as direction of travel. Reviewers report Halo AI demonstrating well and then under delivering in production, pointing at documentation gaps more often than at accuracy. The six month risk is configuration drift and untuned thresholds.
Hikon is a Halo partner working with service desks across the UK and Israel, and here the commercial side is the easy part. What matters is which connection type suits your data handling position, how the embedding field and insights context are designed, and where the minimum match score sits. We tune those against your ticket history, not a demo data set, and we do not switch suggestions on for agents until indexing has produced matches worth trusting.
Halo's configuration guide is the reference for setting names and prerequisites: the Halo documentation
Before switching suggestions on for agents, index your ticket history and check your connection type, embedding field and minimum match score against the guide above.
Talk it through with the people who wrote it — a free 20-minute call, no pitch.
Let's talk →Book a free 20-minute call about your platform and your plans — no pitch, no obligation.
Start the conversation