What executive teams should ask before approving an AI project

Why many AI projects fail before they deliver business value

An AI project should never begin with the technology. It should begin with a business problem that is worth solving.

We've seen organisations invest heavily in generative AI platforms because competitors were doing the same. Six months later, they had licences, demonstrations and enthusiastic presentations, but very little had changed inside their operations. Staff were still performing repetitive work, reporting remained inconsistent, and leadership struggled to identify any measurable return.

That situation is more common than many executives realise.

A successful AI implementation project starts with commercial priorities. Whether the objective is improving operational efficiency, reducing manual administration or increasing forecast accuracy, the outcome should be specific enough that everyone can measure success. If the business case disappears once you remove the words "artificial intelligence", the initiative probably isn't ready for approval.

"Technology rarely causes AI projects to fail. Weak business alignment usually does."

Start with the business problem, not the AI

One lesson repeatedly stands out when reviewing enterprise AI projects: organisations that define the problem clearly make faster progress than those that begin with the software.

Imagine a professional services firm in Melbourne where consultants spend hours preparing proposals every week. Management believes an AI solution could automate proposal writing. At first glance, that sounds sensible. After mapping the workflow, however, they discover the biggest delay is not writing. Staff spend most of their time searching multiple document libraries for previous proposals and pricing information.

The real opportunity becomes knowledge retrieval rather than text generation.

Questions executives should ask include:

• What business outcome are we trying to improve?
• How will customers or employees benefit?
• Which KPI will demonstrate success?
• What happens if we do nothing?

Those questions create a stronger AI business case than simply approving another AI initiative.

Ask whether AI is the best solution

Not every operational challenge requires artificial intelligence.

Sometimes a workflow redesign, better reporting process or existing automation platform delivers the same result for far less cost and complexity. I've been involved in conversations where leadership assumed machine learning was necessary, only to discover that standard business rules removed most of the manual effort.

A good executive team challenges assumptions before approving an investment.

Choosing AI because it is fashionable rarely stacks up. Choosing AI because it solves a problem that traditional software cannot is a much stronger foundation.

Decide your organisation's AI ambition before spending money

One of the most overlooked parts of AI strategy is agreeing on how ambitious the organisation wants to be.

Some Australian businesses want carefully governed AI adoption with a handful of high-value use cases. Others want enterprise AI embedded across multiple departments over several years. Neither approach is wrong, provided leadership agrees before projects begin.

Without that shared direction, marketing may invest in generative AI, operations may prioritise automation, finance may pursue predictive analytics and IT may introduce entirely different platforms. Everyone stays busy, but nobody moves in the same direction.

A practical AI roadmap should answer three simple questions:

• Where will AI create measurable business value first?
• Which projects deserve investment this year?
• How will success be measured before additional AI deployment is approved?

Those conversations often determine whether an AI transformation becomes a lasting business capability or simply another short-lived pilot project.

Build an AI project on strong data before investing in new technology

Many executives assume choosing the right AI platform is the hardest decision. Data quality is where most AI implementation projects succeed or fail.

I've seen organisations compare Microsoft Copilot, ChatGPT Enterprise and custom AI development for weeks, only to discover that their customer records, pricing data and operational reports all used different definitions. The technology was ready. The information feeding it was not.

That is why strong AI implementation starts with data, not software.

Is your data ready for an enterprise AI project?

An experienced analyst should be able to answer the business question before AI enters the conversation. If they cannot trust the available data, an AI model will not magically produce reliable answers.

Consider a national retailer operating across Sydney, Brisbane and Perth. Sales, finance and operations each maintain their own reporting process. Revenue figures vary slightly between departments because discounts, returns and promotional sales are classified differently.

Introducing AI into that environment only magnifies the inconsistency.

Before approving an AI development project, executives should ask:

• Is our data accurate enough for AI decision-making?
• Are key systems connected?
• Do departments use consistent business definitions?
• Is our information current enough for real-time decisions?

Many organisations rush into AI deployment before answering those questions. The project may look impressive at first, but problems appear quickly.

Create one shared version of business information

Enterprise AI relies on shared business context.

When one department defines an "active customer" differently from another, AI systems cannot consistently interpret requests. The result is conflicting reports, unreliable insights and lower confidence in AI-generated recommendations.

One practical way to avoid this is by creating shared business definitions before expanding AI use cases.

For example:

This may sound basic, but shared definitions improve AI governance, reporting accuracy and decision-making far more than many executives expect.

Spend more on data foundations than AI software

A common mistake is allocating most of the AI investment to software licences while leaving little budget for preparation.

Successful organisations often reverse that thinking.

Instead of spending everything on AI platforms, they prioritise:

• Data quality improvements.
• AI governance.
• Integration with existing business systems.
• Staff capability and AI training.
• Change management.
• AI security and compliance.

The software becomes only one part of a much larger AI operating model.

When executive teams view AI implementation as a business capability rather than a technology purchase, projects are far more likely to produce measurable business outcomes that continue long after the initial deployment.

AI governance questions every executive team should answer before approval

Governance often gets a bad reputation. Some executives worry it will slow down delivery, create unnecessary paperwork or discourage innovation. In practice, the opposite is usually true.

The organisations that move fastest with AI are rarely the ones taking shortcuts. They are the ones that establish clear ownership, sensible guardrails and decision-making processes before the first AI solution goes live. When everyone knows who is responsible, projects keep moving instead of getting bogged down in uncertainty.

"Governance is the steering wheel that helps organisations move faster without losing control."

Who owns the AI project when something goes wrong?

One question I always encourage leadership teams to ask is simple:

"If this AI system produces the wrong outcome tomorrow morning, whose phone rings first?"

If nobody can answer, the project is not ready.

An AI project manager can coordinate delivery, but accountability belongs with the business. Operations leaders should own operational workflows. Finance should own financial models. HR should own people-related AI solutions. Technical teams build and support the platform, but they should not carry responsibility for business decisions made using AI outputs.

A practical governance framework assigns clear responsibilities before development begins.

Clear ownership also strengthens executive sponsorship. Instead of several committees sharing responsibility, one accountable leader drives the AI initiative from planning through ongoing operation.

How will you manage privacy, bias and compliance?

Australian organisations cannot afford to treat AI governance as an afterthought.

Whether customer information sits in Microsoft 365, Google Workspace or cloud AI platforms, executives should understand how personal information is handled, who has access and what safeguards are in place.

Responsible AI starts with practical questions rather than legal jargon.

Ask questions such as:

• Could this AI system expose confidential information?
• Are sensitive records protected appropriately?
• How will we identify biased or misleading outputs?
• Does the project align with organisational policies and Australian privacy obligations?
• Have we documented acceptable use for employees?

Many executives focus heavily on AI capability while giving little attention to compliance or security. That imbalance often creates more risk than the technology itself.

Have you planned for the worst-case scenario?

Every AI implementation project deserves the same level of contingency planning as any other major business initiative.

Imagine your preferred AI vendor experiences an extended outage. Alternatively, a model update changes response quality overnight, affecting customer communications or operational decisions. Without preparation, teams scramble to respond while confidence quickly disappears.

Before approving an AI investment, leadership should work through scenarios such as:

• Critical AI services becoming unavailable.
• Incorrect recommendations affecting business decisions.
• Data quality deteriorating over time.
• Third-party vendors changing pricing or service conditions.
• Regulatory requirements changing during the project lifecycle.

A straightforward risk assessment completed before AI deployment is usually far less expensive than trying to repair trust after something goes wrong.

Good governance does not eliminate risk altogether. It creates enough visibility that leaders can make informed decisions while keeping AI initiatives moving with confidence.

Why people determine whether AI adoption succeeds or fails

Technology rarely changes a business on its own. People do.

I've worked with organisations where the AI solution performed exactly as designed, yet adoption stalled within weeks. The software worked. The problem was that employees were unsure when to use it, managers continued relying on old processes, and nobody had explained how success would be measured. Before long, staff slipped back into familiar habits.

The strongest AI implementation projects invest just as much in people as they do in technology.

Decide where humans stay in control

One of the first governance decisions executives should make is determining where AI supports decisions and where humans remain responsible.

For example, an AI assistant might draft customer responses, summarise lengthy reports or recommend inventory levels. That does not mean employees should stop reviewing important decisions.

A useful framework is to classify each AI use case into one of three categories:

Keeping people involved in higher-risk decisions builds trust and improves AI governance while reducing operational risk.

Prepare teams before AI deployment begins

Many organisations approve the technology budget but overlook the people budget.

That can become expensive.

Imagine a logistics company introducing AI automation to manage customer enquiries. The software performs well during testing, yet customer satisfaction drops after launch because frontline staff were never trained to review AI-generated responses or resolve exceptions.

Technology alone cannot fix poor adoption.

Before AI deployment, executive teams should plan for:

• Role-based AI training.
• Clear usage guidelines.
• Internal champions across departments.
• Regular feedback sessions.
• Practical examples using real business workflows.
• Ongoing support after rollout.

Training works best when employees immediately apply new skills to their daily work rather than attending generic demonstrations they forget by Friday.

Build an AI project team that represents the whole business

Successful AI projects are business-led, not IT-led.

Technology specialists remain essential, but operational leaders, finance teams, legal advisers and frontline employees should all contribute during planning. Each group understands risks and opportunities that others may overlook.

A balanced AI project team often includes:

• Executive sponsor.
• Business owner.
• AI project manager.
• Data owner or steward.
• Technical lead.
• Security or compliance representative.
• Operations representative.
• End-user representative.

When these stakeholders contribute early, the AI solution becomes more practical because it reflects how work happens instead of how people assume it happens.

That collaborative approach also makes AI adoption much smoother. Employees are far more likely to trust a solution when colleagues have helped shape it from the beginning.

How executive teams should measure AI ROI before approving an AI project

One of the biggest mistakes leadership teams make is treating AI deployment as the finish line. It is only the beginning.

I've seen businesses proudly announce the completion of an AI implementation project, only to struggle months later when executives asked a simple question: "What has improved?" Nobody had defined success before the project started, so there was little evidence of business value.

That is why AI performance measurement belongs in the planning phase, not the post-project review.

Define success metrics before the AI development project begins

A successful AI project should be measured like any other business investment.

Instead of vague objectives such as "improve efficiency", establish measurable outcomes before approving the budget.

Examples include:

• Reduce proposal preparation time by 40%.
• Cut customer response times by 25%.
• Reduce manual document processing by 60%.
• Improve forecast accuracy by 15%.
• Increase first-contact resolution by 10%.

These targets make AI ROI easier to measure and help executive teams decide whether further AI investment is justified.

Set realistic AI project timelines

Many organisations expect dramatic improvements within a few weeks. That expectation rarely reflects reality.

A practical AI implementation roadmap usually looks like this:

Some early productivity gains may appear within the first few months, but meaningful organisational change often takes longer. Executive teams should plan for steady improvement rather than overnight transformation.

Plan for AI maintenance instead of treating deployment as the finish line

AI systems are never truly finished.

Business processes change, customer behaviour shifts, regulations evolve and new data becomes available. An AI solution that performs well today may gradually lose accuracy if nobody monitors it.

Before approving an AI project, leadership should confirm that someone is responsible for:

• Monitoring AI performance.
• Reviewing model outputs.
• Updating business rules when required.
• Measuring business outcomes.
• Identifying new opportunities for optimisation.

When continuous improvement becomes part of the AI operating model, organisations gain far more value from every project they approve. Instead of collecting isolated AI initiatives, they build a growing capability that supports better decisions, stronger productivity and measurable commercial outcomes year after year.

Executive AI project approval checklist for Australian organisations

Before approving any AI project, executive teams should pause and confirm that the fundamentals are in place. Strong AI implementation is rarely about finding another impressive tool. It is about approving the right project, at the right time, with the right governance and ownership.

A practical checklist helps separate projects that are ready to move forward from those that need more work.

Questions every executive team should answer before signing off

Before approving an AI investment, confirm that you can answer "yes" to the following:

• We have defined a measurable business problem.
• AI is the best solution for this challenge.
• Our data is accurate and suitable for the proposed use case.
• A business owner is accountable for outcomes.
• Governance, privacy and security requirements have been addressed.
• Employees will receive role-based training.
• Success metrics and reporting have been agreed.
• A plan exists for ongoing monitoring and improvement.

If several of these questions remain unanswered, delaying approval is usually the wiser decision.

Common warning signs that an AI project is not ready

The following table provides a simple executive review before approving any artificial intelligence project.

The most successful AI initiatives are rarely the most ambitious on day one. They are the ones that solve a genuine business problem, demonstrate measurable value and build confidence for the next stage of adoption.

Executive teams that ask the right questions before approving an AI project give themselves a far better chance of turning AI strategy into practical business outcomes. Rather than chasing isolated pilots, they create governed, repeatable capabilities that improve productivity, support better decisions and deliver value long after the initial deployment.

TL;DR

Approving an AI project requires far more than selecting the right technology. Executive teams should begin with a measurable business problem, confirm AI is the right solution, strengthen data quality, establish governance, prepare employees and define success metrics before deployment. Organisations that prioritise clear ownership, practical AI adoption and continuous improvement are far more likely to achieve measurable ROI and build AI capabilities that deliver lasting business value.