Practical guide
Draft an investment downside case with AI
Turn fictional costs, adoption assumptions and dependencies into a transparent three-scenario investment memo without inventing benefits or approval.
Financial analysts can use AI to structure an investment case when source facts, assumptions and scenario choices remain visibly separate. This fictional enablement proposal compares base, downside and upside cases and tests the adoption level needed to recover its cost. It is educational arithmetic, not investment advice, a forecast about a real product or approval to spend.
State the decision and evaluation window
An investment memo should name the choice, owner and time horizon before presenting benefits.
The fictional decision is whether to fund a one-year internal enablement programme. The accountable owner may approve, reject or request a smaller pilot. The evaluation window is twelve months, and the exercise ignores financing, tax, depreciation and effects beyond that period.
Ask AI to create a decision header with option, owner, horizon, currency and excluded treatments. Verify that the memo does not silently turn a pilot question into a full-rollout recommendation. A complete-looking template cannot fill policy choices that the brief leaves open.
Separate sourced costs from assumptions
Each input should reveal whether it is quoted, calculated or assumed.
The fictional proposal contains a 60-unit platform quote and 30 units of internal delivery time, for a twelve-month cost of 90. The delivery-time value is supplied by the programme owner for the exercise; it is not a market rate. No residual value is included.
Ask AI to build an input register with value, unit, period, source and status. Recalculate the 90-unit total and flag missing implementation dependencies. Do not add avoided costs, productivity gains or retention effects merely because they are common in investment templates.
Define benefits as scenario assumptions
A benefit model should show its mechanism and avoid presenting assumed adoption as measured behavior.
The base case assumes 30 active participants, each completing four relevant tasks per month, with a supplied value of 0.25 units per completed task. Across twelve months, the gross modeled benefit is 30 times 4 times 12 times 0.25, or 360 units. This is a fictional scenario, not an observed productivity result.
Ask AI to express the formula and units, then verify multiplication independently. Record that active participant, relevant task and task value require operational definitions before measurement. The calculation shows what the assumption implies; it does not prove that the assumption is credible.
Build downside and upside cases
Scenarios should vary consequential inputs without hiding which combination produces the answer.
The downside uses 12 active participants, two tasks per month and 0.20 units per task, producing 57.6 units. The upside uses 40 participants, five tasks and 0.30 units, producing 720 units. Against the 90-unit cost, the downside is negative 32.4 while the base and upside are positive before excluded effects.
Ask AI to generate the table only from the stated inputs. Check each formula, label decimals consistently and keep the cases symmetric in structure. Do not attach probabilities to the scenarios unless a source supports them, and do not average them into a supposedly expected result.
Find the reversal condition
A break-even condition makes the recommendation's dependence on adoption inspectable.
At four tasks per month and 0.25 units per task, one active participant produces 12 units across the year. Recovering 90 units therefore requires 7.5 participant-years, so at least eight consistently active participants under the exercise assumptions. Partial participation should be calculated rather than rounded away.
Ask AI to solve the formula, then substitute the result back into the model. State that eight is a model condition, not a validated adoption target. If task value or frequency changes, recompute the threshold and preserve the prior version.
Write a conditional recommendation
The memo should tell the owner what must be true and what evidence is still missing.
Recommend a bounded pilot only if the owner accepts the cost basis and can measure active participation, relevant tasks and the supplied value mechanism. Include the downside loss, the eight-participant break-even condition, unmodeled dependencies and a stop or review point.
AI may shorten the memo, but verify that assumptions, scenario labels and excluded effects survive. Do not claim the programme will create 360 units of value. The defensible conclusion is that the base case does so only under stated fictional assumptions that require testing.
Sources and scope
NIST: AI risk management and human review. Skills England: workplace AI foundation skills.
These references provide general background, not accounting policy, financial advice, validation or endorsement of this exercise. Every organization should use its authorized records, policies, controls and qualified reviewers. The entities, figures, assumptions and conclusions in the worked example are original fictional instructional material. AI can help organize supplied evidence, but it cannot approve a journal, certify a forecast or authorize a financial decision.
Sources: [1] [2]
Sources
- 1.AI RMF Core · NIST
- 2.AI foundation skills for work benchmark · Skills England