An AI Agent for Stalled Jobs That Names the Owner, Not Just the Problem

Most Stalled Jobs Have a Reason on File

When a job goes idle, the instinct is to assume it fell through a crack. So the team asks who dropped it, and someone apologizes, and the job moves for a week and stalls again. The question that finds the fix is narrower: why did this job stop.

A third of the time it is a part. A quarter of the time nobody owns the next step. Each cause has a different fix and a different owner, and only one of them is a person who forgot. A follow-up routine that treats every stall as neglect is chasing the wrong thing two-thirds of the time.

That is the difference between a reminder and an agent. A reminder says the job is old. An agent reads the job, finds the dependency or the missing owner, and puts the one next step in front of the right person.

The reframe

A stall is a question with an answer on file. The agent's job is to read the answer and hand someone the next line, not to escalate volume.

How Much Is Sitting in Your Open Jobs

Move the sliders to match your board. The model counts only the stalled jobs that are actually reachable, because a job waiting on a dated part is not a stall you can work today.

Stalled jobs you can work

19

of 210 open, after removing the ones waiting on a dated dependency

Value in play

$385,860

per year in reachable stalled work

Days recovered

job-days saved if each unblocks today

Why jobs stall

The parts order and its vendor status against the job

The most legitimate stall and the easiest to miss, because the job looks idle while a real dependency is outstanding.

The assigned employee field and the last activity touch

The job moved between people and landed nowhere. This is the stall a nudge fixes, and the one a report never catches.

The estimate and its approval status, plus the conversation history

The business is waiting on a decision it never actually asked for. The stall is an unmade ask, not an unmade decision.

The schedule and any inspection line on the job

External, dated, and checkable. It should be excluded from the stall list rather than chased.

Nothing. It is the absence of any record at all

The smallest category, and the one people assume is the whole problem. Most stalled jobs have a reason on file.

Note the bottom two bars. The permit hold should be excluded, not chased, and the genuinely dropped job is the smallest category and the loudest worry.

Illustrative estimate from the inputs above. Your own numbers will differ; this is not a measured result or client data.

One Stalled Job, Six Stages

Each stage names the system it touches, the payload it leaves behind, and the concrete failure it prevents.

Target System

ServiceTitan V2 jobs, the baseline window

Stall Confirmed

A job that has not moved is flagged only after the monitoring baseline says it should have. The agent starts from a confirmed stall rather than a suspicion, so it does not chase jobs that are simply moving at a normal pace.

The agent nags a job that is on schedule, and the team learns to ignore it.

Stage payload

A Job Leaves the List When It Moves, Not When It Is Marked Done

Every task system has a way to clear an item without doing it. The one that matters here is the resolved checkbox. A stalled job marked resolved with no motion is the same job, a week older, with a clean status.

So the agent watches the record, not the checkbox. A status change, an estimate approval, a parts update, a schedule move. If nothing moved, the job stays on the list, and the person who closed it sees it come back. That is not a punishment. It is the only definition of done that holds.

It is also what makes the cause log worth reading. If jobs keep coming back under the same cause, the cause is the defect, not the person, and the log says which.

The Agent Stages. A Person Acts.

Recovery work sits next to customer promises, so the boundary matters. The agent reads, classifies, and drafts. Anything a customer would rely on stays with a person.

Autonomous

• Confirm a stall against the baseline window
• Read the job for a dated external dependency
• Classify the cause and test whether it is reachable
• Name the owner who should act next
• Draft the unblocking nudge and stage it

Supervised

• Send the drafted nudge once a person approves it
• Propose a reassignment when the owner is unavailable
• Recommend a schedule change rather than making it
• Escalate a genuinely dropped job to a manager

Never

• Promise a customer a new date or time
• Order a part or commit to a vendor
• Change a price, a scope, or an estimate
• Close a job or mark it resolved
• Reassign a person's work without consent

Two Things We Will Tell You to Buy Elsewhere

Not a reminder app

A reminder fires a notification on a fixed interval whether or not the reason is read. It is a volume tool, and volume is the thing that makes a team stop reading the list. If you want a push notification on a timer, you already have it in every platform you own. This agent earns its place by reading the job first.

Not a substitute for owning the parts process

If a third of your stalls are parts, the fix is the parts process, not a better nudge. The agent will tell you that, clearly and every month, and no amount of automation will make an order arrive sooner. Read the cause log as a repair list.

Straight Answers

Where This Fits in Operations

This agent works the stalls that monitoring finds, inside the day the operations pillar coordinates.

Find Out Why Your Jobs Actually Stop

Send us your open board. In thirty minutes we will read every stalled job, name the cause, and show you how much is waiting on a part, on an owner, or on a customer you never asked. You keep the analysis whether or not you hire us.