The Blind Leading the Blind

The Blind Leading the Blind

How AI sales hype damages legitimate AI capability

When “AI transformation” becomes a sales label before it becomes a defined organisational capability, the client is not buying technology. The client is buying problems: vague promises, undefined delivery, governance risk, and supplier rhetoric dressed up as strategic progress.

By Dr Danie Adendorff

It is false to say that all AI work is nonsense. That is not a serious argument. It is lazy, technically wrong, and strategically harmful.

Serious AI work exists. Serious AI engineering exists. Serious AI governance exists. So do model evaluation, data architecture, automation design, workflow analysis, risk management, cyber assurance, and decision-support systems. These are not marketing slogans. They are professional disciplines. They require technical competence, evidence, testing, accountability, and an honest understanding of what the system can and cannot do.

The issue is not whether AI is useful. That question has already been answered. AI is useful when it is engineered, governed, tested, bounded, and placed inside a responsible operating model. The issue is whether the market can still distinguish legitimate AI capability from AI-labelled sales hype.

That distinction matters because the hype economy does not merely mislead buyers. It damages the credibility of serious AI work by allowing vague commercial promises to borrow the authority of real technical disciplines.

The problem is not AI. The problem is the commercial fog now surrounding AI. The public-facing AI market has been heavily contaminated by people selling “AI transformation” before they can define the capability, evidence base, data boundary, operating model, risk control, or accountability structure.

That is not innovation. It is sales theatre.

This is where the phrase “the blind leading the blind” becomes accurate.

The buyer does not properly understand what is being bought. The seller often does not properly understand what is being sold. The word “AI” then becomes a convenient fog that hides the competence gap between them.

The result is not transformation. It is risk transfer.

The supplier sells confidence. The client inherits the consequence.

The danger is not artificial intelligence itself. The danger is AI-labelled intermediation without professional standards.

The Market Created by the Confidence Gap

Many SME owners are under pressure. They are told that AI will change everything. They are told that competitors are already using it. They are warned that delay means decline. They are told that if they do not adopt AI now, they will be left behind.

That pressure creates urgency. Urgency creates confusion. Confusion creates a market for anyone who sounds certain.

The confidence gap is not created because clients are stupid. It is created because sellers can sound certain in a field where many buyers do not yet know how to test the claim.

This is the entry point for the new AI intermediary.

Some intermediaries are competent. They understand workflow design. They know the limits of the tools they use. They can explain what the system will do, what it will not do, what data it touches, where human review remains necessary, and who carries responsibility when something fails.

Others are not competent. They understand sales funnels, scripts, audits, retainers, pain-point discovery, and lead capture far better than they understand data, systems, governance, operational risk, or accountability.

The distinction matters.

A competent AI service provider defines the problem before selling the solution. An inflated AI intermediary sells the future before defining the work. One begins with capability. The other begins with persuasion.

This is not always fraud. Often it is professional overreach. A consultant learns a few tools and becomes an “AI strategist”. A marketing adviser adds “AI automation” to an existing service. A freelancer packages templates, CRM triggers, Zapier links, Canva assets, email sequences, and generic productivity language into something that sounds much larger than it is.

Some of those tools are useful. That is not the issue.

The issue is inflation.

Small tactical support is sold as transformation. Micro-automations are presented as strategic architecture. Templates are sold as systems. Prompting is sold as intelligence. Confidence is presented as capability. The seller appears advanced because the buyer is even further behind.

That asymmetry creates the market.

The buyer hears “AI transformation”. The seller sells momentum. The actual deliverable is often a thin automation layer, a generic workflow, a bundle of third-party tools, or a retainer wrapped in strategic language.

That is not transformation. It is AI-labelled sales packaging.

Why the Label Misleads

The phrase “AI service” is now too loose to mean very much on its own.

Prompting is not automation. Automation is not data integration. Data integration is not machine learning. Machine learning is not governance. A ChatGPT workflow is not an accountable decision-support system. A chatbot is not an operating model. A dashboard is not intelligence. A few productivity hacks are not transformation.

These distinctions matter because each category carries different capability, risk, cost, maintenance, and accountability implications.

A business owner may understand the operational problem very well: too many enquiries, slow administration, weak follow-up, poor reporting, repetitive paperwork, inconsistent client communication, fragmented records, or poor visibility over performance. What the owner may not understand is whether the proposed AI solution is a genuine system improvement or merely a consultant’s sales funnel dressed up as strategy.

This is where vague AI selling becomes dangerous.

The client does not necessarily lack intelligence. The client lacks the specialist vocabulary and technical frame needed to interrogate the offer. The seller is not necessarily malicious. But if the seller cannot define the capability, data boundary, risk controls, toolchain, maintenance burden, and accountability model, the offer is not mature enough to be sold as transformation.

“AI-related services” sounds sophisticated enough to sell and vague enough to avoid responsibility.

That is the problem.

Real AI Service Versus AI Theatre

The clean distinction is not between “AI people” and “non-AI people”. That is too crude.

The real distinction is between defined capability and vague promise.

A real AI service begins with a defined problem. It identifies the workflow being improved. It defines the data boundary. It names the toolchain. It specifies where human review remains necessary. It explains what failure looks like. It identifies risk controls. It assigns accountability.

A vague AI service begins with slogans.

It promises to save time, cut costs, automate the business, optimise operations, scale output, unlock growth, and transform the organisation. Then it moves quickly to an audit, a discovery call, a monthly retainer, or a generic implementation package.

The difference is structural.

A real service can be tested.

A vague promise can only be believed.

A real service explains what the system will do on Monday morning inside the business. It states what the system will not do. It identifies what data it touches. It names which platforms, models, applications, automations, databases, plugins, and third-party services are involved. It defines where human review remains necessary and what happens when the workflow fails.

A vague promise relies on the buyer not knowing which questions to ask.

That is why vague AI intermediation is not a harmless market phase. It damages trust in legitimate AI. It teaches business owners that AI is another consultancy trick. It blurs the line between useful automation and inflated sales language. It allows underqualified sellers to trade on the credibility built by serious engineers, researchers, risk specialists, and governance professionals.

The people hyping everything as AI for quick commercial gain are not defending AI. They are damaging it.

The credibility of AI depends on separating real capability from bullshit.

The Operational Hangover

The operational problem begins after the sale.

When a brittle sequence of API calls, third-party prompts, template workflows, and loosely connected automation tools is rebranded as institutional capability, the architecture becomes inherently unstable. The client is left with an unmapped, unvetted stack of dependencies. Each dependency becomes a potential point of failure for data privacy, business continuity, client communication, auditability, and regulatory compliance.

This is the operational hangover hidden inside many micro-automation packages.

The sales pitch focuses on speed. The business inherits fragility.

The pitch promises efficiency. The business inherits maintenance.

The pitch sells transformation. The business inherits a stack it does not understand, cannot properly audit, and may struggle to control when something goes wrong.

A poorly designed micro-automation can work well enough during a demonstration and still fail under real operating conditions. It can break when a platform changes its pricing, when an API changes, when a staff member leaves, when a prompt produces unreliable output, when data moves into the wrong environment, when a plugin behaves unpredictably, or when the business tries to scale a workflow that was never designed as a governed system.

This is not a minor technical issue. It is a governance liability.

The SME owner does not merely buy a tool. The SME owner inherits a dependency structure. That structure affects continuity, information security, customer trust, record-keeping, legal exposure, and management control.

True operational capability cannot be bought off the shelf from an intermediary whose primary qualification is an optimised sales script. It requires disciplined alignment between data integrity, workflow design, human oversight, security controls, and explicit accountability models.

Until organisations decouple technological utility from marketing choreography, they will continue to inherit the consequences of a market where enthusiasm routinely outpaces competence.

Where the Risk Lands

The risk usually lands with the client.

If an automation sends inaccurate messages to customers, the customer relationship belongs to the business, not to the consultant. If customer data is handled badly, the reputational and compliance exposure sits with the organisation using the system. If AI-generated content misleads customers, the public-facing consequence lands on the business. If a fragile workflow breaks, the operational damage occurs inside the company, not inside the sales pitch that created it.

That is why vague AI selling matters.

The seller may collect the retainer. The client carries the consequence.

The damage may be financial, where the promised efficiency never materialises. It may be operational, where poor workflow design creates more friction than it removes. It may be reputational, where automated communication weakens trust. It may be legal or compliance-related, where data handling, consent, records, or review controls were never properly considered. It may be strategic, where a business believes it has “implemented AI” when it has only added a brittle process layer with no governance.

Language makes this worse.

Words such as “transformation”, “automation”, “agent”, “workflow”, “audit”, “AI strategy”, and “intelligent system” can sound substantial even when they remain undefined. The buyer hears modernisation. The seller sells certainty. The actual capability may be narrow, fragile, poorly governed, and dependent on tools the client does not understand.

That is the blindness problem.

One side does not know how to evaluate the claim. The other side does not know how to ground the claim. Both proceed under the protective cover of a fashionable term.

This is not progress. It is unmanaged risk wrapped in modern language.

Five Questions Before Buying AI

An SME owner does not need to become an AI engineer to avoid this trap. But the owner does need to ask disciplined questions before paying for an AI-labelled service.

The first question is: what exactly will this system do?

Not what will it unlock, scale, optimise, revolutionise, or transform. What will it actually do inside the business? Will it draft emails? Classify enquiries? Move data between systems? Summarise documents? Generate marketing content? Support customer service? Analyse sales records? Trigger follow-up actions? Produce reports for human review?

If the seller cannot answer this clearly, the offer is not defined.

The second question is: what data will it touch?

This includes customer data, employee data, financial data, operational records, emails, contracts, call transcripts, website forms, CRM entries, and uploaded documents. If the data boundary is unclear, the risk boundary is also unclear.

The third question is: what toolchain is being used?

The answer cannot simply be “AI”. AI is not a toolchain. The buyer should know which platforms, applications, integrations, models, databases, plugins, automations, and third-party services are involved. Each layer introduces cost, dependency, privacy exposure, security risk, maintenance burden, or operational failure.

The fourth question is: where does human review remain necessary?

This is not a decorative control. It is central to accountability. Some AI-supported tasks can tolerate error. Others cannot. A draft social-media caption is not the same as a client instruction, legal communication, financial recommendation, safety-relevant process, compliance record, or operational decision.

A serious provider must explain where the machine stops and human authority re-enters.

The fifth question is: who is accountable when it fails?

If the workflow sends the wrong message, mishandles data, produces inaccurate advice, creates a misleading document, or triggers the wrong action, who detects it? Who corrects it? Who carries responsibility? What safeguards were built in before the failure occurred?

A serious provider will not resent these questions. A serious provider will welcome them because they separate professional work from theatre.

From AI Label to Executive Control

This is where the Decision Before Consequence discipline becomes relevant.

The issue is not whether a business can use AI. The issue is whether AI is placed inside a decision system before it is allowed to create consequence.

DBC does not begin with the question, “How can we monetise AI?” That is the wrong starting point. It begins with more serious questions:

· What decision is being supported?

· What evidence is valid?

· What consequence follows?

· Who remains accountable?

· Where must human authority re-enter?

That is the difference between buying a label and governing a capability.

AI should not be treated as a magic commercial layer that can be pasted over a business. It must be placed inside a decision system. A tool that increases speed while weakening accountability is not progress. A workflow that produces output while hiding risk is not transformation. A service that sells confidence without defining capability is not strategy.

The future will not belong to organisations that merely “use AI”. That phrase is already too vague to carry serious meaning. The more important divide will be between organisations that place AI under disciplined control and organisations that buy AI-labelled confidence from whoever sounds furthest ahead.

The blind leading the blind is not merely a metaphor. It is a diagnostic condition. It exists when neither buyer nor seller can define the capability, boundary, risk, dependency structure, or accountability of the thing being sold.

AI under executive control is different.

It begins with the decision. It tests the evidence. It defines the risk. It maps the dependency stack. It protects the data boundary. It preserves human authority where consequence matters. It treats AI as a governed capability, not a sales label.

That is how legitimate AI should be defended.

Not by excusing the hype. Not by protecting vague claims. Not by pretending that every AI-labelled service is innovation.

Legitimate AI is defended by drawing a hard line between real capability and commercial bullshit.

Author workflow disclosure: This article was developed through an AI-assisted but human-directed workflow. AI was used for structuring, language refinement, and editorial support. The author retained responsibility for the argument, judgement, and final wording. AI-generated material was not treated as empirical evidence.

contact@rumbls.com

Terms and Conditions

Privacy Policy

Cookie Policy

© 2026 WDA Publishing. All rights reserved.

Disclaimer

Contact .