Before Your AI Hire Starts

Half of whether your AI hire succeeds will be decided before they start—and it will be decided by the executive team.

You’re about to give one person a broad mandate: find where AI has value, change how work gets done, manage the risk, and produce results the business can see.

The job description can describe that work. It cannot create the conditions required to do it. A capable AI leader can select tools, run trials, build workflows, and help people adopt them. They cannot define a mandate leadership never defined, measure work nobody measured, or recover decisions that exist only in people’s heads.

This brief covers what the executive team should put in place before day one.

Part one

Four moves, one preparation

Define the mandate before you hand it over

Most AI job descriptions list the same work: assess tools, collect use cases, set governance, train employees, run trials, and report usage.

All of it is necessary. None of it answers the main business question:

What should this company be able to do because AI is available?

“Save time” is not a mandate. It is one possible result, and on its own it is how a company ends up with a faster version of yesterday: the same assumptions, with more throughput.

Ask instead:

Which work exists only because of limits AI can now remove?
Where would additional capacity create a capability the company has never had?
Which customers, services, or decisions become economical once the work changes?

The new hire should sharpen the executive team’s answers, not invent them alone.

The division of responsibility is familiar: leadership owns the direction; the hire owns the tactics. You do not delegate finance and then stop having an opinion about capital. AI is no different.

Before day one, write a short mandate that states:

the business result the company wants
the limits of the role
the authority the role carries
the first result the company will count as a success

Without it, the AI leader chooses the program’s direction by default and is later judged against expectations nobody wrote down.

Decide which job you’re actually hiring for

Most initial AI roles are four jobs wearing one title.

Four jobs, one title: the required work

PlatformSelect tools, manage suppliers, define architecture, and control cost.
AdoptionChange workflows, train people, and build internal support.
GovernanceSet data rules, make risk decisions, and assign accountability.
BuildCreate engineering workflows, automations, and integrations.

One person may cover all four initially. But one of them will become the main job, and each requires different skills, time, and authority.

Before you close the search, write down the answers:

Which job is the main job?
Who owns each of the others?
What decisions can the new hire make?
What work is outside the role?

Give candidates these answers before they accept.

It prevents a common failure: the company hires someone to do technical work, expects them to change the enterprise, and gives them the authority for neither.

Give the new hire an inheritance, not an investigation

Left alone, a new AI leader can spend three months discovering what already exists: current systems, informal workflows, data owners, approval paths, and AI use the company has never formally acknowledged.

The executive team can hand over much of that information on day one.

Begin four small inventories before the hire starts.

The data mapWhere the source of truth is, and who can say yes to using it
The work baselineTwo or three recurring workflows, measured before you change them
The decision mapWhere the reasons behind recurring decisions live
The existing-use inventoryWhere employees already use AI

Four small inventories, begun before the hire starts.

The data map

List the important systems. For each one, record:

the data it contains
the person or team that owns it
the person who can approve access
the authoritative source when more than one system holds the same fact

Do not turn this into a data-platform project.

The map answers one practical question: Where is the source of truth, and who can say yes to using it?

The work baseline

Choose two or three recurring workflows that cause visible delay or rework: a monthly report, a committee pack, a client brief, or a recurring reconciliation.

Record how the work actually moves through the company. Measure it before you change it:

preparation hours per cycle
corrections and rework
time required to answer a standard request
handoffs and approval delays
repeated questions

The program cannot show improvement without a baseline. “It feels faster” will not survive budget review.

The decision map

Record where the reasons behind important recurring decisions live: pricing, priorities, supplier choices, escalations, and exceptions.

In most companies, the answer is a person, not a record.

That list of people becomes the new hire’s interview plan. It is also the context a future workflow will need.

Without it, a new system can produce general answers, but not answers that carry the company’s own judgment.

The existing-use inventory

Find out where employees are already using AI, including personal accounts, unofficial tools, embedded product features, and workflows built without formal approval.

Record:

the work being done
the information being used
who depends on the result
whether the workflow is experimental or operational
what would break if it disappeared

Do not begin with enforcement. Begin with visibility.

The new hire needs to know what the company is already learning—and what risk it is already carrying.

Protect the first 90 days

The company will want a visible AI demonstration immediately. Resist that pressure.

Use this sequence instead:

Days 1–30

Main workConfirm the inventories, meet informal users, and identify constraints.

Executive-team actionGive access to people. Resolve ownership questions. Do not demand a demo.
Days 31–60

Main workSelect one measurable workflow and establish its baseline.

Executive-team actionSupport a useful project, not merely a high-status one.
Days 61–90

Main workPut one workflow into use and apply basic controls.

Executive-team actionResolve access decisions quickly. Protect the team’s time to learn.

The right first workflow is rarely the one leadership most wants demonstrated.

Board packs and executive dashboards come with disputed definitions, multiple owners, and political risk. They are highly visible and offer little room for a clean first win.

The first project should optimize for adoption and evidence, not executive visibility.

Choose one that meets these conditions:

It repeats on a known cycle.
One owner can judge the result.
The people doing the work want it to change.
The starting condition can be measured.
It can go into use without a company-wide redesign.

The first goal is not an impressive demonstration. It is a useful system people trust—one the company can measure and explain.

Before the system goes into use, apply basic controls:

Name a person who can approve data access quickly.
Keep credentials and secrets secure.
Keep a record of each run, when it occurred, and the data it used.
Require human approval for important external output.
Use real test cases to identify regressions after workflow changes.
Keep prompts, workflows, and supporting documents in one known location.
Name the people who can change or remove those items.

AI collapses the time required to build. Everything else becomes the constraint: access, verification, context quality, and decision speed.

Illustrative, not to scale.

If an access decision takes six weeks, those six weeks are part of your product-development cycle.

Part two

In brief

Six predictable failures

Preventable before and during the first quarterNeeds continuing executive attention

Build before you map. The team ships a visible project quickly. Later, it discovers duplicate tools, missing context, or a better solution that already existed.
Skip the baseline. The work improves, but nobody can prove it.
Choose the high-status workflow. The project receives plenty of attention but has little chance of producing a clean win.
Stop after the demonstration. The demo works. Missing access, records, ownership, or testing keeps it out of production.
Make the AI leader the AI helpdesk. Every request routes to the new hire, who has no authority to set priorities or refuse work.
Treat the hire as the strategy. The role gets filled, the box gets checked, and the executive team stops thinking. The judgment about what the company should become stays with its executives—or it does not get made at all.

The executive team can prevent the first four before and during the first quarter. The last two require continuing executive attention.

What the new hire needs from the executive team

Before day one:

Before day oneTick what’s already in place

These are not favors. They are what make accountability possible.

When the mandate, authority, baseline, and access path are clear, the company can judge results. Without them, even an excellent hire can spend the first year negotiating the job while trying to do it.

Seven questions for the first week

What must this company be able to do in two years that it cannot do now?
Who already uses AI, and for what work?
Which recurring work consumes more time than leadership realizes?
Where are the reasons behind the last three important decisions?
Who can approve access to each data source, and how long does that approval take?
What would make the first production workflow trustworthy?
What result will the company accept as meaningful at 90 days?

The executive team does not need final answers before day one. It does need to recognize that these are leadership questions and be prepared to work through them with the new hire.