Example: AI Business Case

Prepared for the board of Meridian Brokers, a commercial insurance brokerage with 92 staff across two offices, to answer one question: should we adopt AI across the business, and what do we get for the money?

1. Executive summary

Meridian's people spend a large share of every week producing documents from other documents: claim summaries from adjuster reports, submission packs from client files, renewal comparisons from insurer responses, and a very large volume of explanatory email. This is precisely the category of work current AI tools do well, and it is spread across every department rather than concentrated in one.

We assessed 14 candidate workflows. Six are worth doing now, four are worth doing later, and four we recommend you do not attempt with AI at all. The six recommended workflows account for an estimated 228 hours per week of current staff effort across 92 people, or about 2.5 hours per person per week.

Applying a deliberately conservative 40% realisation factor (assuming most recovered time is absorbed rather than converted to output), the estimated year-one value is $251,700 against a total year-one cost of $43,200. That is a payback period of roughly 2.1 months.

The material risk is not cost or accuracy. It is uncontrolled adoption: staff are already using consumer AI accounts with client data, outside your policies, with no audit trail. That is happening today and is the strongest argument for acting now rather than next financial year.

2. What we looked at

We did not start with the technology. We started with a walk through the working week of each department, asking what people actually produce, what they wait on, and what they redo. The assessment covered:

• Broking (28 people): submission preparation, market comparison, renewal reports
• Claims (19): file review, insurer correspondence, claimant updates
• Client Services (14): inbound email triage, policy questions, certificate requests
• Finance (8): reconciliation commentary, month-end reporting narrative
• Operations & Compliance (9): procedure documents, file audits, regulatory updates
• Marketing (5): content production, case studies, campaign copy
• Leadership & Administration (9): board papers, meeting notes, internal comms

We also reviewed your existing footprint: three departmental subscriptions bought on personal cards, an unknown number of free consumer accounts, and no admin visibility over any of it.

3. Where AI pays off at Meridian

These are the six workflows we recommend for the initial rollout, ranked by value and ease. Hours are current effort per week across the whole team performing that work.

A further four workflows (proposal design, training material, tender responses and new-starter onboarding packs) are good candidates, but they depend on the first six bedding in. We have deferred them to a phase two rather than inflating the first business case with them.

4. Where AI will not help you

This section matters as much as the last one. Four things came up in discovery that we recommend you rule out now, so nobody spends six months discovering it the expensive way:

• Automated claims decisioning. Regulatory exposure and the cost of being wrong are both far too high. AI can summarise a claim file; it should not decide the claim.
• Direct-to-client chatbots. Your client relationships are the product. A bot answering coverage questions creates liability and erodes exactly what clients pay you for.
• Premium calculation or rating. This belongs in deterministic systems that produce the same number twice. Language models do not.
• Replacing headcount in year one. The realistic outcome is the same team absorbing growth without adding roles, not a reduction. Any business case built on immediate redundancies will not survive contact with reality.

5. What it costs

Full year-one cost, including our fee and your own people's time. Licence pricing is the blended per-seat business tier rate for the provider recommended in the accompanying plan.

Note that 14 of your 92 staff are not licensed in year one. Warehouse, reception and part-time field roles do not have workflows in scope, and licensing them anyway is the most common way this spend quietly doubles.

The $600 monthly usage ceiling is a hard limit configured at the provider, not a budget you hope holds. It cannot be exceeded without an administrator deliberately raising it.

6. What it returns

We have modelled this conservatively on purpose. The board should be able to defend the number under challenge, not be embarrassed by it in twelve months.

The 40% realisation factor is the assumption most worth arguing about, so we have made it explicit rather than burying it. Even at a severe 15%, meaning six of every seven hours saved simply evaporate, the return is $94,400 against $43,180, and the project still pays for itself inside six months.

There are two further benefits we have deliberately excluded from the numbers because they cannot be defended with a calculation: faster turnaround on submissions during renewal season, and the recruitment and retention effect of not being the brokerage that still does everything by hand.

7. Risks, and how each one is controlled

8. The risk of doing nothing

Deferring this decision is not a neutral act. During discovery we confirmed that staff in at least three departments are already pasting client information into free consumer AI accounts. That is unsanctioned, unlogged, outside your data policies, and it continues every day the decision is deferred.

The choice in front of the board is not whether AI is used at Meridian. It is whether it is used inside a governed setup with spending limits and an audit trail, or outside one.

9. Recommendation

We recommend Meridian proceed with a full implementation across the six workflows identified, on a single governed provider, with hard spending limits and SSO, delivered within 30 days.

The specific decision we are asking the board to approve:

• A year-one budget of $43,180, of which $35,280 is recurring
• 78 licensed seats, reviewed at 90 days before any expansion
• A nominated executive owner (we suggest the COO) and an IT contact for the identity work
• An implementation window with a fixed end date, not an open-ended evaluation

The exact setup this budget buys, covering provider, models, controls, permissions, security settings and the week-by-week rollout, is set out in the companion document.