Four times the work came off the plate and stayed off

A digital agency running 15+ client accounts stopped hand-creating its work. Onboarding, website builds, campaigns, monthly and quarterly deliverables, meetings, holidays, PTO, and QA gates now generate themselves, on schedule, with the right checklist attached.

When I arrived, every recurring piece of work was created by a person remembering to create it. Client onboarding epics, website process checklists, monthly advertising and reporting tickets, weekly syncs, holiday coverage, QA hand-offs: all manual, all inconsistent, all landing late in busy weeks.

I built it in three layers.

Jira Automation. Templated epics that fan out into full checklists (a website build-out generates its five-phase checklist; a new contract generates the onboarding epic), scheduled rules that create each month's and quarter's deliverables from the client roster, sprints created nine weeks out, and Slack alerts for overdue tickets, stale QA, and mentions.

ScriptRunner. A daily 8 AM stand-up DM to each person with their blocked, in-progress, overdue, and due-today tickets in a ready-to-post format, and validation that a ticket has a sprint, an assignee, and a due date before it can slip through.

Google Apps Script. Intake forms for PTO and client support that create Jira tickets and Slack notifications, calendar events and meeting transcripts synced into Jira, and ticket-health monitoring that flags or reschedules stale work. Two Make.com workflows were retired along the way. Project charters, timelines, and SOPs ran on the same rails, with timelines in Structure PPM over Tempo.

What the 133 rules actually containRunning a client month by the ticketThe 63-server auditStand-up summaries and status reporting

Leadership went from not knowing whether a given month was profitable to a monthly financial report, a 12-month projection built line by line from the client roster, and a live view of where the overhead was going.

The books were messy enough that leadership could not say with confidence whether the company was making money in a given month. Roughly $150,000 a month was flowing through QuickBooks and Stripe, but categorisation, reconciliation, and reporting had not kept up with it.

First, tracking: I brought the transactions current, categorised them, and started producing a monthly report leadership could actually read. Then, projection: using prior-year expenses as the baseline and current expenses, overheads, and the client roster (who was renewing, who was near expiration), I built a 12-month projection with a per-month transaction view underneath it, so every projected line traced back to a client, a contract, or a recurring cost. Alongside it, a subscription cost tracker synced daily from QuickBooks against a monthly spend target, because tooling was where the leaks were.

Recreated extracts of the projection, ledger, and trackerThe capability map, which flags single points of failureSubscription and renewal tracking

Two cold-start deliverables in one weekend: a messy 67-response consumer survey turned into a formula-driven summary and a nine-slide report, and a September 30 beverage launch mapped backward into a 16-task, dependency-coded timeline with the critical path made visible.

The brief was a consulting firm's new-hire assessment, and it looked like real client work on purpose: one raw survey export and one page of launch constraints, nothing else. I treated it as a project, not a test.

The messy part first. The export arrived with unnamed columns, headers buried in row two, multi-select answers scattered across nine columns, and free-text complaints ranging from "too dense" to a paragraph about sweeteners. I rebuilt it as a clean Data sheet, then added one derived column that coded every open-text dislike into themes (Price/cost, Sweetness, Texture, Availability, Melts in shipping). That single coding step is what turned 67 opinions into a rankable barrier list.

The summary that computes itself. The Summary tab holds no typed numbers: every count is a COUNTIF against the Data sheet and every percentage divides off those counts, so correcting one response upstream re-states the whole analysis. Findings held up: 88% had tried the brand, price was the top barrier by a wide margin (19 of 45 coded complaints), natural ingredients and low sugar ranked as the attributes that matter, and vegan ranked dead last.

Click any slide to preview it full size; the links below download the real workbooks and deck.

Then the launch plan, built backward. The second brief gave a target date, lead times, and phase rules, and asked one question: how would you manage this? I anchored on September 30 and scheduled backward, which surfaced the real headline: the required start was June 15, about fifteen weeks out, with almost no slack. The plan orders the long-lead cans and cartons first and in parallel, overlaps artwork approval with validation, runs the 2,400-can trial into the fixed four-week approval gate, and tucks full-run procurement inside that gate so the 400,000-can run lands the week before launch. Every task carries an owner function, a T-coded dependency, and a critical-path flag; the Gantt bars are painted by those codes.