What Does It Take To Become An AI-First Company?
Giving employees access to AI does not make a company AI-first.
Executives are told to “give AI your knowledge,” “teach it your processes,” “train it on how decisions are made” and “provide examples of good decisions.”
That sounds sensible.
But in a company with hundreds or thousands of employees, how do you actually do it?
This is the part I have struggled to translate from good advice into something I could picture implementing in practice.
Much of the most valuable knowledge in an organisation has never been documented.
It appears when people challenge assumptions, explain exceptions, debate alternatives and make decisions based on experience that may never appear in a procedure manual.
The company usually preserves the outcome.
It often loses the reasoning that produced it.
So perhaps the first step towards becoming AI-first is not to ask employees to document everything for AI.
It is to change the working environment so that the knowledge the organisation already produces stops disappearing.
First, digitise important face-to-face decision-making
In many traditional businesses, important decisions are still made face-to-face.
There is nothing wrong with that.
The problem is what disappears when the meeting ends.
Imagine a management meeting deciding whether to discontinue Product A.
Ten people review a presentation, challenge assumptions, debate alternatives and finally decide:
Keep the product for another six months.
The minutes preserve the decision.
But why six months?
What alternatives were considered?
Which assumption changed somebody’s mind?
What did the sales director know that was not in the presentation?
The practical change is simple.
Digitise the meeting.
People can still sit together in the same room, but conduct important meetings through the company’s digital meeting environment as well, so the presentation, conversation and decision-making can be captured and transcribed.
AI can then turn that discussion into a structured record of the decision, alternatives, reasoning, evidence and unresolved questions.
The purpose is not to replace face-to-face interaction.
It is to stop valuable organisational intelligence from disappearing when the meeting ends.
But formal meetings are only where some of that intelligence is created.
Second, bring everyday company conversations into the company’s digital environment
Much of the organisation’s practical knowledge is exchanged between meetings.
A manager explains an exception. A salesperson shares something learned from a customer. Two colleagues solve a problem together.
When those discussions disappear into personal messaging apps or private exchanges, the organisation loses them too.
Large companies may already have platforms such as Microsoft Teams or Slack. The change may be less about buying new technology and more about reinforcing where work-related discussions should take place.
Smaller companies may need to invest in that environment.
The objective is not to give AI unrestricted access to every employee conversation. Data, privacy and access rules still apply.
It is to keep useful company knowledge inside a company-controlled digital environment where it can be retained, searched, shared and, where appropriate, made available to AI.
Now the company has captured more of its thinking.
But capture alone is not enough.
Third, connect the company’s thinking across channels
Return to Product A.
The decision starts in a management meeting.
Two weeks later, the sales team discusses it in Slack or Teams.
A month later, finance produces another analysis.
Three months later, someone raises a new concern.
Six months later, management reviews the decision.
These are not five unrelated pieces of information.
They are one continuing organisational conversation.
This is where I find the idea of omnichannel useful.
Companies already try to preserve continuity when a customer moves between a store, website, app and call centre.
The organisation needs the same continuity for its own thinking.
An issue may move between a meeting room, online meeting, chat, document and business system, but the context should remain connected.
At its simplest, that could begin with a shared reference such as
#Product-A-Retention
A project number, customer number, product code or decision ID could serve the same purpose. More sophisticated systems may create those relationships automatically.
The principle is simple:
Give important organisational subjects a persistent identity so the company’s reasoning about them remains connected across channels and over time.
Think of it as an organisational decision thread.
The company is no longer simply accumulating documents, transcripts and messages.
It is accumulating a history of how it thinks.
That creates something AI can potentially work with.
But becoming AI-first also requires AI to enter the work itself.
Then give AI its first job in a controlled decision process
Start with a well-scoped project involving people with sufficient authority, domain knowledge and AI fluency.
Choose a recurring decision where the inputs are reliable and the boundaries are clear.
Business software might already produce a deterministic output, such as an Excel dashboard.
Give AI that output together with clear rules and relevant context.
For example:
If A, recommend B. Otherwise, recommend C.
AI makes the first recommendation.
The responsible human evaluates it.
If AI recommends B and the human agrees, record the decision.
If AI recommends B and the human chooses C, AI asks why.
Perhaps the human knows something the system has never seen before. Perhaps there is an exception. Perhaps circumstances have changed.
Record that explanation with the information AI saw, the recommendation it made and the decision the human took.
The employee has not been asked to become an AI trainer.
They have simply done their normal job with AI participating in the workflow.
Where the outcome is measurable, record the expected result and review it later against what actually happened.
Now the organisation begins to create a learning loop:
Business data → AI recommendation → human judgement → explanation or correction → outcome → organisational memory → better future AI assistance
AI is no longer learning only what the company decided.
It can increasingly be given the context to understand why it decided it and what happened afterwards.
A different way of thinking about “training your AI”
This is a different meaning of “training your AI.”
It does not necessarily mean building a language model specifically for the company or asking thousands of employees to write training examples.
It means designing work so that useful organisational knowledge is produced as a by-product of normal activity.
Meeting dialogues stop disappearing when people leave the room.
Everyday discussions stop disappearing into inaccessible conversations.
Related decisions stop becoming disconnected fragments scattered across the organisation.
And selected AI systems begin participating in real work where human judgement can add knowledge the system did not have before.
Over time, the organisation develops a record not only of what it decided, but of why it decided it and what happened afterwards.
That may be the real foundation of an AI-first company.
The scarce resource is not necessarily the AI.
It is the organisational intelligence the company already produces every day but never records.
What Does It Take To Become An AI-First Company?
Giving employees access to AI does not make a company AI-first.
Executives are told to “give AI your knowledge,” “teach it your processes,” “train it on how decisions are made” and “provide examples of good decisions.”
That sounds sensible.
But in a company with hundreds or thousands of employees, how do you actually do it?
This is the part I have struggled to translate from good advice into something I could picture implementing in practice.
Much of the most valuable knowledge in an organisation has never been documented.
It appears when people challenge assumptions, explain exceptions, debate alternatives and make decisions based on experience that may never appear in a procedure manual.
The company usually preserves the outcome.
It often loses the reasoning that produced it.
So perhaps the first step towards becoming AI-first is not to ask employees to document everything for AI.
It is to change the working environment so that the knowledge the organisation already produces stops disappearing.
First, digitise important face-to-face decision-making
In many traditional businesses, important decisions are still made face-to-face.
There is nothing wrong with that.
The problem is what disappears when the meeting ends.
Imagine a management meeting deciding whether to discontinue Product A.
Ten people review a presentation, challenge assumptions, debate alternatives and finally decide:
Keep the product for another six months.
The minutes preserve the decision.
But why six months?
What alternatives were considered?
Which assumption changed somebody’s mind?
What did the sales director know that was not in the presentation?
The practical change is simple.
Digitise the meeting.
People can still sit together in the same room, but conduct important meetings through the company’s digital meeting environment as well, so the presentation, conversation and decision-making can be captured and transcribed.
AI can then turn that discussion into a structured record of the decision, alternatives, reasoning, evidence and unresolved questions.
The purpose is not to replace face-to-face interaction.
It is to stop valuable organisational intelligence from disappearing when the meeting ends.
But formal meetings are only where some of that intelligence is created.
Second, bring everyday company conversations into the company’s digital environment
Much of the organisation’s practical knowledge is exchanged between meetings.
A manager explains an exception. A salesperson shares something learned from a customer. Two colleagues solve a problem together.
When those discussions disappear into personal messaging apps or private exchanges, the organisation loses them too.
Large companies may already have platforms such as Microsoft Teams or Slack. The change may be less about buying new technology and more about reinforcing where work-related discussions should take place.
Smaller companies may need to invest in that environment.
The objective is not to give AI unrestricted access to every employee conversation. Data, privacy and access rules still apply.
It is to keep useful company knowledge inside a company-controlled digital environment where it can be retained, searched, shared and, where appropriate, made available to AI.
Now the company has captured more of its thinking.
But capture alone is not enough.
Third, connect the company’s thinking across channels
Return to Product A.
The decision starts in a management meeting.
Two weeks later, the sales team discusses it in Slack or Teams.
A month later, finance produces another analysis.
Three months later, someone raises a new concern.
Six months later, management reviews the decision.
These are not five unrelated pieces of information.
They are one continuing organisational conversation.
This is where I find the idea of omnichannel useful.
Companies already try to preserve continuity when a customer moves between a store, website, app and call centre.
The organisation needs the same continuity for its own thinking.
An issue may move between a meeting room, online meeting, chat, document and business system, but the context should remain connected.
At its simplest, that could begin with a shared reference such as
#Product-A-Retention
A project number, customer number, product code or decision ID could serve the same purpose. More sophisticated systems may create those relationships automatically.
The principle is simple:
Give important organisational subjects a persistent identity so the company’s reasoning about them remains connected across channels and over time.
Think of it as an organisational decision thread.
The company is no longer simply accumulating documents, transcripts and messages.
It is accumulating a history of how it thinks.
That creates something AI can potentially work with.
But becoming AI-first also requires AI to enter the work itself.
Then give AI its first job in a controlled decision process
Start with a well-scoped project involving people with sufficient authority, domain knowledge and AI fluency.
Choose a recurring decision where the inputs are reliable and the boundaries are clear.
Business software might already produce a deterministic output, such as an Excel dashboard.
Give AI that output together with clear rules and relevant context.
For example:
If A, recommend B. Otherwise, recommend C.
AI makes the first recommendation.
The responsible human evaluates it.
If AI recommends B and the human agrees, record the decision.
If AI recommends B and the human chooses C, AI asks why.
Perhaps the human knows something the system has never seen before. Perhaps there is an exception. Perhaps circumstances have changed.
Record that explanation with the information AI saw, the recommendation it made and the decision the human took.
The employee has not been asked to become an AI trainer.
They have simply done their normal job with AI participating in the workflow.
Where the outcome is measurable, record the expected result and review it later against what actually happened.
Now the organisation begins to create a learning loop:
Business data → AI recommendation → human judgement → explanation or correction → outcome → organisational memory → better future AI assistance
AI is no longer learning only what the company decided.
It can increasingly be given the context to understand why it decided it and what happened afterwards.
A different way of thinking about “training your AI”
This is a different meaning of “training your AI.”
It does not necessarily mean building a language model specifically for the company or asking thousands of employees to write training examples.
It means designing work so that useful organisational knowledge is produced as a by-product of normal activity.
Meeting dialogues stop disappearing when people leave the room.
Everyday discussions stop disappearing into inaccessible conversations.
Related decisions stop becoming disconnected fragments scattered across the organisation.
And selected AI systems begin participating in real work where human judgement can add knowledge the system did not have before.
Over time, the organisation develops a record not only of what it decided, but of why it decided it and what happened afterwards.
That may be the real foundation of an AI-first company.
The scarce resource is not necessarily the AI.
It is the organisational intelligence the company already produces every day but never records.
Continue reading