The function
We run your AI operating model
We sit in the model. We hold living decision rights, run the cadence, and keep who-builds / who-approves / who-kills honest. We stay. This is the model itself, not a consulting sprint to design one.
- the model
- not a RACI slide
- rights
- who builds, approves, kills
- cadence
- the week the model runs
- we stay
- the model continues
The model, not a RACI.
An AI operating model is living decision rights and a cadence that uses them. Work is already starting in five places at once, and nobody can say who may start it, fund it, or stop it. AI strategy and advisory decides where to play. AI governance and compliance writes the boards and the binders. The AI Center of Excellence is the intake door. The Chief AI Officer is the officer. The AI Transformation Office funds and kills workstreams. This page is the model those seats sit inside: who builds, who approves, who kills.
Most companies searching for an AI operating model already know the RACI they commissioned is dead. They also know they will not hire a Head of AI at public 2026 mid-market packages just to keep a rights chart honest. Someone still has to name the builder, approve the spend, and kill the thing that should not be alive.
arosplatforms is a Toronto-based firm that serves companies worldwide. When we take this engagement, we run your AI operating model. We sit in the forums where rights get used. We have decision rights. We do not suggest a target operating model and leave. We do not audit you, find gaps, and sell you the build. We are the function that holds the model, and we stay.
What the model looks like in motion.
Scroll through it, the screens move with you.
Who builds. Who approves. Who kills.
A short chart the company will actually use this week. If two functions both think they approve, the model says which one wins.
Platform team
Shadow copilot
Spend over cap
Pilot, no owner
A short list someone will defend
What the model can decide without a committee. Written by the people who will still be in the room when someone tries to skip it.
Build rights sit with the named owner, not the loudest function.
any team may start a pilot
Kill rights are used on the date the model set.[src]
A cadence the model can run on
Rights used, exceptions written, one kill or one stay. The model is in it every week, not at the quarterly redesign.
Platform team
Shadow copilot
Spend over cap
Pilot, no owner
Everything in the engagement.
Living decision rights
A short, enforced chart: who may start work, who may spend, who may stop it. A right nobody uses is a RACI. We use it.
Cadence that happens
A weekly forum where the model is run, not presented. If a week passes and no right was used, the model is theater.
Who builds
The second function does not get a second stack because it asked louder. The model names the builder and refuses the duplicate start.
Who approves
Spend, vendors, and go-lives have a named approver and a threshold. Approval is a right, not a meeting that hopes for consensus.
Who kills
Every piece of work that is not yet a decided capability has a date and a person who may end it. Killing it is the model doing its job.
The model stays live
A written mandate for the model itself: what it can change, what goes to the officer, what is an exception. Without this, “operating model” is a slide with a nicer name.
A clear path from kickoff to value.
Take the model
We agree the living rights: who may build, who may approve, who may kill, and on what cadence those calls get made. You get a model already in the rooms, not a RACI slide and a waiting period.
First 30 days, rights are live
The chart is short and used. The week has a forum where a build, an approval, or a kill actually happens. We start making the calls that were sitting in a target-operating-model deck nobody opened.
First 90 days, the cadence is running
Rights, exceptions, and overrides have a weekly rhythm. Conflicts between functions get resolved in the model, not in Slack. Leadership can see who decided, who built, and what died.
First 180 days, the model holds
The model is not a project that ends. We stay. The rights and the cadence compound because someone is still there to hold them when the org chart moves.
Results you can measure.
Rights
who builds, who approves, who kills is written and used
Cadence
the model runs in the rooms, not in a redesign
The model
the rights are still live after the first quarter
Built around your starting point.
COO, CIO, operating partners
The company with AI everywhere and no rights
Work is already starting. Vendors are already in. Nobody can say who was allowed to build, approve, or kill. You need the model this month, not a twelve-month hire plan.
Mid-market and growth-stage
The company that will not hire a Head of AI this year
A Head of AI plus a bench is a real payroll event. Public 2026 guides put that lead in the same band as a senior AI executive. Running the model is how you get the function without pretending you can fill the job.
If you want a SOW
Not for teams shopping a build
If the real ask is “design the operating model, find the gaps, and build those gaps,” this is the wrong page. We run the model. We do not sell the implementation from this seat.
What an AI operating model actually holds
The phrase exists because the work outgrew every other room. A CAIO can own the agenda and still not name who may start. A CoE can own the door and still not name who builds. A transformation office can fund a portfolio and still not run the weekly cadence. An AI operating model exists so those rights have a home. That is a function, not a project. Projects end. Models remain.
Decision rights, as a standing chart
Decision rights in this seat are not a RACI on a share drive. They are the live answer to who may start AI work, who may spend on it, and who may stop it. When a unit wants to build or a team has already shipped, the model names the builder, the approver, and the killer. If the right is not used, the model is dead and we say so.
Cadence, as something that happens every week
Most companies already have an AI operating model. They will not call it that. A quarterly steering pack, a working group that skips, a RACI last updated when the last consultant left. We treat cadence as a week the model will still be in the room to run. A project produces a redesign. A function runs a week — including the no when two functions both claim the same right.
Who builds
The second team always wants to build. The vendor always wants to be the stack. Naming the builder is a decision, not a suggestion. If nobody can name who builds, you do not have an operating model. You have a collection of first times.
Who approves
Approval without a threshold is a meeting. The model puts a name and a number on spend, vendors, and go-lives. If no one has ever refused an approval, you do not have a model. You have a rubber stamp. Deeper AI governance and compliance work exists as its own service. On this page, approval is a right that gets used.
Who kills
A pilot without a named killer is a permanent employee with worse benefits. The model puts a person and a date on anything that is not yet a decided capability. The AI Transformation Office kills workstreams in a portfolio. The CoE holds kill-criteria on items in its door. This page is the right itself: who is allowed to end the thing, and whether that right was used on time.
The other chairs are different chairs
A Chief AI Officer owns the agenda. A CoE owns the door. A transformation office owns the funded portfolio. Strategy owns where to play. Governance owns the written controls. Training and enablement owns fluency. We sell those on different pages. Do not buy this page if what you wanted was one of those seats. Do not buy those pages if what you wanted was the living rights.
Why companies look for an operating model and never run it
The search terms are honest: AI operating model, target operating model for AI, operating model as a service. People type them because they already flinched at the full-time hire and already watched a RACI die. The need does not go away because the hire is hard. Big-firm articles describe a model you should have, then sell you a program to redesign it.
The full-time owner of the model is a real hire
A model that can assign build, refuse approval, and kill work is not a rebadged transformation lead plus a slide. Public 2026 US guides put a Head of AI / VP AI base roughly between $300,000 and $525,000 (Christian & Timbers). Director-level AI engineering and platform roles often sit $275,000 to $475,000. Those are not arosplatforms prices. They are the public market for the seat you are comparing us to.
The company is not ready for that hire — and still needs the rights
This is the usual case. AI is in the building. Starts are already happening. There is not a loaded line for a Head of AI. Published 2026 fractional executive retainers commonly run $5,000 to $30,000 a month — about $60,000 to $360,000 a year — depending on days in the seat (Olofsson). That is not a workshop. It is a slice of the week the model needs.
“Rewire the operating model” is the wrong instrument
Deloitte Insights 2026 published “Rewiring the enterprise operating model for AI scale.” It is the kind of article that tells you to redesign the model. It is not a firm that will sit in it. A redesign assumes you have people who can inhabit the chart after the team leaves. Most companies on this page need the rights live this month. We sell the narrower thing: a function that will hold the rights and stay. Not a target operating model.
Leaving the model empty has a cost you are already paying
No operating model does not mean no AI. It means AI without rights: five functions that all think they approve, kills that never land, a cadence that is a quarterly readout. We will not invent a dollar figure. The comparison is not “hire us or save the money.” It is “pay for the model, or keep paying for the absence of one.”
What “we run the model” means, week to week
The sentence is literal. We run your AI operating model. We sit in the rights. If a week goes by and we were not in the rooms where build, approval, and kill got decided, we were not doing the job — no matter how many decks we produced.
We are in the rooms where rights get used
The job happens on your calendar. We join the forum where a start is allowed or sent back, and the date a pilot is supposed to die. We do not invent a steering committee so we have somewhere to present. We arrive having read the chart, and we use the meeting to decide.
We have decision rights
A model without rights is consulting with a better noun. Before we start, we write the mandate: what the model can assign, approve, refuse, or kill, and what goes upstairs. If you want a smart guest, do not buy this. If you want the model, the model has to be able to say no, and have it stick.
We are accountable for the model
If the same start happened twice, a right was skipped in public, or a pilot that should have died is still burning cloud, that is our problem. Functions do not get to say “we recommended.” They decide, and they live with it. Accountability is the weekly record — decisions, owners, dates.
What a week actually contains
A pass through the rights that were used or skipped. At least one build assignment or refusal. At least one approval or send-up. At least one kill-date conversation. A short written trail. And the unscheduled thing, which is why you wanted a function instead of a slide. Some companies need a day a week. Some need more. We set that in the open.
Not a consultant. Not an agency. Not an audit shop.
If you have bought AI help before, you have met the other three. They are legitimate businesses. They are not this job. Each will wear an operating-model badge for the length of a statement of work. A redesign is still not the function sitting in the model.
A consultant writes the RACI, then leaves
A consultant is paid to improve your thinking and get out of the way. You have a target operating model. You are still the person who has to sit in the rights forum on Monday. If your last engagement produced slides and a vacuum, you bought the redesign. You did not buy the model.
An agency builds what you brief
An agency is paid to make a thing. arosplatforms does that work on other pages. This page is not a door into it. If the first serious question you are asked is “what should we build?”, you are talking to a builder. We will ask who is allowed to start, and what happens if we say no. We will not convert this seat into a statement of work.
An audit shop finds gaps and sells the fix
Assess the client. Publish the gaps. Convert the gaps into a development statement of work. We will not do that from this seat. Deciding a start should be refused is a decision, not a lead. Suggest, audit, find gaps, build those gaps — that sequence is the thing this page refuses.
A job posting is a hire, not a model
“Head of AI operating model” often means “we would like an employee, then a team.” That is a clean need. It is not what we sell. We are not a recruiter. We take the model as the model. If the company later grows into a full-time owner, saying so is part of the job.
30 / 90 / 180 days in the model
Companies ask for a 30 / 90 / 180 because they want to know what changes. Here is what changes when you bought the model, not a diagnostic. There is no phase called “discover the gaps,” and no phase called “build the gaps.” There are rights, and they get more real.
The first 30 days: the rights are already the model
Week one is the mandate: decision rights, the cadence, what we can refuse, the name on build / approve / kill. We take the keys that exist the way a finance team takes a close. By month’s end the chart is short and we have made at least one real refusal or kill. We will not run an interview tour or produce a backlog for our engineers.
The first 90 days: the cadence is being run
There is a standing forum. The chart is shorter, because some starts were refused and some things died. Approvals are a habit. At 90 days we can tell leadership who built, who approved, what we killed, and what we will decide next. If we cannot, we have been busy, not in the model.
The first 180 days: the model holds
The operating model should now be boring in the way a good finance close is boring: cadence, rights, a chart, fewer surprises. New starts go through a named builder. If the company has grown into a full-time hire, we will say so. If it has not, we stay. One hundred and eighty days is not the end of a project.
Rights, cadence, who builds, who approves, who kills
These are the parts of the job people try to peel off and turn into projects. They are not projects. They are how the model does the work, and how this page stays distinct from the officer, the door, the program office, strategy, and governance.
Rights as a regime, not a slide
The model can tell you what was allowed to start this month: the named builder, the approver, the killer, the reason it was refused. Also the harder pieces: stopping work that started without a right. Implementing a workflow tool is engineering. Holding the rights is the function. Engineering later is another page. The seat never becomes that page.
Cadence as a standing judgment
We run cadence as a function: the week is short, the exceptions are written, the skip is a decision we will still be here to own. A company with a live model should be harder to start AI work in, not easier. That is the point. Policy-writing at regulator depth is AI governance and compliance. On this page, the week is what the model will actually run.
Build, approve, and kill as three different verbs
The model names the three verbs and refuses to let one person hold all of them without writing that down. The vendor does not approve its own renewal. The team that built the pilot is not the only one who can kill it. Build, when it happens, is another page. The decision that it should happen is this one.
The model and the other pages, held apart
Running rights and teaching people are not the same job. The function that refuses a start points at training and enablement when fluency is the gap, and at the AI Center of Excellence when the gap is the door. We stay for the rights. That is the product. If you wanted only a poster, buy a workshop.
Who this is for, and who should leave the page
We would rather lose a conversation than take a model we cannot hold. If you are not in the fit, the honest next step is a different page, or a different firm.
Who this is for
A COO, CIO, or operating partner who wants an AI operating model and will give it rights on who builds, who approves, and who kills. A company that already has AI in the building and no chart that owns it. A mid-market company that knows the public Head of AI band is not this year’s hire, and is done pretending a RACI is the same thing.
It is also for leadership that wants the truth more than the theater. If you want a model that will refuse a start and kill a pilot, we can do that. If you want an operating model for the annual report, we cannot.
Who this is not for
Anyone shopping for a custom-build, an “AI OS,” or an agent-build partner and using this language to start the funnel. Anyone who wants an audit, a gap list, and a proposal to close the gaps. Anyone who wants a RACI workshop they can file. Anyone who will not grant decision rights. Anyone looking for a hire or a contractor to convert.
It is also not for a company that already has a strong model and wants extra hands. That is staff augmentation. We are the function that runs the model. If the model is filled, we should not be in it. And it is not the Chief AI Officer, the AI Center of Excellence, or the AI Transformation Office.
Cost of the model, cost of no model
We will not invent an arosplatforms price, package, or retainer on this page. Pricing the seat is a conversation. What we can put in writing is the public 2026 market, and the cost of pretending you do not need the comparison.
Read the ranges as a market, not a quote
Those numbers move with company size and whether AI is the product or a function. The honest comparison is the person who would hold the rights, not a workshop. When we talk, we talk about days in the model, not a SKU. If you can hire a full-time owner, we will say hire one. If you wanted a build team, we will send you away from this page.
The cost of no model is not theoretical
You are already paying it: duplicate starts, unused licenses, approvals nobody owns, kills that never land. We will not invent a dollar figure. Without the model, the company chooses by inertia and by whoever booked the last vendor meeting.
What we will not do with money on this page
We will not show three packages or a bundle of development hours. Those patterns turn the model into a salesperson. If engineering happens later, it is another page, commissioned by the model. We will not convert the seat into a statement of work.
| What you are comparing | Public 2026 range | What it is |
|---|---|---|
| Head of AI / VP AI, US base | $300k – $525k | Christian & Timbers 2026 corporate AI study |
| Director, AI engineering / platform | $275k – $475k | Same study, director band, before the bench |
| Fractional executive retainers | $5k – $30k / month | Common published monthly range (Olofsson and peers) |
| Fractional, annualized | $60k – $360k / year | Same retainers, expressed per year |
| “Rewire the model” programs | Sold as projects | Articles and redesigns — not a published substitute for the seat |
Ranges compiled from public 2026 compensation and fractional-retainer writeups (Christian & Timbers, Olofsson, and other open market guides). They are not arosplatforms fees or packages. Deloitte Insights 2026 “Rewiring the enterprise operating model for AI scale” is the kind of article that tells you to redesign the model. It is not a firm that will sit in it.
Book the conversation about the model
If you need an AI operating model and you are not going to staff one this year, the honest move is to put the function in the building anyway. arosplatforms will run that model. We will hold the rights, run the cadence, name who builds / who approves / who kills, and stay.
If you need a plan, a binder, a door, an officer, a program office, or a system, say that. Those are other pages. If you need the model, book a demo or contact us. That is the only ask on this page.
Tools we work with
Frequently asked.
An AI operating model is living decision rights and the cadence that uses them: who may build, who may approve, who may kill. It is not a RACI. We sit in it, we have decision rights, and we stay. Strategy, governance, the CoE, the CAIO, and the transformation office are other seats. This page is the model those seats sit inside.
Put Aros in the model
Book a conversation about putting arosplatforms in as the function that runs your AI operating model. We hold the rights, we run the cadence, and we stay. [Book a conversation](/demo) or [write us](/contact).