Automation
AI Strategy for Business Leaders: What to Do Before Buying Tools
AI strategy should start with value, workflows, governance, data, adoption, and measurement before business leaders buy tools or launch pilots.

Many AI initiatives start with the wrong question.
Which tool should we buy?
That question feels practical. It creates motion. It gives the team something to compare, test, budget, and deploy.
But it is often too early.
A company can buy a strong AI tool and still fail to create value. The tool may sit outside the real workflow. The data may be weak. The risk rules may be unclear. Managers may not know what behaviour should change. Employees may experiment privately. Leaders may track usage instead of outcomes. The pilot may work, but nobody knows how to scale it.
AI strategy for business leaders should start with the value, workflow, governance, data, adoption, and measurement questions that decide whether tools can create business impact.
The tool matters.
But it should not lead the strategy.
In this article
The Short Answer
Before buying AI tools, business leaders should answer seven questions.
1. What business value are we trying to create? 2. Which workflow or decision process should change? 3. What data, context, and knowledge does the tool need? 4. Who owns the outcome, risk, adoption, and improvement loop? 5. What guardrails are required before use? 6. How will people learn and change their work? 7. How will we measure value beyond tool usage?
| Tool-first AI | Strategy-first AI |
|---|---|
| Starts with vendor demos | Starts with value pools |
| Optimizes isolated tasks | Redesigns workflows |
| Measures logins and prompts | Measures business outcomes |
| Leaves governance vague | Defines decision rights and risk rules |
| Assumes adoption will happen | Builds learning and manager support |
| Treats data as a later issue | Tests data readiness early |
| Scales enthusiasm | Scales proven workflows |
The goal is not to slow everything down.
The goal is to avoid fast waste.
Start With A Value Pool
AI strategy should begin with a value pool.
A value pool is an area of the business where better speed, quality, cost, decision-making, personalisation, risk control, or scalability would matter.
Examples include:
- faster customer support resolution;
- better sales research and proposal quality;
- lower operational rework;
- faster management reporting;
- improved content production and review;
- better compliance monitoring;
- faster product feedback synthesis;
- more consistent decision preparation.
| Value pool | Leadership question |
|---|---|
| Revenue growth | Where could AI improve conversion, retention, or expansion? |
| Cost reduction | Where is skilled time wasted on repeatable work? |
| Quality | Where do outputs vary too much between people? |
| Speed | Where do delays block customers or decisions? |
| Risk | Where does human review need better support? |
| Learning | Where does the organisation keep relearning the same thing? |
This connects to <a href="https://julienflorkin.com/business/business-metrics-improve-execution/">business metrics that actually improve execution</a>.
If leaders cannot name the value pool, the AI initiative will drift toward activity.
Activity is not strategy.
Map The Workflow Before Choosing The Tool
AI creates more value when it changes the workflow, not only one task.
A tool may help someone write faster, summarise faster, analyse faster, or search faster. That can be useful. But bigger value often appears when the whole workflow is redesigned.
Before buying, map the current workflow:
- where work starts;
- who provides inputs;
- where information is stored;
- which decisions are made;
- where handoffs happen;
- where quality breaks;
- where delays appear;
- what gets reviewed;
- what happens after output is produced.
| Workflow question | Why it matters |
|---|---|
| What is the actual sequence of work? | Prevents automating the wrong step |
| Where is the bottleneck? | Focuses AI on leverage |
| Who uses the output? | Clarifies adoption requirements |
| What must stay human? | Defines review and judgement |
| What changes after AI is introduced? | Prevents parallel old and new processes |
This is the practical foundation of <a href="https://julienflorkin.com/business/workflow-mapping-business-leaders/">workflow mapping for business leaders</a>.
If the workflow is unclear, the tool will be blamed for a design problem.
Decide What Should Not Be Automated
A serious AI strategy defines boundaries.
Not every task should be automated. Some work requires human judgement, empathy, accountability, regulatory interpretation, negotiation, or strategic choice.
The question is not, "Can AI do this?"
The better question is:
What role should AI play in this decision or workflow?
| AI role | Example |
|---|---|
| Assistant | Draft, summarise, classify, retrieve, compare |
| Analyst | Identify patterns, anomalies, options, risks |
| Coach | Help users improve prompts, writing, or decisions |
| Control layer | Flag compliance issues or missing evidence |
| Automation agent | Execute bounded tasks under rules |
| Human replacement | Only where risk, quality, and accountability are acceptable |
Leaders should define:
- what AI can produce;
- what humans must review;
- what requires approval;
- what data cannot be used;
- what outputs cannot go directly to customers;
- when escalation is required.
This avoids both fear-based paralysis and reckless automation.
Check Data And Context Readiness
AI tools do not create good answers from poor context.
Many businesses want AI performance before they have clean data, useful documentation, clear ownership, or reliable source material.
Before buying a tool, ask:
- Where is the relevant data?
- Who owns it?
- Is it current?
- Is it structured enough?
- Is sensitive information protected?
- Are source documents reliable?
- Can the tool access the right context?
- Who checks output quality?
| Data issue | AI impact |
|---|---|
| Outdated documents | Wrong or stale answers |
| Fragmented systems | Incomplete context |
| Poor ownership | Slow improvement |
| Sensitive data exposure | Security and compliance risk |
| No source standard | Weak trust in output |
| Inconsistent taxonomy | Harder retrieval and automation |
AI strategy is often data strategy in disguise.
If leaders ignore this, the company may buy tools that demonstrate well but perform badly in real work.
Define Governance Before Scale
Governance should not be a bureaucratic wall.
It should be a set of practical rules that make responsible use easier.
Good AI governance defines:
- approved tools;
- prohibited data;
- human review standards;
- customer-facing use rules;
- model and vendor risk;
- security requirements;
- auditability;
- escalation paths;
- ownership of incidents;
- how tools are evaluated over time.
| Governance question | Decision needed |
|---|---|
| Who approves use cases? | Governance owner |
| Who owns risk? | Legal, security, business, or shared |
| Who reviews outputs? | Human accountability |
| What must be logged? | Audit standard |
| What happens when AI fails? | Escalation and correction path |
This supports <a href="https://julienflorkin.com/business/ai-alignment-business-governance/">AI alignment and business governance</a> because strategic control matters more as AI moves closer to decisions and execution.
Build Adoption Into The Strategy
AI adoption is not automatic.
People may ignore the tool, misuse it, overtrust it, hide their use, resist it, or use it only for shallow tasks.
Leaders should plan adoption before buying.
That means answering:
- Which roles will use the tool?
- What work will change?
- What training is role-specific?
- Which managers will coach adoption?
- What examples will show good use?
- How will feedback improve the workflow?
- What incentives support the new behaviour?
| Adoption layer | What to design |
|---|---|
| Role clarity | Who uses AI for what |
| Training | Practical workflow-specific practice |
| Manager support | Coaching and expectation setting |
| Examples | Before-and-after use cases |
| Feedback | Improvement loop for outputs and process |
| Trust | Clear limits, review rules, and transparency |
This links to <a href="https://julienflorkin.com/business/why-ai-adoption-human-behavior-problem/">why AI adoption is a human behaviour problem</a>.
A tool rollout without behaviour design is not an AI strategy.
Choose The Right Pilot
A pilot should be small enough to learn quickly and real enough to matter.
Weak pilots are isolated demos. Strong pilots test a workflow, a business outcome, a risk model, and an adoption path.
| Weak pilot | Strong pilot |
|---|---|
| Tests tool features | Tests business workflow |
| Uses artificial examples | Uses real work under controlled conditions |
| Measures user excitement | Measures time, quality, risk, and adoption |
| Has no owner | Has business, technical, and risk ownership |
| Ends with a presentation | Ends with scale, adapt, or stop decision |
A good pilot defines:
1. The workflow. 2. The owner. 3. The users. 4. The data. 5. The guardrails. 6. The success metrics. 7. The scale path.
If there is no plausible scale path, the pilot may still be useful for learning.
But leaders should not confuse learning with transformation.
Measure Value, Not Tool Usage
Tool usage matters, but it is not enough.
High usage can mean the tool is useful. It can also mean people are generating more low-value output. Low usage can mean resistance. It can also mean the tool was poorly matched to the workflow.
Better metrics include:
- cycle time;
- rework;
- quality scores;
- customer response time;
- conversion rate;
- error reduction;
- decision speed;
- employee confidence;
- review burden;
- cost per process;
- revenue or margin impact.
| AI metric | Better question |
|---|---|
| Number of users | Who changed their workflow? |
| Number of prompts | What output improved? |
| Time saved | Where was saved time reinvested? |
| Pilot count | Which pilots scaled? |
| Tool spend | What value pool improved? |
This connects to <a href="https://julienflorkin.com/business/accountability-operating-rhythm/">accountability as an operating rhythm</a>.
AI value should be reviewed in cadence, not celebrated once at launch.
Decide Build, Buy, Or Configure
Only after the previous questions should leaders compare tools.
The choice is not simply build versus buy. Many companies need a mix:
- standard SaaS AI features;
- configured workflows;
- internal knowledge assistants;
- automation layers;
- custom models or agents;
- governance and monitoring tools;
- integration with existing systems.
| Option | Best when |
|---|---|
| Buy | The workflow is common and the vendor is mature |
| Configure | The process is specific but not strategically unique |
| Build | The capability is core, proprietary, or strategically differentiating |
| Partner | Internal capability is missing but ownership must remain clear |
The decision should follow strategic relevance, risk, data sensitivity, integration needs, and total cost of change.
The cheapest license may be expensive if it creates workflow confusion.
A Pre-Purchase AI Strategy Checklist
Before signing a contract, leaders should be able to answer this checklist.
| Area | Question |
|---|---|
| Value | What business outcome should improve? |
| Workflow | What work changes, and who owns it? |
| Data | What context does AI need, and is it ready? |
| Risk | What can go wrong, and who is accountable? |
| Governance | What rules apply before deployment? |
| Adoption | How will people learn, trust, and use it? |
| Measurement | How will value be proven? |
| Scale | What happens after the pilot? |
| Cost | What is the full cost of tool, change, training, and maintenance? |
If these answers are weak, buying the tool will not fix them.
Common Mistakes
The most common AI strategy mistakes are predictable.
| Mistake | Better approach |
|---|---|
| Starting with vendor demos | Start with value pools |
| Buying before workflow mapping | Map the workflow first |
| Treating AI as an IT project | Make it business-led with technical support |
| Measuring adoption only | Measure workflow and business impact |
| Ignoring managers | Use managers to coach behaviour change |
| Skipping governance | Define guardrails before scale |
| Assuming data is ready | Test data and context early |
| Running pilots without a scale path | Decide scale, adapt, or stop criteria |
AI strategy fails when leaders buy possibility and postpone responsibility.
Final Thought
Business leaders do not need to become AI engineers before making good AI decisions.
They do need to ask better questions.
Before buying tools, clarify the value pool, workflow, data, governance, adoption, measurement, ownership, and scale path.
Then the tool conversation becomes much easier.
The right AI tool can create leverage.
But only after the business knows where leverage is supposed to appear.
FAQs
What should business leaders do before buying AI tools?
Business leaders should define the target business value, map the workflow, check data readiness, clarify governance, assign ownership, design adoption, choose success metrics, and decide how a pilot could scale.
Why do AI initiatives fail after tool purchase?
AI initiatives often fail because the tool is not connected to a real workflow, data is weak, governance is unclear, adoption is unmanaged, managers are not prepared, or success is measured by usage instead of business impact.
Should AI strategy be led by IT or the business?
AI strategy should be business-led and technically enabled. Business leaders must own value, workflow change, adoption, and accountability, while technical teams support architecture, security, integration, and data.
What is a good first AI pilot?
A good first pilot targets a real workflow, has a clear owner, uses controlled data, includes guardrails, measures business outcomes, and ends with a decision to scale, adapt, or stop.
How should leaders measure AI value?
Leaders should measure outcomes such as cycle time, quality, rework, customer response, decision speed, cost, revenue, risk reduction, and workflow adoption, not only logins, prompts, or tool usage.