Will AI commoditise your current competitive advantage?

Competitive advantage under capability diffusion

AI changes the economics of knowledge work in a way that makes imitation cheaper. The question is less about whether AI is useful, and more about whether it erodes the scarcity that your advantage depends on.

Competitive advantage often rests on something hard to copy: tacit expertise, operational routines, access to customers, regulatory competence, or a cost position built over years. AI alters the copying cost, because many knowledge tasks can be replicated from text, examples, and feedback loops.

How does capability diffusion happen?

When a capability becomes standardised, competitors do not need to understand it deeply to adopt it. They need a workflow that can absorb it. Customer emails can be triaged, marketing copy drafted, contracts summarised, and code scaffolds generated. The limiting factor shifts from producing output to assuring quality, consistency, and compliance.

Early signals in the UK and beyond

Consider the pace at which customer support has moved from scripts to AI-assisted resolution in retail and travel, or how professional services firms have introduced AI-assisted document review. The baseline service level rises across the market, compressing differentiation. In parallel, regulators are raising expectations around transparency and control, from the EU AI Act to UK sector regulators emphasising model risk management and consumer duty outcomes.

So the uncomfortable possibility is not that AI makes everyone identical. It is that it makes a large part of what looked special become table stakes.

Advantages AI tends to commoditise

Not every advantage is equally exposed. The most vulnerable forms are those that can be expressed as patterns over data, language, or common workflows, and therefore reproduced quickly by models and templates.

Where the floor rises fastest

Advantages built on speed of producing standard content or analysis are often exposed. Examples include first drafts of proposals, routine management reporting narratives, basic market scans, or generating internal policy text. If the value proposition is largely that these artefacts exist, competitors can now match that quickly.

When expertise becomes a feature

In some markets, specialist knowledge is being productised. In-house tax guidance, HR policy interpretation, basic legal issue-spotting, or procurement negotiations can be embedded in tools and copilots. The risk is a shift from paying for expertise to paying for assurance, liability, and outcomes.

Unit economics that invite imitation

Once an organisation shows that a process can be done with fewer hours, rivals can reverse-engineer the economics. If an AI-assisted claims process reduces handling time from 20 minutes to 12 minutes, that is a 40 percent productivity gain. That improvement can be copied if it depends mostly on generic models and a common workflow design.

Counter-argument worth holding

Some leaders argue commoditisation fears are overstated because implementation is difficult: data quality, change management, and legacy systems slow diffusion. That is true in the short run. The question is how long the organisation expects friction to protect margins, and what happens when a competitor redesigns its operating model to remove that friction.

Defensible advantage in AI-native organisations

If AI makes certain outputs cheap, advantage migrates to what is harder to copy: trustworthy decisions, integrated workflows, and the institutional ability to learn faster without breaking things that matter.

Where differentiation can re-form

AI-native advantage tends to sit in the system, not the model. It emerges from how decisions are made, audited, improved, and connected to value. A competitor can buy similar model access, but cannot easily replicate a mature governance muscle, a high-quality feedback loop, or a workforce that knows when not to automate.

Examples of harder-to-copy assets

Proprietary process data: not just having data, but having labelled outcomes tied to real decisions, such as underwriting performance by segment or root causes of service failures.

Embedded assurance: evidence trails, human oversight design, and clear accountability. In regulated sectors, this becomes a commercial asset as well as a compliance requirement.

Workflow integration: AI that sits inside core systems of work, with clear hand-offs, escalation paths, and exception management, rather than a standalone assistant.

Trust and brand: not marketing trust, but operational trust, such as consistent advice quality, explainable declines, and fair treatment.

A learning analogy from education

In passing, LSI’s own AI-native learning platform is designed around mastery evidence and continuous feedback rather than time served. The point is not the technology; it is the operating logic: measurable outcomes, tight feedback loops, and governance around how judgement is formed. That logic translates well beyond education into any AI-enabled organisation.

Operating model redesign, not tools

The move from pilots to production is mostly an organisational design challenge. The difficult choices are about who owns risk, how work is redesigned, and how incentives change so that adoption becomes real behaviour.

Centralise what must be consistent

Some capabilities benefit from central ownership: model risk management, data governance, vendor due diligence, and a shared measurement framework. Consistency matters for reputational control, regulatory defensibility, and reuse economics.

Federate what must be close to work

Many use cases need domain nuance: claims, collections, HR, product operations, supply chain. Federated teams can own workflow design and adoption, provided they operate within clear guardrails and common assurance standards.

Role changes that show up in the org chart

AI-native work often creates new accountability boundaries. Examples include product owners for AI-enabled processes, control owners for automated decisions, and operational SMEs who curate feedback and edge cases. It also changes existing roles: managers spend less time compiling information, more time validating exceptions and coaching judgement.

From pilots to production cadence

Pilots optimise for learning. Production optimises for reliability. The transition requires an operating cadence: use case portfolio reviews, measurable benefit tracking, incident management, and retraining triggers when performance drifts. A practical question is whether the organisation has a defined path from prototype to controlled release, or whether prototypes accumulate as unmanaged risk.

ROI, resilience, and decision tests

Measuring AI value is easy to overstate if benefits are counted but risks are not priced in. AI-native organisations treat ROI as a portfolio question, tied to resilience and reputation, with explicit boundaries around where automation is acceptable.

Leading and lagging metrics that matter

Leading indicators: cycle time, first-contact resolution, error rate in assisted outputs, percentage of work routed through the new workflow, human override frequency, and model drift signals. These show whether the system is behaving.

Lagging indicators: cost per case, revenue conversion, customer complaints, regulatory findings, and loss events. These show whether value is accruing sustainably.

Example economics without hype

If a 500-person operations function handles 2 million cases annually at £4 per case, total direct cost is roughly £8m. A credible 10 to 20 percent reduction from AI-assisted triage and drafting is £0.8m to £1.6m per year, before platform, change, and assurance costs. That can be material, but only if rework, escalations, and risk controls do not erase the gain.

Risks that convert into reputational cost

Hallucinated advice, biased outcomes, confidentiality leakage, and unclear accountability are not theoretical. They become complaints, press coverage, regulator attention, and staff distrust. Mitigations are rarely glamorous: clear decision rights, auditable logs, red-team testing, human-in-the-loop thresholds, and strong third-party controls.

A decision test for commoditisation pressure

If a competitor had access to the same models and a competent team, could they replicate the customer-visible part of your advantage within 6 to 12 months? If yes, the advantage may already be commoditising. The remaining question is whether differentiation sits in the organisation’s learning speed and control environment, or in a product and operating system that is still hard to copy.

Uncomfortable question: if the organisation keeps optimising the current operating model and merely adds AI on top, is it silently choosing to compete on price and scale, even if the brand story says otherwise?