Before You Buy AI: Is It an AI Problem or a Process Problem?

Before investing in AI, a manufacturing or engineering SME should first establish whether the problem actually requires AI. If the underlying issue is unclear ownership, inconsistent processes, poor-quality data or unnecessary administration, adding AI can simply automate the weakness rather than remove it.

The better starting question is therefore not "Where can we use AI?"

It is:

What commercial problem are we trying to solve, and what is preventing us solving it today?

What is the difference between an AI problem and a process problem?

An AI problem exists where technology can perform or support a defined task more effectively because it can analyse, interpret, generate or process information at a scale or speed that would otherwise consume significant human resource.

A process problem exists where the fundamental way work is organised is unclear, inconsistent or unnecessarily complicated.

Examples of potential AI applications might include:

• Summarising large volumes of customer or market information
• Researching prospective accounts
• Categorising enquiries
• Drafting routine commercial content
• Extracting information from documents
• Identifying patterns within sufficiently reliable commercial data
• Supporting meeting notes and action capture
• Helping users interrogate structured internal information

Process problems look different.

For example:

• Nobody clearly owns quotation follow-up
• Different salespeople use different pipeline definitions
• Discounts are approved inconsistently
• CRM information is rarely updated
• Reports are manually recreated because source data cannot be trusted
• The same information is entered into several systems
• Management meetings produce actions that are not subsequently tracked

Applying AI to those situations may make individual tasks faster.

It does not necessarily fix the underlying commercial problem.

Why does this matter for manufacturing SMEs?

Smaller and mid-sized manufacturers rarely have unlimited management capacity, technology budgets or implementation resource.

Every improvement project therefore has an opportunity cost.

Time spent introducing a new AI system is time not spent resolving another commercial issue.

The test should consequently be commercial rather than technological:

Will this improve revenue, margin, capacity, decision-making or customer performance sufficiently to justify the cost and disruption?

If the answer cannot be articulated, the proposed use case probably needs further definition.

Start with the problem, not the tool

Imagine a manufacturer where salespeople spend several hours each week updating quotations and manually preparing reports.

It would be easy to conclude:

"We need AI to automate sales reporting."

But further examination might reveal:

• Quotation statuses are inconsistent
• Several people maintain separate spreadsheets
• CRM fields are not mandatory
• Nobody owns data quality
• The weekly report contains information management rarely uses

AI could certainly produce the report more quickly.

But the company may get a greater return by first simplifying the process, defining the required measures and removing duplicate administration.

Only then can automation address the work that genuinely remains.

When is AI likely to add value?

AI becomes more attractive when five conditions are reasonably clear.

1. The task is understood

The business can describe what happens today and what outcome is required.

2. The inputs are available

The information required to perform the task exists and can legitimately be accessed.

3. The task consumes meaningful capacity

Improving it will release time, improve quality or reduce delay sufficiently to matter.

4. Errors can be controlled

There is a proportionate way to check important outputs before they affect customers or commercial decisions.

5. There is a measurable commercial outcome

The business knows what improvement should result.

That might be:

• Faster quotation turnaround
• Reduced administration
• More account-development time
• Better opportunity qualification
• Faster customer response
• More consistent commercial reporting

Without a defined outcome, it becomes difficult to know whether the technology has created value.

Where should manufacturers be cautious about AI?

The greater the commercial consequence of an error, the more control is required.

A system drafting an internal meeting summary carries a very different level of risk from one automatically changing customer prices.

Likewise, using AI to identify possible dormant accounts for a salesperson to review is different from automatically deciding which customers should receive revised commercial terms.

A useful distinction is:

AI supporting a decision

versus:

AI making the decision

The second requires substantially greater confidence in the underlying data, rules and controls.

What if the underlying data is poor?

AI does not remove the need for reliable commercial data.

Suppose a business asks an AI system:

Which opportunities are most likely to close this quarter?

But the CRM contains:

• Out-of-date expected dates
• Inconsistent sales stages
• Duplicate opportunities
• Missing values
• Quotations that have not been followed up
• Opportunities that should already have been closed

The sophistication of the analysis is less important than the quality of the information beneath it.

The same principle applies to normal management reporting. Our guide to what a manufacturing MD should review every Monday explains why defining useful commercial information should normally come before selecting the system that produces it.

Should you automate an inefficient process?

Usually not without challenging the process first.

Consider a quotation that currently requires eight internal approvals.

Automating those eight approvals may reduce administration.

But management should first ask:

Do we actually need eight approvals?

If three would provide adequate commercial control, simplifying the process creates value before any technology is introduced.

This is an important distinction because technology can make inefficient processes more efficient without making them effective.

A practical test before approving an AI project

Before investing, management should be able to answer six questions:

If those answers are unclear, a pilot may still be worthwhile.

But it should be treated as exploration rather than an established business case.

Where might AI release commercial capacity?

For many manufacturing SMEs, the most attractive early applications may not be dramatic.

They are often tasks that consume commercial time without requiring somebody's full expertise.

Examples could include:

• Initial prospect research
• Summarising account history before a meeting
• Converting meeting notes into draft actions
• Categorising inbound enquiries
• Producing first drafts of routine communications
• Summarising market information
• Identifying gaps in CRM records
• Preparing first-pass analysis for management review

The objective is not to remove people from the commercial process.

It is to reduce low-value activity so that people can spend more time where judgement, technical knowledge, customer relationships and negotiation actually matter.

That connects directly with the wider question of how commercial capacity can be released before simply increasing headcount.

AI should follow commercial priorities

Manufacturers do not need an AI strategy disconnected from the rest of the business.

They need a commercial and operational strategy that identifies where technology can create useful leverage.

That means establishing:

Problem → Process → Data → Technology → Outcome

rather than:

Technology → Find somewhere to use it

The distinction may sound simple, but it can prevent considerable wasted effort.

What should you do next?

Before approving an AI project, write down the commercial problem in one sentence.

Then establish:

• What currently causes it
• What it costs in time, margin, delay or lost opportunity
• Whether the existing process is appropriate
• Whether the required data can be trusted
• What role AI would actually perform
• How improvement would be measured

If the business cannot answer those questions, the priority may be improving the process rather than introducing another tool.

Fox Healey's Commercial Performance framework considers Data and Systems alongside Market and Growth, Customers, Sales Execution, Pricing and Margin, and People and Capability.

Technology should support commercial performance.

It should not become the objective in its own right.