How Do We Prioritise Competing AI Opportunities in Delegated Authority?

Most delegated authority teams do not struggle to find ideas for AI. They struggle to choose between them.

Once oversight teams, operations leads and underwriters start discussing what AI could do, the list grows quickly: bordereaux validation, coverholder query triage, anomaly detection in reporting, automated committee summaries, and more. Each idea has genuine merit.

The real challenge is that budget, technical capacity and change management bandwidth are all limited. Pursuing every opportunity at once usually means none of them get done well.

This article sets out a practical way to compare competing AI opportunities in delegated authority and decide what to tackle first, without relying on guesswork or internal politics.

The Growing List of AI Possibilities in DA

Once a DA team starts exploring AI, ideas tend to multiply.

Bordereaux validation is an obvious candidate, given the manual effort involved. But so is flagging anomalies in monthly reporting, triaging routine coverholder queries, generating oversight summaries for committee packs, and monitoring compliance against binder terms.

Each of these represents a genuine operational pain point. None of them is wrong to consider.

The difficulty is that identifying opportunities is a different task from prioritising them. A long list of plausible ideas does not tell a team where to start, how much capacity each option will consume, or which will deliver value quickly enough to justify continued investment. Without a structured way to compare them, teams can end up pursuing the loudest request or the most fashionable idea rather than the one that will do the most good.

How DA Teams Traditionally Choose What to Tackle First

Before introducing a more structured framework, it is worth recognising how these decisions typically get made today.

One common approach is squeaky-wheel prioritisation: whichever stakeholder complains most persistently gets their problem addressed first. This has the advantage of responding to genuine frustration, but it can just as easily reward volume of complaint over actual operational impact.

Another common approach is cost-based ranking, where the cheapest or fastest-to-build option is chosen first simply because it can be delivered with minimal friction. This can produce quick wins, but those wins are not always meaningful ones.

A third approach is volume-based triage: tackling whichever process handles the most transactions, on the assumption that bigger volume automatically means bigger benefit. This is a reasonable starting instinct, but it ignores whether the underlying data and processes are actually ready for automation.

Each of these traditional approaches contains a useful signal, but none of them, used alone, reliably identifies the best opportunity. They need to be combined and weighed against each other deliberately.

A Framework for Comparing AI Opportunities

A more reliable approach scores each candidate opportunity against three consistent criteria: operational impact, feasibility and strategic fit.

Operational impact asks how much time, cost or risk the initiative would genuinely remove, and how many people or processes it would affect. A high-impact opportunity touches a meaningful volume of work or addresses a recurring source of error or delay.

Feasibility asks how realistic successful delivery actually is. This includes data readiness (is the underlying information consistent enough to work with), the complexity of the change required, and the availability of the right stakeholders to support it. An idea with excellent potential impact but poor data quality or unclear ownership may not be feasible in its current form.

Strategic fit asks whether the initiative aligns with wider organisational priorities, such as regulatory expectations, coverholder relationship management or planned system changes. An initiative that supports a broader strategic direction is generally easier to gain sponsorship for and sustain over time.

AI can genuinely help at this stage, not by making the decision, but by analysing available data across candidate processes (volumes, error rates, data structure consistency) to give teams an objective, evidence-based view of feasibility and impact rather than relying purely on anecdote. The final judgement, weighing these scores against organisational context and risk appetite, remains with experienced DA professionals.

Scoring each opportunity against these three criteria, even informally on a simple grid, usually reveals a clearer picture than intuition alone. The best first project is rarely the opportunity with the highest impact score in isolation. It is the one that combines meaningful impact with realistic, achievable delivery.

Sequencing and Governing Ongoing Prioritisation

Once a first opportunity is identified, sequencing becomes the next consideration.

Early initiatives should be chosen partly for their ability to build organisational confidence. A well-delivered first project, even a modest one, demonstrates that AI can be applied responsibly and effectively within DA, making it considerably easier to gain support for more ambitious initiatives later.

This does not mean permanently avoiding harder problems in favour of easy wins. It means sequencing deliberately: address a credible, achievable opportunity first, then use the capability and trust that builds to tackle more complex initiatives.

Prioritisation decisions should also involve the right mix of stakeholders. Operational teams understand where the pain is genuinely felt day to day. Technical and data specialists understand what is realistically feasible given current data quality and systems. Decisions made by only one group tend to miss something important.

Finally, prioritisation is not a single decision made once and then forgotten. Data quality improves, coverholder relationships change, regulatory expectations shift and organisational capacity evolves. Revisiting the prioritised list periodically, as part of ongoing governance rather than a one-off planning exercise, ensures that the sequence of initiatives continues to reflect current reality rather than assumptions made months earlier.

Example

A London managing agent's DA oversight team has identified four candidate AI opportunities: automating bordereaux validation across twelve coverholders, flagging anomalies in monthly reporting, triaging coverholder queries, and generating oversight summaries for committee packs. With limited change capacity, they must choose where to start.

Using an impact-versus-feasibility framework, the team recognises that bordereaux validation offers the highest combined score: it affects the most coverholders, the underlying data is relatively well structured, and success would be visible and easy to communicate internally. They prioritise this first, deferring the oversight summary automation until the validation project has demonstrated value and freed up team capacity.

FAQs

• Should we always start with the highest-volume DA problem?

  Not necessarily. Volume alone is not a reliable prioritisation criterion. A high-volume problem with poor underlying data quality or unclear ownership can be harder to solve successfully than a smaller problem with well-structured data. Feasibility and operational impact need to be weighed together, not volume in isolation.

• How many AI opportunities should we pursue at once?

  Most DA teams are better served by focusing on one or two initiatives at a time initially. This preserves change management capacity and allows early projects to demonstrate value clearly, which builds the internal confidence needed to expand scope later.

• Who should be involved in prioritising AI opportunities in DA?

  Both operational and technical stakeholders should be involved. Operational teams understand where the genuine pain points lie, while technical or data specialists can assess whether an initiative is realistically feasible given current data quality and systems. Leaving the decision to only one group risks missing important context.