Left Brain Most AI projects fail before
anyone writes a line of code.
They get picked because someone asked loudly, aimed at a process nobody has written down, and staffed from a team that was already full. None of that is a technology problem. All of it is fixable in ten working days.
Bring the AI list you haven't started. We'll rank it together in under thirty minutes.
Why AI stalls
The model was never
the hard part.
Building an agent is a weekend. Getting a company to run on one is a year — and almost everything that goes wrong in between is organisational.
Nobody picked
Twelve things started because twelve people asked. The three that mattered are competing for the same four people, and none of them will finish.
The process underneath was never written down
You can't automate a process that runs differently depending on who's doing it. Most AI projects discover this in month three.
Nobody owns it
It crosses IT, ops, security and a department. No single budget line, no single name, so it dies in committee rather than in production.
The build was sized, the readiness wasn't
Everyone estimates the model work. Nobody estimates the six weeks of defining and standardising that has to happen first.
It was built for people who never asked
A central team ships something to a department that didn't want it. Adoption becomes a change programme instead of a non-event.
So the second one is harder to fund than the first
A pilot that produced a nice demo and no number makes the next approval slower. That's how AI budgets quietly die.
Every one of those is a decision failure, and none of them is fixed by buying another tool.
The part nobody scores
An AI project isn't one job.
It's a jump between stages.
Every process sits somewhere on this track. What makes an AI initiative big or small isn't the model — it's how many stages it has to cross before an agent can run against it.
Undefined
Nobody has written down how this actually works today.
Documented
Written down, but still done differently by different people.
Standardised
One way of doing it. The exceptions are named rather than improvised.
Instrumented
It produces a record something else can actually read.
Automatable
An agent can run against it, and you'd trust the output.
Every initiative gets a current stage and a target stage. That single pair is the difference between a six-week project and an eighteen-month one — and it's the number that turns an ambition into a bandwidth conversation.
This is where the room gets a third option. Not yes or no — approve the first move only. Take it to Standardised this quarter and revisit automating it when it actually is. Nobody loses face and the capacity cost drops to something real.
What gets installed
Ten working days.
You own all of it.
Your initiative portfolio doesn't exist as data anywhere — it lives in decks, heads and Slack threads. That's the real reason nothing can run on it.
One intake
A single door for every request that changes how the business operates — a tool, a process, an agent, a fix. Nothing becomes business as usual without coming through it.
One scoring standard
Impact, urgency and who's asking, on a fixed scale. Departments argue on one scale instead of five, and the guide defines what each level means.
Reasoning kept with the score
Every score carries the sentence that justifies it. It's why the ranking can be audited six months later, and why inflation gets caught.
Readiness staging
Current stage and target stage on every initiative, so effort reflects the work that actually has to happen rather than the build alone.
An honest capacity limit
What each team can absorb while still doing its day job, and who is already past it. One line, and the room goes quiet.
A decision log
What was decided, when, by whom, and why. The record your board asks for and almost nobody has — and the reason a no costs nothing to defend.
You keep the standard, the data and the cadence. It runs without the person who built it.
How it pays for itself
You don't need budget.
You need to stop something.
Duplicated, stalled and unowned work is already on the payroll. It just isn't producing anything. That's where the hours for the AI work come from.
Stop the waste
Work nobody owns and nobody finished is already being paid for.
Hours come back
Not a budget request. A reallocation of money you already spend.
Point it at the right build
The people who run the process define what it has to do. Engineers make it production-ready. Neither half works alone.
That frees more hours
Which funds the next one. People who automate a process pick up the next.
Nobody has to find money. Somebody has to decide out loud to stop something — which is a different, and much easier, conversation.
What I need from you
Four things. None of them
need system access.
- Your AI list
- However it exists. A spreadsheet, a board slide, an export, or three people's memory.
- Who leads each one
- Names, not teams. One person quietly carrying five things is invisible at the team level.
- One number per lead
- Roughly what share of their week is already taken by running the business. A thirty-second estimate is enough.
- What's been tried and died
- Ten minutes of conversation. It tells me where the failure pattern is before I see a single record.
Enterprise process tools need event logs from your ERP and six months of data hygiene before they say anything. I need a spreadsheet and four conversations.
The guarantee
In ten working days I'll find more change spend going to duplicated, stalled or unowned work than this costs.
If I don't, you don't pay.
Over 25% of effort was duplicated across initiatives at the thousand-person enterprise below. In any company running fifteen or more initiatives at once, one merged duplicate covers the fee.
Scoped honestly: the guarantee is on finding it and putting a number on it. Acting on it is your decision, not mine.
Proof
Restoring commercial stability
A thousand-person, $100M ARR enterprise. Automation, IT, process, data and compliance work all entering through different doors, with no shared intake and no prioritisation.
I was brought in to modernise performance reporting.
The assessment said otherwise: you cannot measure performance if you don't control how work enters the system. Another dashboard would have codified the dysfunction. So I built one intake, one review board, one release calendar, and standards ahead of execution — and left the reporting until last.
Leadership regained confidence — not because the dashboards improved, but because execution did.
Installed while I was an employee, leading the data side of a business unit turnaround. Now delivered independently, from the outside, in ten working days.
Whether this is for you
It works when
- You have eight or more initiatives running at once
- More than one team is competing for the same delivery people
- Somebody senior will enforce the decisions afterwards
- You'd struggle to name what's in flight without opening a file
It doesn't when
- You're under about fifty people and one person holds it all
- You want a tool rather than a way of deciding
- You want someone to run delivery for you — different job
- Nobody above the room is willing to say no to anything
If the first list doesn't describe you, I'll say so on the call rather than sell you an install you don't need.
Nathan Harris
Founder, Left Brain Consulting
Eight years of enterprise transformation at MetLife and IQVIA — ERP migration recovery, PMO buildouts, and executive change control boards. Fortune 50 and Fortune 275 scale, which mostly taught me that change fails for organisational reasons rather than technical ones.
I'm not an AI consultant. I don't build your agents and I won't tell you which model to use. I make sure the thing you build is the right one, that the people who need it asked for it, and that there's room in the week to actually finish it.
Left Brain Consulting is an independent practice based outside Denver. I take a small number of installs at a time, which is the honest reason the founding rate exists.
Bring the AI list you haven't started.
Thirty minutes, your real initiatives, ranked against one standard and against what your teams can carry. You'll know by the end of the call whether the install is worth having.