AI investment raises a question that adoption figures cannot answer. What will the organisation do better than its competitors, why will that difference matter, and how long could it last? A portfolio of successful deployments can improve performance without changing the organisation’s competitive position. That can still justify investment, but it requires a different strategic argument.
Consider two banks using similar AI tools to help employees prepare client meetings. Both reduce preparation time and improve the consistency of their materials. Those benefits matter. Yet if their competitors can obtain the same tools and achieve comparable results, the improvement may become an expectation rather than a reason for clients to choose one bank over another.
Competitive necessities protect the organisation’s ability to participate, while differentiating capabilities improve its position relative to an alternative. Both deserve funding when their economics justify it. The resource-based view of the firm provides a familiar test for the latter, asking whether valuable assets are difficult to reproduce and can be put to work effectively. AI makes that test more demanding as widely available models accelerate imitation and operational use creates opportunities for proprietary learning. The lending case that follows examines whether a bank can turn that learning into profitable service for a segment it currently struggles to reach.
Start with the basis of competition
Strategic advantage begins with understanding why customers choose the organisation and what enables it to serve them profitably. Depending on the business, that may involve better advice, faster service, dependable execution, distinctive products, or a superior understanding of risk. AI deserves strategic attention where it can materially strengthen one of those factors.
An illustrative bank seeks to serve small businesses with seasonal cash flows. Its current lending process relies heavily on annual accounts and manual assessment. Some viable borrowers are too expensive to assess for the loan size, while others wait long enough to choose a faster provider. The strategic opportunity is to make this segment economically serviceable without relaxing credit standards.
The alternative is not necessarily another bank. An embedded lender connected to a merchant platform might recognise a funding need through sales activity and present an offer within the customer’s existing workflow. Its distribution and acquisition economics would challenge a bank that measures itself only against traditional lenders. The bank must compare its proposition with that combination of speed, access, and convenience.
In this example, the bank has a potential information advantage through permissioned account history, recurring receipts, payment obligations, and observed repayment outcomes. A cash-flow model helps assess the amount and timing of affordable credit. Integration with lending operations allows suitable applications to proceed quickly, while uncertain cases receive further review. The proposed advantage is profitable access to a previously underserved segment through a timely, appropriately sized offer.
The investment case needs measurable targets covering application-to-decision time, assessment cost per funded loan, customer acceptance, and net contribution after funding, credit losses, and servicing. Validation must compare these outcomes with the current process and relevant alternatives. More approvals alone would provide weak evidence, particularly if losses or customer harm increased.
Identify what competitors cannot readily reproduce
The bank’s potential advantage rests on longitudinal information, calibrated risk judgement, and the ability to deliver within a controlled operation. None is presumed exclusive. A platform may have better sales information, and a rival may obtain account data with customer permission. The bank has to establish where its combined capability is superior and how long that difference is likely to last.
Distribution remains its weak flank. The embedded lender already reaches the customer within a commercial workflow, while the bank must secure attention and complete the application. Existing account relationships provide a possible route, but their acquisition economics and customer response need testing. A platform partnership is another option, with revenue sharing, data rights, and dependence on the partner included in the assessment. Better underwriting will not establish an advantage if the bank cannot reach borrowers at an economic cost.
Possessing extensive data is insufficient. The information must be useful for the decision, accessible under appropriate permissions, and connected to reliable outcomes. An institution with decades of records may struggle to develop an advantage if those records are fragmented or provide little evidence about which actions worked. Another institution with less information may be better positioned if it can use that information consistently and learn from the result.
Domain knowledge also matters because it determines how evidence is interpreted and what can be done with it. In lending, recognising a pattern is only part of the capability. The bank must understand the customer’s circumstances, assess the opportunity within its risk appetite, and deliver an appropriate product. Competitors may obtain comparable models while still finding that combination difficult to reproduce.
The ability to deploy under banking controls belongs in this assessment. Independent validation, documented decision policies, fairness assessment, monitoring, and reliable handling of exceptions are demanding operating capabilities. A bank that performs this work effectively has a stronger route from analytical insight to responsible lending at scale.
Regulation itself does not confer an advantage, and an incumbent’s control processes are not automatically superior. A fintech can build equivalent capabilities or partner with an institution that supplies them. The strategic asset is the demonstrated ability to deliver good outcomes under scrutiny, with proportionate controls that support adoption rather than create avoidable delay.
This assessment requires scepticism. An established relationship or a large transaction base can support an advantage, but neither guarantees one. Leaders need evidence that the asset contributes to better outcomes and a credible view of how rivals might respond. The relevant question is whether competitors can recreate the complete capability within a commercially meaningful period.
Make deliberate choices about what to own
Ownership decisions follow the elements that determine distinctive performance, with flexibility retained elsewhere. A generic model is a fragile basis for differentiation when rival vendors can offer comparable capabilities. Improvements in widely available models can turn today’s technical lead into tomorrow’s standard feature. An investment case must survive that possibility.
For the lending example, the distinctive elements include permissioned context, cash-flow features, evaluation against relevant borrower outcomes, decision policies, and operational integration. The bank can obtain general-purpose language capabilities from suppliers while retaining control of these elements. A specialised predictive model deserves internal development when it demonstrably strengthens credit assessment and the bank has the capacity to maintain it.
These choices require separating the functions involved. A language model helping assemble application information and a predictive model assessing repayment risk perform different tasks. Better document processing does not establish a credit advantage. Each ownership decision depends on the contribution the component makes to the complete capability.
Owning the lending model commits the bank to ongoing validation, monitoring, redevelopment, and the specialist skills those activities require. Those costs belong in the economics of serving the segment. Internal development earns its place when the additional control and performance justify that commitment, while supplier arrangements need to preserve access to the information and decisions on which the bank’s advantage depends.
The choice turns on advantage, control, delivery speed, and lifetime economics. For this bank, protecting its cash-flow assessment and lending policies matters because those capabilities determine whom it serves and at what risk. Owning every underlying technology adds little unless it strengthens that result.
Build a capability that improves through use
An initial performance lead is vulnerable to imitation. Sustaining it requires learning from operational use and turning that evidence into better decisions. The organisation needs to capture outcomes, assess the effect of its interventions, and translate that evidence into better models, policies, and workflows.
In the lending example, repayment, utilisation, seasonal cash-flow changes, and the outcomes of servicing actions provide evidence for improving offers and risk assessment. The bank needs to test whether that evidence improves performance in subsequent borrower cohorts and whether the benefit persists after operating costs. The learning mechanism needs to be measured as carefully as the initial deployment.
That process requires care. A bank that investigates only cases flagged by a model may learn a great deal about those cases while remaining uncertain about what it missed. A lending model trained on previously approved borrowers may provide incomplete evidence about applicants who were declined. Feedback becomes strategically useful when its limitations are understood and its quality is actively managed.
Learning also depends on the organisation acting on what it discovers. Product owners, domain specialists, data teams, and operational leaders need a shared way to assess results and make changes. Otherwise, useful evidence accumulates without improving the capability. Effective governance enables appropriate adjustment while retaining oversight of consequential decisions.
The investment case must explain how the proposed capability will become more valuable over time. That might involve better outcome information, more effective interventions, or wider application of a proven method. Claims that advantage will compound simply because a system uses AI deserve scepticism.
Direct investment towards a defensible position
An effective portfolio contains both competitive necessities and opportunities for differentiation. Each investment needs to be judged against its purpose. A broadly available employee assistant deserves scrutiny of adoption, work quality, and realised productivity. The lending proposition requires evidence that the bank can serve the target segment more effectively and profitably than the alternatives.
Before funding an investment presented as strategic, the committee needs answers to four questions. Who would the customer otherwise choose? What measurable performance gap will we close? What prevents a rival from reproducing the improvement? How will operational use make the capability better? In the lending case, those answers must cover distribution economics as well as credit performance, and acknowledge the limits of the outcome information available.
The final challenge is a competitive response scenario. What happens if a rival gains comparable model capabilities, secures access to similar data, or reaches the customer through a cheaper channel within 18 months? That period is a stress test rather than a forecast. Its purpose is to expose whether the proposed advantage rests on a durable capability or a temporary feature.
The answer determines whether the strategic claim holds. A readily copied improvement can still warrant investment as a competitive necessity. A claim of strategic advantage requires evidence that the organisation can establish a lead, exploit it economically, and continue strengthening its position as rivals respond.
