01 the business
Residential property manager
1,800 rental homes, 14 support staff
Tenant emails, portal requests and chats land in one queue. Someone reads each one to decide who should take it. An urgent leak can sit behind routine questions, and tags are so inconsistent that nobody can say what tenants contact the team about most.
02 before
The problem
Tickets land in one shared queue and someone has to read each one to decide who should take it. Urgent problems, like a broken checkout or a burst pipe at a rental, can sit unread behind password resets. Tags are applied inconsistently, so reporting on what customers actually contact you about is guesswork.
- Urgent issues wait behind routine ones.
- One coordinator spends half the day sorting the queue.
- Reports on contact reasons are unreliable.
03 the build
What the AI does
- Step 1: Reads each incoming ticket from email, web form or chat as soon as it arrives in Zendesk, Freshdesk or Jira Service Management.
- Step 2: Labels the topic, product or property, language and urgency using your own categories.
- Step 3: Spots signs of a serious problem, such as safety issues, outages or an angry repeat contact, and raises the priority.
- Step 4: Routes the ticket to the right team or person based on skills, workload and working hours.
- Step 5: Posts urgent tickets to a Slack or Microsoft Teams channel so someone picks them up straight away.
paused: waiting for a person to approve
How it was put together
- Labelled each ticket with topic, property, language and urgency using the team's own categories.
- Raised priority for safety words, outages and angry repeat contacts.
- Routed by skill, workload and working hours.
- Posted urgent tickets to a Teams channel for immediate pickup.
04 human in the loop
Where a person stays in charge
Your support lead sets the categories and routing rules and can correct any label in one click. Corrections are reviewed weekly and used to tune how tickets are sorted.
- The support lead owns the categories and routing rules.
- Any label can be corrected in one click, and corrections are reviewed weekly.
- Safety-related tickets are never deprioritised automatically.
05 rollout
From kickoff to live in 3 weeks
- Week 1Agree categories and urgency rules; label a sample of past tickets by hand.
- Week 2Run in shadow mode and compare with the coordinator's choices.
- Week 3Switch on routing and the urgent channel.
06 after
What changes
Tickets reach the right person first time, urgent ones get seen quickly, and nobody spends the morning sorting the queue. Tags become consistent enough to show which issues come up most.
Assume 3,000 tickets a month and 1.5 minutes for a person to read and route each one.
3,000 × 1.5 minutes = 4,500 minutes, or 75 hours a month of sorting, before counting the cost of urgent tickets waiting.
What to measure
- Time to first touch for urgent tickets.
- Share of tickets re-routed after the first assignment.
- Label corrections per week.
- Coordinator time spent sorting.
07 evidence
What the research says
Across 5,179 customer support agents, an AI assistant raised issues resolved per hour by 14% on average and by 34% for novice and lower-skilled agents.
NBER, Generative AI at Work (Brynjolfsson, Li and Raymond) (opens in a new tab) · 2023
About 75% of generative AI's potential value sits in four areas: customer operations, marketing and sales, software engineering, and R&D.
08 lessons
Mistakes to avoid
- Copying a vendor's category list. Use the categories your team actually works with.
- Ignoring working hours. Routing to someone who is off duty is not routing.
- Never reviewing corrections. That is how the sorting improves.
09 questions
Common questions
- Will it work in more than one language?
- Yes. It labels the language and can route to people who speak it.
- Which help desks does it support?
- Zendesk, Freshdesk, Jira Service Management and HubSpot are common, plus most tools with an API.