Build an AI adoption business case around a decision someone can authorize. Name the work that should improve, compare credible alternatives, show the full cost and explain who will turn any improvement into a useful outcome. Ask for a bounded commitment with evidence requirements and a review point.
For an adoption lead working with finance, the useful output is a proposal whose assumptions can be challenged. A large return-on-investment figure is less useful if nobody can explain how saved minutes become capacity, lower spending or better service.
Start with the decision and the alternative
Write the funding question before the benefits slide. You might be asking for a limited trial in sales support, continued funding for an established workflow or shared support across several departments. Those requests need different evidence and commitments.
Describe what will happen without the proposal. Include expected demand, existing process improvements and contracts already in place. A baseline that assumes today's problems continue unchanged can make almost any intervention look attractive.
HM Treasury's business-case guidance uses business as usual as a comparison and considers alternative options. It is UK public-sector guidance, but the comparison principle is useful in a company too. Your case should explain why the proposed approach is preferable to a smaller, cheaper change.
Suppose Jamie Lee manages sales support at a distributor. The team prepares quotations from customer quantities, an approved price list and stock records. Jamie wants customers to receive accurate quotations sooner. Every quotation must still have the correct price, availability and terms before a salesperson sends it.
| Option | What changes | What Jamie must establish |
|---|---|---|
Continue the current process | Existing tools, staffing and planned improvements remain | Expected demand and the current delay |
Improve the process without AI | Repair the quotation template and access to approved records | Whether those changes remove the bottleneck cheaply |
Add a supported AI workflow | Approved drafting, staff practice and human checking | Additional benefit beyond the simpler changes, including review effort |
If better source records are necessary for both improvement options, include their cost consistently. Do not assign all cleanup costs to one option while giving the other its benefits for free. If the AI proposal includes coaching and a new template, describe the whole package being funded.
Give each benefit an owner and a route to realization
A business case needs someone who can make its expected benefit happen. Jamie can reorganize quotation work; finance can agree how costs are treated. Neither responsibility belongs to the software alone.
In a public discussion about reporting AI value, a Reddit user described connecting each initiative to a workflow, a payback period and a profit-and-loss line. That is an unverified account of one person's practice, rather than evidence of financial results. Its useful detail is the connection between a local change and a budget decision.
Not every worthwhile benefit will release cash. The UK Government Efficiency Framework distinguishes reductions in spending from improvements such as staff time available for other work. Keep that distinction explicit in your own proposal.
For Jamie's quotation team, record the intended benefit in one of these forms:
- Available capacity. Staff spend less total effort on accepted quotations. Jamie assigns the usable time to waiting requests and checks whether the queue shrinks.
- Avoided spending. An external support invoice becomes smaller. The contract owner confirms which charges can actually be removed and when.
- Better service. Customers receive correct quotations sooner. The service owner tracks turnaround and errors, without assuming that every faster quotation wins a sale.
Attach an owner, evidence source, timing and dependency to each benefit. If the same hour supports more quotations, do not also count its salary equivalent as money removed from payroll. If additional sales are part of the case, explain how they will be distinguished from changes in prices, demand or sales activity.
Show which assumptions change the funding decision
Keep observed results separate from estimates. Before a pilot, adoption and time savings are assumptions to test. After a pilot, the measured result still needs its population, work conditions and uncertainty attached.
For Jamie's proposal, suppose the team expects 1,200 comparable quotations a month, with costs valued in US dollars. The following scenarios use assumed uptake and net time savings after checking and correction. They illustrate how to expose uncertainty, not how much any particular tool will save.
| Monthly assumption or result | Low | Base | High |
|---|---|---|---|
Share of quotations using the assisted workflow | 40% | 60% | 80% |
Net minutes released per assisted quotation | 2 | 4 | 6 |
Total hours released | 16 | 48 | 96 |
Share Jamie can put to useful work | 50% | 75% | 75% |
Usable hours | 8 | 36 | 72 |
Capacity equivalent at an assumed $40 per hour | $320 | $1,440 | $2,880 |
The base calculation is 1,200 quotations × 60% × 4 minutes ÷ 60 = 48 hours. Applying the assumed 75% usable share gives 36 hours. The resulting capacity equivalent does not establish a cash saving.
The usable share matters because scattered spare minutes may not fit the next job. Jamie must explain how work will be reassigned, whether enough suitable requests exist and whether another approval queue will absorb the gain. Use the net-time measurement guide to count the whole workflow before estimating this next step. Do not subtract review effort twice if it is already included in the net saving.
Now suppose the proposal requires:
- $900 in additional monthly cash spending.
- $3,600 in one-off cash setup.
- 80 hours of existing staff time for preparation and practice.
Keep the staff commitment visible even if it does not increase payroll. The first three months require $6,300 in additional cash, plus those staff hours, before any verified cash benefit.
The Digital and Data Benefits framework recommends sensitivity analysis around assumptions such as adoption, productivity and cost. Use the scenarios to identify the weak point in your own decision. If approval depends on nearly every eligible quotation using AI, test that assumption before buying access for the whole company.
Include licensing or usage fees, integration, data preparation, training, support, evaluation, security review and ongoing maintenance where relevant. Identify which are incremental cash costs and which consume existing staff capacity. Use one time period and currency, state the source of each estimate, and allocate shared costs once.
Ask for the next commitment the evidence supports
A weak estimate can justify a bounded learning investment if the uncertainty matters and the test is affordable. It cannot justify presenting an unobserved benefit as a result.
For Jamie, the request could be a capped trial of quotation drafting, with the template-only approach as a credible comparison. Agree with an analyst what observations and duration are needed. The credible-pilot guide explains why a fixed number of weeks is not enough to establish a reliable effect.
Put these decisions in the authorization:
- Scope and limit. Which quotation types, people and approved data are included, and what is the maximum cash and staff commitment?
- Acceptance and pause conditions. Who checks prices, stock information and terms, and what failures require stopping the workflow?
- Evidence for renewal. Which quality, total-effort and service results would justify further spending, and who evaluates uncertainty?
- Accountability and exit. Who owns the benefit, the operating work and the next decision? What can be cancelled, and what support remains necessary if funding ends?
For a company-wide adoption program, combine the workflow cases with the shared work that makes them possible. A common training or support team may serve several departments. Show its cost once, explain how the departments use it and avoid adding overlapping benefit claims together. Stage the commitment where possible so untested demand does not become a permanent budget by default.
Bring the proposal to finance with the assumption most likely to change the answer highlighted. Ask the workflow owner what would prevent the benefit from being realized. Resolving that question is more useful than polishing the optimistic percentage.
Questions about an AI adoption business case
What should an AI adoption business case include?
Include the business problem, proposed scope, credible alternatives, full costs, expected benefits and the evidence behind them. Name the person responsible for realizing each benefit and specify the funding decision, limits and review point. A proposal for quotation drafting, for example, should state who checks prices and terms, what improvement matters and when further spending will be reconsidered. Use the authorization checklist to turn the analysis into a decision.
Can time saved be counted as AI ROI?
Time saved can support a capacity valuation when the estimate includes checking and correction and the time can be used productively. It is not automatically a cash return. For a cash-based return-on-investment calculation, identify actual incremental financial benefits and costs over the same period, with finance agreeing their treatment. If the benefit is faster service with unchanged spending, report that outcome and its cost explicitly. The quotation scenarios show why capacity and cash should remain separate.
Can we seek funding before we have pilot results?
Yes, for a bounded test whose cost and uncertainty are clear. Label expected uptake, productivity and business benefits as assumptions; specify what the test must establish before expansion. Include quality checks, a credible comparison and a person able to stop the work. Where AI would enable a genuinely new service, a same-task time baseline may not exist. Test demand, feasibility and delivery cost rather than inventing a manual process that nobody currently performs.
How should we calculate payback for an AI program?
For a simple cash case with a one-off setup cost and stable monthly flows, divide setup cash by the positive monthly net cash benefit after recurring costs. That shortcut is unsuitable when benefits ramp up, costs change or the net benefit is zero or negative. Build a period-by-period cash forecast in those cases, including commitments that cannot be cancelled. Keep capacity equivalents out of cash payback unless a specific spending change is supported, and ask finance to apply the organization's investment-appraisal method.
