Skip to content
QRKTech

Case study · Managed IT · Project management

An AI project manager that gives the team back 30+ hours a week

Hazel turns a signed scope of work into a staffed, scheduled project in the PSA, then chases it to done: technician work lists, follow-ups, customer updates, and a direct message to management the moment a deadline is at risk.

Client
A Minnesota managed IT services provider, about 35 people
Coworker
Hazel, Project Manager
Reports to
Director of Projects
QRK TechDigital coworker
H

Hazel

Project Manager

Department
Projects
Reports to
Director of Projects
Shift
Always on · work lists Mon/Wed/Fri 08:30 · overdue scans twice daily

Scoped access

  • PSA projects, tasks & milestonesRead / write, role-scoped
  • Technician calendars & workloadRead
  • Microsoft TeamsMessage internal staff
  • Project & customer recordsOwn workspace

Memory

Project and customer dossiers, a team roster built from delivery history, decision log

QRK-PM-0001

of project-management work handled every week
30+ hrsof project-management work handled every weekReported by the projects team
to draft a technician-ready project plan
4h → 15mto draft a technician-ready project planEstimated; a PM reviews every plan
to review a ticket’s status and next step
45m → 30sto review a ticket’s status and next stepEstimated, per ticket
projects planned in the first three months
30+projects planned in the first three months
§ 01The job before

What the work looked like.

Context

The provider runs dozens of concurrent customer projects: server migrations, Microsoft 365 moves, hardware refreshes, office relocations. All of it is tracked in HaloPSA.

One senior project manager coordinated the work. That meant turning each scope into a plan, staffing it, chasing technicians, answering customers, and validating closeouts.

The problem

A first-draft plan took about four hours per project: reading the scope, building milestones and child tasks, setting budgets and dates, and finding the right people.

Follow-up was worse. Checking one ticket’s status and next step took about 45 minutes, across roughly 20 tickets a week. Stale tasks and slipping dates often surfaced at the weekly meeting, when they were already late.

Closeouts meant manually checking every child task, deliverable, and hour against the scope. The job was full of judgment, but most of the hours went to retrieval, retyping, and reminders.

§ 02The coworker

What Hazel does.

  1. 01

    Scope to plan

    Reads the statement of work and the parent project, then locks scope and approved hours with the PM before anything else. It drafts milestones and child tasks with goals, deliverables, budgets, and target dates. If the task hours exceed what was sold, it stops and says so.

  2. 02

    Staffing that respects reality

    Matches each task to qualified technicians using a roster built from real delivery history. Before proposing dates, it checks working hours, leave, and calendars. It won’t silently swap in someone unqualified because the first choice is busy.

  3. 03

    Build once, verify everything

    After approval, it creates each record once and reads every one back: linkage, milestone, owner, dates, and budget. It also catches PSA automations that quietly overwrite owners or dates, and corrects them before technicians are notified.

  4. 04

    Chase work to done

    Every Monday, Wednesday, and Friday, each technician gets a Teams list of their open project tasks, with past-due work first. Twice-daily scans catch new overdue work. Follow-up notes are collaborative, ask focused questions, and are read back to confirm they posted.

  5. 05

    Keep customers informed

    Answers “where are we?” questions with the current, verified schedule, within the communication authority the team has granted.

  6. 06

    Escalate early, with evidence

    Messages senior management directly when work is blocked, a customer is unhappy, or a committed date is at risk. The message covers the evidence, the impact, what has been tried, and the decision needed.

  7. 07

    Close out properly

    Validates every child task, deliverable, and budget against the scope. It assigns residual work to owners and drafts an auditable handover. A status change on its own doesn’t count as a closure.

§ 03Walkthrough

One project, start to finish

A typical run, step by step. Customer details are changed.
  1. Mon 09:04

    A signed scope arrives in Teams: office relocation, 42 seats, 86 approved hours. Hazel reads the parent project and checks its records for a similar past move.

  2. Mon 09:20

    Draft plan: 11 tasks across Preparation, Execution, and Clean-up, totaling 84 hours, two under budget. It flags that the cutover depends on a new internet circuit that isn’t installed yet.

  3. Mon 09:31

    A staffing proposal with reasons for each owner. One preferred technician is on leave during the cabling window, so Hazel names a qualified alternative and explains the tradeoff.

  4. Mon 10:02

    The PM approves the plan card in Teams. Hazel creates the tasks and reads each one back. A PSA automation had reset one target date, so Hazel corrects it before anyone is notified.

  5. Wed 08:30

    The new tasks appear in each technician’s work list, in due-date order.

  6. Day 9

    The circuit install slips. Hazel messages the Director of Projects with the new risk to Friday’s cutover, two options, and the decision needed by Wednesday.

The PM’s time went into two decisions instead of a morning of data entry.

§ 04Under the hood

How the work flows.

The same platform runs every QRK coworker. For how each layer works, take the platform tour.
  1. 01

    Trigger

    A scope in Teams, a schedule (Mon/Wed/Fri 08:30), or a PSA event

  2. 02

    Context

    Project and customer dossiers, team roster, shared project memory

  3. 03

    Reason

    Scope, dependencies, capacity, and risk, leading to a recommended plan

  4. 04

    Approve

    The PM approves the exact plan version in Teams

  5. 05

    Act

    Role-scoped PSA tools: tasks, milestones, notes, appointments

  6. 06

    Verify

    Every record read back; automation side effects reconciled

  7. 07

    Report

    Work lists, follow-ups, escalations, and an updated project record

§ 05Guardrails

What keeps it in its lane.

These aren’t suggestions in a prompt. They’re built into the role, the tools, and the procedure.
  • Creating tasks or changing assignments, dates, budgets, or status needs an approved, concrete plan. Having the capability is not the same as having permission.
  • Every write is checked with a dry run, performed once, then independently read back. A tool returning no error is not proof of success.
  • Standing permission covers only routine private follow-ups and internal escalations to verified staff. Customer messages and new recurring automations need their own authorization.
  • If a write’s outcome is uncertain, it stops and reconciles instead of retrying and risking duplicates.
  • Cross-customer information, internal costs, and private notes never leave the internal audience.
  • It signs every note as an AI project assistant. It never impersonates a person.
§ 06Results

What changed, and what we learned.

Outcomes

  • The PM’s time goes to decisions and customers instead of data entry and reminders.
  • Projects start faster, with a technician-ready plan the same day the scope arrives.
  • Fewer duplicate and orphaned tasks, and budgets reconciled before work begins.
  • Stale work surfaces within days instead of at the weekly meeting.
  • Management hears about at-risk deadlines while there is still time to act.
  • Closeouts are backed by evidence and can be audited.

Lessons

  • Lock scope and hours before planning. Most project pain starts with tasks that add up to more than what was sold.
  • Read-back catches what logs miss. PSA automations quietly overwrite owners and dates, and checking every record is how they get found.
  • An early overdue scan reported “nothing overdue” every run because it read only the first page of results. The rewrite checks every technician’s queue, labels truncated results, and found 14 overdue tasks the old version had missed.
  • Adoption came from the team shaping the coworker. Managers extend its procedures directly, so it works the way they work.

Next step

Want a project manager like Hazel?

We’ll sit down with the people who do it, map the steps, and put a number on the time. If a digital coworker is the right tool, you’ll know what it would return before you commit. If it isn’t, we’ll say so.