Nobody starts with AI. They start with a process somebody already runs by hand, every week, at a cost the finance team can find. The mandate says otherwise, so the leader who catches it goes shopping for a technology while the answer sits in an operations spreadsheet. The first decision is not which AI to buy. It is which of our own processes goes first, and what does it cost us today?
Where to Start With AI
How Do You Get Started With AI
in an Organization That Has None?
The order of the first decisions, the readiness check in front of them, and where each one actually gets made.
Built from real buyer questions in our sales meetings
Direct Answer
Start from a process done by hand that costs you something measurable. Choose the work before the technology, state what it costs to run by hand now, and prove the result on your own material before buying at scale. The order of those decisions matters, and so does the gate in front of them: you hold the data, an executive will clear the path, one process has a stated current cost. Iternal supplies tooling and people for the middle. The sequence is yours.
The limit: the order below does not survive an organization that has decided nothing. Iternal learned that the hard and painful way. Opening with general AI education proved a rabbit hole that burns enormous amounts of time. Product demonstrations turned into education conversations after which nothing happened. Time bled away on unqualified AI conversations until a fast qualify-and-disqualify step went in front of them. Name the process, its owner and what running it by hand costs, or you are not ready to start.
Three answers separate a start from a rehearsal. Name the material the work runs on and confirm you hold it. Name the executive who will clear obstacles rather than merely approve a line item, because buyers told us a sponsor who is also a blocker leaves the work dead where it stands. Name the hours the process consumes today, since a proof of value needs a number on both sides. The targeted questions below do the rest.
The sequence and the decisions inside it live on different pages. Setting the order and the entry gate is one job; making the individual calls is six others. Which candidates are yours and which goes first is decided on the use case selection page; what comparable organizations run sits in the use case catalog; whether to commit now belongs to the timing page; who performs the work is the staffing page; what a guided discovery produces is the AI Blueprint Builder page; and what an engagement costs in calendar time is the engagement page.
The Question Underneath the Mandate
Only two constraints reach us more often: material that may not go to a hosted service, and enterprise data too disorganized to answer a question. Third in line, and stated with more embarrassment than either, is the admission that nobody here knows where to begin.
The admission usually arrives with a deadline attached. A chief executive tells a technology leader the company needs AI, and that leader does not know where to start. A newly appointed director of AI needs quick wins fast. One buyer named the gap exactly: the chief executive’s vision runs ahead of what the team can execute. A mandate is not an objective, and somebody has to convert it.
The First Ninety Days: One Owner, One Exit Test, One Page per Step
A program stalls where nobody can say what finishing looks like. Give every step an owner and a test it has to pass, and the sequence stops being a strategy document and becomes a calendar. Six steps carry a mandate to a build, each decided elsewhere.
| Step | Who owns it | The exit test | Where it gets decided |
|---|---|---|---|
| 1. Set the floor — the shapes of work organizations like yours run. | The leader holding the mandate. | You can name three shapes of work in your own words, no product name in them. | the use case catalog |
| 2. Decide the shape of the commitment — move now, or hold. | The executive sponsor. | You have written down which parts of a first purchase survive a model change. | the timing page |
| 3. Name and rank the candidates — the scored list, and the one that goes first. | The manager who feels the pain. | The top candidate has an owner and a stated cost of running it by hand today. | the use case selection page |
| 4. Run the discovery — turn that candidate into written deliverables. | The sponsor, plus whoever knows the business. | You hold an architecture, a governance view and a business case finance will read. | the AI Blueprint Builder page |
| 5. Settle who builds it — internal people, delivered work, or both. | The executive sponsor. | Every task has a named person on both sides of the agreement. | the staffing page |
| 6. Scope the engagement — phases, deliverables and dates. | The sponsor with procurement. | A stated time to a real outcome in writing, not an open-ended retainer. | the engagement page |
The shaded rows are where a program is won or lost; the others can be repaired later. A badly chosen candidate costs you the quarter, and the credibility that came with it.
Education belongs in the sequence, and not at the front of it. Iternal runs a substantial education arm and states just as plainly that opening with general AI education is a rabbit hole. Both hold; they describe different moments. Literacy offered before anyone has named a process generates interest and no decision, while the same material attached to a chosen workflow makes a tool stick. For more information visit the training curriculum page.
Pin it down: questions for your evaluation
-
Which steps carry no charge, which are billable, and what does each billable one deliver in writing?Where the money starts, before anyone commits a calendar.
-
Who writes the exit test for each phase, and will those tests appear in the statement of work?That a phase can end, and both sides agreed in advance what ending means.
-
Which parts of the training are tied to the process we chose, and which are general AI literacy?Whether enablement attaches to the work or is sold alongside it.
Which Decision Comes First, and Where the Rest Get Made
The order of a program and the content of it are two different problems, and one document that carries both gets finished by nobody. The order is the smaller of the two: which decision comes first, who owns it, and what has to be true before that first decision is worth taking.
The content sits one level down. Naming a candidate process, scoring one against another and choosing which to prove first is the work of the use case selection page, where the scored list carries an owner and a cost baseline on each row. Cataloging a sector, quoting a schedule and pricing a build each belong to their own page for the same reason: six questions answered under one headline are six questions answered badly.
The Floor Is Lower Than the Room Admits
Executives arrive convinced their organization is uniquely behind. They are usually on the same floor as everyone else. Buyers describe it without decoration: none of the internal systems have any generative AI on them, nothing is in production, no AI PCs are in the environment. One leader expects the first year to be people testing AI to see what it does.
A zero baseline changes what a sensible first move looks like. An organization with nothing running needs one visible win, not a transformation program. Iternal recommends the same shape to almost everyone: crawl, walk, run — prove value at small scale, then extend. Small is cheap enough to survive being wrong, which matters more in month one than being right.
The Readiness Check, and What Each Failure Actually Means
Three questions decide whether the sequence above is available this quarter. Ask them out loud, with the people who would do the work.
Do you hold the data the work runs on?
A pass is a named repository, a named owner and permission to put a sample in front of a system; Iternal is consistent that value arrives with your own material rather than a shipped sample. A fail is a data problem wearing an AI costume.
Will an executive clear the path, not just sign the line?
A pass is an executive who will move a blocked meeting, release an expert for an afternoon and answer for the result. A fail is worse than an absence: buyers told us a sponsor who is also a blocker leaves the work dead where it stands.
Can you state what one process costs to run by hand today?
A pass is hours a month, headcount, and what one error costs when it escapes. A proof of value weighs the manual cost against the automated one, so a missing baseline removes the number that wins the argument. A fail is a measurement task of a week.
Read the verdict literally. Three passes and you have a start date. Two passes and you have a task rather than a strategy problem, so close it before a statement of work exists. One or none, and the next useful conversation is with your own leadership.
Answered elsewhere
- Whether this year is the year to commit at all — see the timing page.
- How to staff a build when hiring will not conclude in time — see the staffing page.
- Where the money comes from when no line exists — see the funding page.
- Whether your documents are fit to be asked questions — see the data readiness section.
- When the material may not go to a hosted service — see the confidential data section.
- How to bring people along once a first workflow exists — see the adoption section.
FAQ
FAQ: Beginning an AI Program From Nothing
Start from one process people already run by hand, state what it costs you now, and prove the result on your own material before buying at scale. The order matters more than the product: set the floor, rank the candidates and pick one, run a discovery, settle who builds it, then scope the engagement. Give every step an owner and an exit test.
Convert the mandate into an objective before you convert it into a purchase. A mandate names a technology; an objective names a process, its owner and what running it by hand costs today. Walk one process, one sponsor and one number into the next leadership meeting, and the mandate stops being abstract.
Begin where most organizations begin: a zero baseline is the normal starting position, not a failure. Buyers routinely report no generative AI on internal systems at all. Iternal recommends crawl, walk, run: prove value at small scale on one fast win, then extend. One visible win beats a transformation program nobody has funded.
Three things. You hold the data the work runs on, with permission to put a sample in front of a system this month. An executive will clear obstacles rather than only approve a budget line, because a sponsor who is also a blocker leaves the work dead where it stands. And you can state what one process costs to run by hand today.
Attach the training to a process instead. Iternal runs a substantial education arm and states just as plainly that opening with general AI education is a rabbit hole that burns enormous amounts of time. Literacy offered before anyone has named a process produces interest and no decision; attached to a chosen workflow, the same material makes a tool stick.
Start With the Process, Not the Product
Name one process, its owner, and what it costs to run by hand. The rest is a sequence of ordinary decisions with ordinary owners.