From AI capability to measurable economic value

Define the workflow, its volume, its current cost and the quality standard required. Identify the budget holder and the operational owner. Record whether the problem creates a direct expense, a delay, a constraint on capacity or a measurable loss of revenue.

The comparison must use equivalent work. An AI-assisted process that completes more tasks but requires materially more checking is not delivering the same result at the headline speed.

For a bounded test, an 8-week pilot could examine a 15% reduction in handling time at an equivalent quality standard. Both figures are illustrative test-design assumptions. The result must be measured against the baseline after human review and exception handling; neither is an achieved Westminster saving.

The software price is only one input. The analysis should include implementation, integration, data preparation, training, recurring model or infrastructure use, oversight, support and the cost of errors.

Human review should either reduce the net time saved or appear as an incremental cost. It should not be counted twice. Existing costs that continue unchanged are different from costs the customer can actually avoid.

In the anonymised dollar-denominated client case on this page, $30,000 is the full incremental cost of year one, including setup and recurring delivery—not simply a software licence price. Showing the complete cost makes the $6,000 net benefit and 20% customer ROI meaningful.

Time released can be valuable without reducing payroll. The case for cash savings requires an avoided cost. The case for redeployment requires a credible use for that capacity and evidence of the resulting economic benefit.

Additional revenue should be translated into the relevant incremental profit after delivery costs. Do not add both the value of released capacity and revenue generated by that same capacity without reconciling the overlap.

This is a deliberately hypothetical example to explain the method. It is not a result achieved by Westminster, a customer or a portfolio company, and it is not a fund return.

Customer ROI here means (incremental benefit − incremental cost) ÷ incremental cost. Base payback is £6,000 divided by the £1,000 monthly net benefit.

The illustration assumes setup is paid at inception, full benefit from the first month, constant workload and costs, and consistent VAT/tax treatment. It excludes a ramp-up period and discounting. A real assessment should test both. Where benefits are noncash, the result is economic payback—not cash payback.

The same product changes from a positive case to a negative one when less capacity becomes usable economic benefit and delivery costs rise. That is why adoption and implementation deserve as much attention as model performance.

The customer’s return supports a purchasing decision. It does not establish the supplier’s profitability or defensibility.

Evidence to assess:

• Revenue quality: paid versus unpaid activity, recurring versus implementation revenue, collections and concentration.
• Retention: comparable customer cohorts, renewal and expansion, reasons for loss and dependence on a small number of large customers.
• Delivery margin: model and compute consumption, third-party costs, customer support, human intervention and integration work.
• Distribution: the actual buyer, procurement time, acquisition channels, channel concentration and the cost of acquiring and supporting a customer.
• Capital efficiency: cash burn, runway, working-capital needs and the capital required for the next meaningful milestone.

Metric definitions should remain consistent between periods. Growth that relies on changing the definition is not comparable growth.

Company progress only reaches an investor through the security and ownership position held. Entry price, dilution, liquidation preferences, follow-on funding and exit timing can materially change the result.

A base case should state the operational milestones and financing required. A downside case should include delayed adoption, higher costs, a difficult financing environment and the possibility of loss. An upside case should not substitute for a survivable funding plan.

$500,000 invested is not the same as $500,000 of value created

In a hypothetical priced-equity illustration, $500,000 invested at a $10m fully diluted post-money valuation represents 5% ownership at entry. The same cheque does not guarantee that ownership after future financing; dilution, security rights and subsequent capital requirements must be modelled. This is arithmetic, not an actual Westminster holding.

A useful investment memorandum records what would invalidate the thesis and what evidence would justify more capital. It should also explain why declining the investment may be the right decision.

More capital is a separate decision. Under the proposed fund design, the $6m follow-on reserve would need to support the portfolio as a whole; the $1.5m total cost-exposure ceiling per company is a limit, not an entitlement. Each follow-on would require renewed assessment of value, risk and remaining capital. These are proposed allocation parameters, not existing assets or committed follow-on funding.

The source and date of each material input; observed results versus estimates; assumptions and sensitivities; unresolved questions; the actual security and terms; conflicts; capital requirements; and a responsible owner for follow-up.

That turns a persuasive narrative into a decision that can be revisited when the facts change.

Discuss our investment approach