Total Business Impact

When a city is asked to help pay for a new stadium, the request usually arrives with an economic impact study. The study adds up what fans will spend, the jobs that spending supports, the wages those jobs pay, the taxes the wages generate and the further spending all of that sets off. The total is large, and every line in it describes something real. In 1995 the tourism and recreation scholar John Crompton catalogued eleven ways such studies go wrong. Several come down to two habits: counting money that would have been spent locally anyway, such as the evening out of a local fan who would otherwise have gone to a local cinema, and counting the same money again as it changes hands¹.

Almost three decades later, a survey of more than 130 studies of professional teams and venues confirmed what independent research had found throughout: the local economic impact is very limited, and even after adding civic pride and quality of life, the gains tend to fall well short of the public money spent². The promotional studies were not lying about any single line. They were adding lines that overlapped, against a baseline in which nothing else would have happened.

An AI business case is built the same way, often by the same instinct. One team measures faster work, another measures lower cost per unit, a third measures happier customers, and the sponsor adds them up. Each line is honest. The total may still be fiction.

The core idea

Building the AI Business Case ended with a single number, 10 million of expected benefit over three years, and AI ROI and Value Realization turned that number into a return. Neither chapter asked what was allowed into the 10 million. That question is the subject here.

Total business impact is the most credible picture of how an initiative changes the business: every material effect, across every dimension, with each economic benefit counted once. It is not the sum of every positive effect. AI rarely creates value through one metric alone, and the effects it creates depend on each other. A faster process releases capacity, which lowers unit cost, which may let the business serve customers sooner, which may keep them longer. Those are four descriptions of one chain, not four sums of money.

The European Commission’s guide to appraising public investment puts the point in one sentence: most indirect and wider effects are “transformed, redistributed and capitalised forms of direct effects,” and the appraisal has to limit the opportunities for counting them twice³. An executive’s job is therefore not to make the impact number larger. It is to build a picture whose number survives three questions: what depends on what, what would have happened anyway, and where does the money actually appear?

Six dimensions, each with its own discipline

Earlier chapters of this module examined value one dimension at a time. Total impact sets them side by side, and each brings its own measurement discipline.

Revenue covers conversion, retention, new services and pricing, and it counts only what is incremental: sales that would not have happened without the initiative, as Where AI Creates Revenue showed. Cost has to separate a real reduction from cost avoidance and from capacity that has not yet been used; only the first lowers a bill. Customer effects such as speed, accuracy and consistency should be connected to a business outcome where the evidence allows, and left unconnected where it does not. Workforce effects, such as time freed or less repetitive work, are capacity, and capacity is not money until it is used. Risk value means fewer errors, earlier detection or better compliance, and pricing it requires an assumption about both probability and impact; a case should state both. Strategic value, such as a reusable capability or faster experimentation, should be named explicitly and never used as a vague justification for a weak financial case.

Some organizations add sustainability or brand as further dimensions. That is sensible if they matter to the strategy. The list exists to make the search for effects complete, not to fix a template.

One initiative, many linked effects

A single AI deployment rarely stays inside one dimension. In a staggered rollout of a generative AI assistant to 5,172 customer-support agents, researchers measured gains in productivity, customer sentiment, escalations and agent retention together⁴. The effects were real, and they were intertwined: fewer escalations are part of why more issues get resolved per hour, and calmer customers are part of why fewer ask for a manager. A business case that priced each effect separately and added them would count overlapping gains. Total impact therefore starts by tracing which effect feeds which.

The benefits management tradition gives a practical tool for this. Ward and Daniel’s benefits dependency network draws the chain from the technology, through the changes in work it enables, to the benefits those changes produce and the business objectives they serve⁵. Drawn for an AI case, the map shows at once which effects sit upstream of others. An upstream effect is usually the mechanism or the evidence for a downstream benefit. It is rarely an extra benefit of its own. The map also shows where the chain is long: the further an effect sits from the change itself, the more assumptions stand between them, and the more cautiously it should be treated.

Direct or indirect, counted or described

Two distinctions sort the effects before anyone adds anything. The examples here come from the illustrative chemicals plant whose story closes the chapter. The first is direct versus indirect. A direct effect shows up almost at once in a measurable result: a chemicals plant that changes over its reactors less often buys less cleaning solvent. An indirect effect moves an intermediate metric that may later contribute to value: shorter lead times may make customers more loyal, which may raise reorders. Every arrow in that chain is an assumption.

The second is quantitative versus qualitative. Revenue, cost, hours, defects and retention can be expressed as numbers; audit readiness, team learning and flexibility are harder to price. Qualitative does not mean zero, and it does not mean free to invent a figure either.

Put together, the two distinctions give four treatments. Direct and quantitative effects are counted. Indirect quantitative effects are modeled, with each assumption written down and tested where possible; Baselines, Metrics and Measurement described the comparisons that can test them. Direct qualitative effects are described with evidence. Indirect qualitative effects are tracked through indicators until they become something more.

A benefit can be spent once

Double counting is claiming the same economic benefit more than once. It enters a business case in three ways.

The first is alternative uses of one resource. Capacity released by an AI system can absorb growth, cut overtime, avoid a planned hire or shorten response times. These are choices about the same hours or machine time, not additions. Capacity spent avoiding a hire is not also available to cut overtime.

The second is relabeling. The same freed capacity can be written once as a productivity gain, once as avoided hiring and once as lower cost per unit. Lower cost per unit is a ratio: the same cost spread over more output. It is evidence that the benefit is happening, not a further sum of money.

The third is capitalization: counting both an asset and the income it produces. The Commission’s guide gives the classic example. An irrigation project raises the value of farmland and the income from farming it, but only one may be counted, because the owner can sell the land or keep it and take the income, not both³. The AI equivalent is a case that claims both a one-time release of working capital and the yearly carrying cost of the same inventory.

Behind all three sits the counterfactual, the stadium studies’ blind spot. A benefit is the difference between the world with the initiative and the world without it, as Baselines, Metrics and Measurement argued. If the operating plan already included the growth that freed capacity will now serve, the margin on that growth is not new; the benefit is whatever the plan would otherwise have spent to serve it.

Primary, distinct, secondary, indicator

Counting once does not mean claiming only one thing. An initiative can have several genuine benefits. The test is whether a second line is distinct: it saves or earns a different resource, it would still exist if the first line were zero, and it lands on a different line of the accounts, the budget-line test from Where AI Reduces Cost. A line that fails any of the three is the same benefit under another name.

That test gives every effect one of four roles, with strategic value named alongside them. The primary benefit is the financial benefit the case stands on, measured where the money appears. A distinct benefit passes the test and is counted separately. A secondary outcome is a real or plausible effect, such as customer loyalty, that is reported but kept out of the total until a controlled comparison shows its size⁶. A supporting indicator, such as cost per unit or hours saved, shows that the counted benefit is happening and is never added to it. Guardrail metrics, covered in Leading vs Lagging AI Metrics, sit alongside them to make sure speed is not bought with quality.

Story: two drafts at a chemicals plant

The case that follows is an illustrative composite; the company is unnamed and the figures, all over three years and in millions, are invented to make the counting visible. They are a separate illustration, not the base case of earlier chapters.

A specialty chemicals maker runs a plant whose reactors make dozens of products in batches. Every switch between products means stopping, cleaning and restarting, and the plant was running close to full. Its plan included an expansion of one reactor line, costing 8 million, to serve expected growth. An AI scheduler now sequences the batches so that similar products follow each other, which cuts the number of changeovers. Before the investment committee met, two drafts of the case were written from the same facts.

The sponsor’s draft gathered a line from every team. Sales valued the extra volume the freed reactor time could produce at 9 million of margin. Engineering counted the avoided expansion at 8 million. Finance’s cost model showed cost per tonne falling, worth 6 million. Operations added 1.5 million of solvent and energy no longer used in cleaning. Customer service expected shorter lead times to lift retention by 3 million, and planning valued its saved time at 0.5 million. The total was 28 million. No line was dishonest; each team had measured something real.

The finance lead’s draft asked one question of every line: where does the money appear, compared with the plan? The growth was already in the plan, to be served by the expansion, so its margin was not new; the benefit was the 8 million the plan would otherwise have spent. That became the primary benefit. Cost per tonne was the same freed capacity expressed as a ratio, so it became an indicator. Solvent and energy passed the distinctness test, since it is a different resource, on a different budget line, that would have been saved even with the expansion, and it was counted. Retention became a secondary outcome with a test: compare the reorder rates of customers whose lead times shortened with a matched group whose did not. Planners’ time was capacity, not cash. Off-spec batches at changeover became a guardrail, and the chance to reuse the scheduling model at the company’s other plants was named as strategic value.

The finance draft counted 9.5 million: 8 million of avoided expansion and 1.5 million of solvent and energy, set against the scheduler’s full cost in the way Building the AI Business Case describes. The committee approved it. Eighteen months on, the expansion had not been needed and solvent spend had fallen in line with the case. The retention test had found no measurable difference in reorders yet, so that line stayed out. Had the committee approved the sponsor’s draft, its review would have been explaining a gap of 18.5 million. Instead it saw what it had been promised and funded the scheduler’s rollout to a second plant. The facts were the same in both drafts. The difference was the discipline of counting each benefit once.

What this means for leaders

The value of an AI initiative is not the biggest number that can be attached to it. Leaders should expect every case to show its map of effects, not just its total, and should treat a long list of benefit lines that all trace back to the same freed capacity as a warning rather than a strength. The most useful single question is the finance lead’s: where does the money appear, compared with what we would have done anyway? A large total does not by itself make an initiative a priority, either. Confidence, timing and the risk of delivery matter as much as size; ranking initiatives against each other belongs to the portfolio discipline of Module 09.

Two habits make the discipline stick. Insist on the shared vocabulary of primary benefit, distinct benefit, secondary outcome and indicator, so that teams stop arguing about whether an effect is real and start agreeing on its role. And reward the smaller, defensible number. A sponsor who sees inflated totals approved will keep producing them.

Check yourself

1. Total business impact is the sum of every positive effect an initiative produces.MythFact
2. An initiative can legitimately report two financial benefits that come from the same change.MythFact
3. Lower cost per unit is usually an additional benefit on top of the capacity it reflects.MythFact
4. A benefit should be measured against what the business would have done without the initiative.MythFact
5. A plausible retention uplift can go into the total as long as it is clearly labeled.MythFact
6. Strategic value that cannot be credibly priced should be left out of the case.MythFact

What comes next

A credible total impact still has to be read and acted on by people who will not open the model behind it. The next chapter, AI Value Scorecard, puts the primary benefit, the indicators, the guardrails and the evidence onto one page an executive can decide from.