Case study · Systems · Operating model design

The tools existed. The workflows didn’t.

Designing the systems behind the work.

A company that had grown quickly. Plenty of capable software, plenty of capable people, and an operational layer that hadn’t caught up with either.

What this one proves That I can change how an organization operates — across teams I don’t manage, without replacing anyone’s tools.
The four questions
What was it optimized for?

Completing individual requests.Every application did its own job correctly, and every team knew its own part of the work.

Why was that wrong?

Nobody owned the space between the tools.Requests were completed and handoffs were not. The coordination that held it together lived in people’s memory, which meant it worked exactly as long as the right person was available.

What did I redesign around?

Who decides, and what the requester sees while they wait.Not the tool. The request model came first — what gets asked, who is responsible at each handoff, what happens next — and the platform was configured to serve it.

What proves the change?

One requirement, rebuilt three times.External sharing access went from four human steps, to two, to a single decision — with the security control untouched throughout. The same move appears a decade earlier on a different stack, which is the part that makes it a method rather than an episode.

Beat 01 · The noticing

Everyone knew how to do their job. Nobody could tell you where a request went.

Every application worked. Licences were paid, systems were up, teams were competent. And the ordinary business of asking for something and receiving it ran on memory and goodwill.

  • 01Knowing where a request went was tribal knowledge, not documented.
  • 02Approvals were coordinated by hand, person to person.
  • 03Small requests generated conversations instead of resolving.
  • 04The systems worked individually. They didn’t work as one.
Beat 02 · The map

A ticketing platform was the visible project. An operating model was the actual one.

Replacing one tool with another moves the same problems into a newer interface. The work was deciding how requests should behave — what gets asked, who decides, what happens next, and what the requester sees while they wait.

What a migration would have been

Recreate the existing queues in a new tool. Import the fields. Keep the habits, change the logo.

What was designed instead

A request model built from how people actually ask for things — then the tool configured to serve it.

Owned end to end — business requirements · request types · form and field design · workflow logic · approval paths · notification design · automation · systems integration

Connected surface — ticketing platform · asset management · identity and directory · workspace and collaboration

Status — designed and running in a controlled environment; production migration in preparation, not yet live

Beat 03 · The artifact

External sharing access, in three versions.

Sharing a folder outside the company requires approval. That control is correct and was never in question. The workflow holding it up was the problem — it depended on a person being available.

What follows is the same requirement, redesigned twice. Read the left edge of each step: red is a human doing it by hand.

Version 01Manual
PersonEmployee needs external access
PersonContacts IT directly
PersonIT reviews the request
PersonIT adds the address to the group by hand
ResultAccess granted

Human steps — four · waits on someone being available

Version 02Automated
PersonEmployee needs external access
PersonIT approves and records the address
AutomaticScript reads the record
AutomaticGroup membership updated
ResultAccess granted

Human steps — two · the adding stopped being a task

Version 03 · designedIntegrated
PersonRequest submitted through a form that asks the right questions
PersonApprover decides, in the workflow
AutomaticPlatform automation fires on approval
AutomaticWorkspace integration updates access
ResultAccess granted · requester notified

Human steps — one decision · nothing is coordinated by hand

Manual action became automation. Automation became a system. The security control never changed — only the experience around it did.

Beat 03b · The same move, earlier

A CRM that people were told to use, made into one that helped them work.

Years before any of the above, in a different company and a different stack. The pattern is the point — this is not a thing he started doing recently.

  • Custom fields shaped to the actual work
  • Workflows matched to how requests really moved
  • Dashboards for the people making decisions
  • Reporting that answered standing questions
  • Data cleanup so the system could be trusted

The goal was never to make people use the tool. It was to make the tool work the way people already did.

The tools changed. The pattern didn’t.

Different companies, different stacks, more than a decade apart. The work was the same each time: find where people, process, and software are rubbing against each other, and design the thing that removes the friction.

That work is not administration. Administering software keeps it running. This decides how the work should move — and then builds it.

Employer and product names withheld · systems described by function · no confidential interface, data, or architecture shown · no metrics claimed

ArtifactOwnership and the seams between it

Enterprise systems · Contractor onboarding

Four teams did their part. The process still failed.

Every system in this chain was healthy. Every owner completed their step on time. What follows is a model of where the work actually broke — and what changed when the handoffs were given owners instead of good intentions.

Before · as inherited
People operations Contractor record created Completed on time
Handoff — manual, understood
Hiring manager Engagement approved Completed on time
Handoff — manual, understood
IT · identity Accounts and access provisioned Completed on time
No owner — nothing triggers the next step
Facilities Building access Never requested
No owner — no completion signal exists
Hiring manager Day-one readiness confirmed Never notified
Result

A contractor at a door, on day one, with everything else correct.

After · redesigned
People operations Contractor record created Start of the chain
Trigger — on record created
Hiring manager Engagement approved Approval recorded as an event
Trigger — on approval recorded
IT · identity Accounts and access provisioned Completion emits a signal
Trigger — access request raised automatically
Facilities Building access provisioned Now a step, not a request
Trigger — on all provisioning complete
Hiring manager Day-one readiness confirmed Notified without being asked
Result

No step depends on someone remembering.

The gap wasn’t the tools. It was the space between them.

Every audit of this chain came back clean, because audits examine systems and the failure was not inside a system. It lived in the boundaries — the places where one team’s work ends and nobody has agreed what causes the next thing to start.

The redesign added almost nothing. It gave each boundary a trigger and an owner, so provisioning and notification became steps the process could not skip rather than requests someone had to remember.

Ownership mapping · trigger design · workflow automation · cross-functional rollout

Representative and anonymised · no employer, product, or system named · structure reflects a real engagement, details generalised

ArtifactThe same system at three zoom levels

Enterprise systems · supporting evidence

The same system, at three levels.

One process, the workflow that carries it, and the architecture underneath. Read together they show the thing that is hard to photograph: a set of decisions about how work should move.

Level 01 · The process — external sharing, before and after

The honest difference is who has to wait, and for what.

Both versions grant the same access under the same control. What changed is where the waiting sits. Before, every pause depended on a person being free. After, there is one pause, and it is a decision somebody is supposed to make.

Red segments are waiting. Solid segments are work.

Before — the request lives in a conversation
WorkEmployee writes to IT
WaitUntil someone in IT picks it up
WorkIT asks what it’s for
WaitUntil the employee replies
WorkIT reviews and decides
WaitUntil IT is at a console
WorkAddress added by hand

Waits — three · every one of them caused by human availability, not by the decision

After — the request lives in a system
WorkForm captures the context up front
WaitUntil the approver decides
AutomaticAutomation fires on approval
AutomaticAccess updated, requester told

Waits — one · and it is the only one that ever needed a human

The fix was not speed. It was removing every wait that existed because a person happened to be busy, and keeping the one that exists because somebody has to think.

Level 02 · The workflow — a request type, designed

A request type is a set of decisions, not a form.

This is what “designing the workflow” actually consists of. Every row is a choice that could have gone another way, and each one changes what the requester experiences.

Request type

External collaboration access

DecisionSeparated from general access requests, because the approver and the risk are different.

Fields

Who · what they need reaching · why · for how long

DecisionEvery field exists to prevent one round trip that used to happen by message.

Routing

To the owner of the resource, not to a shared queue

DecisionThe person who knows the answer receives it directly; IT stops relaying.

Approval

Single decision, recorded as an event

DecisionRecorded rather than replied to, so the next step can be triggered by it.

Automation

On approval — grant access, no ticket touched by hand

DecisionThe approval is the trigger. Nobody re-enters what was already decided.

Notification

Requester told what changed, and when it expires

DecisionClosing the loop is part of the design, not an afterthought.

Lifecycle

A requested removal date, captured on the form

Designed · not yet liveThat captured date becomes the trigger that removes the access automatically.

DecisionThe end of the access is recorded at the moment it is asked for — the only moment anyone reliably knows the answer. Today it is a field a human can act on. It was designed so it can later act on itself.

Representative · field labels generalised · no live form, queue, or customer data shown

Level 03 · The architecture — four systems, one fact each

Integration is mostly deciding which system is allowed to be right.

Most environments break because two systems each believe they hold the truth about the same thing. The architecture below assigns one source of truth per fact, and the integrations only move facts — they never create a second version of one.

TicketingLayer · requests

Source of truth forWhat was asked, who approved it, and when.

Emits — approval events
Receives — asset and identity context

Identity & directoryLayer · people

Source of truth forWho someone is and what they may reach.

Emits — group and role membership
Receives — approved access changes

Asset managementLayer · things

Source of truth forWhat hardware exists and who holds it.

Emits — assignment records
Receives — joiner and leaver events

WorkspaceLayer · work

Source of truth forThe documents and groups people work in.

Emits — nothing authoritative
Receives — membership from identity

One rule holds the whole thing together: a system may only be corrected where it is the source of truth. Everywhere else it is a copy, and copies are updated by integration rather than by a person.

None of this is visible in the tools.

Every screenshot of a ticketing system looks the same. What separates one from another is the reasoning above — which requests exist, who decides, what a decision triggers, and which system is permitted to be right about what.

That reasoning is the work. The configuration is just where it ends up.

Representative and anonymised · systems described by function · no employer, vendor, interface, or data shown · no metrics claimed · integrated request workflow designed and running in a controlled environment, not yet in production