The hidden third option: what integrated data can change about management decisions

Many important management decisions fall between two obvious reactions. Accommodate the customer or hold firm. Add capacity or wait. Invest or postpone. The real question is rarely asked: is there a third option we cannot see because the information we need sits in separate systems?

An example, explicitly hypothetical

Picture an industrial company. Its largest customer escalates a late, slightly off-spec delivery and openly considers a re-tender. Two proposals are on the leadership table:

• Reaction A: accommodate the customer with a permanent discount and additional capacity.
• Reaction B: refuse the demand and keep the existing process.

Both treat the escalation as a local customer problem. Both are understandable.

Now suppose someone connects information that was previously separate: the customer’s emails, statements from other customers, lost quotes, complaints, delivery data, lead times and margins. It turns out the same requirement appears with several customers. The bottleneck is not in production but before it, between requirement, technical clarification and internal approval. And the customer actually wants reliability, not a lower price.

Suddenly there is an option C: a standardised variant with a shorter approval path, tested at small scale first. It solves a segment problem instead of a single case, and it is reversible.

This scenario is a fully worked but hypothetical thinking model, not a client case and not an average. It illustrates a value logic, not a promised effect.

Four diagnostic questions before option C counts

A “new” option is only worth something once it is confirmed with original data:

1. Does the problem really start before production?
2. Is the requirement more than a one-off?
3. Does the planned measure address the actual criticism? Does a discount solve the problem or just cut margin permanently?
4. Is a reversible alternative realistically feasible?

Only when all four are answered with evidence may A and C be compared economically, with disclosed assumptions and traceable calculation.

What the research says

The mechanism is plausible, but not automatic:

• Customer needs from text can be identified by machine at least as well as from interviews, if the pre-selection is right (Timoshenko & Hauser, Marketing Science 2019).
• In a field experiment with 791 professionals at P&G, individual plus AI improved solution quality over individuals without AI. When selecting the best idea, however, the AI conditions did not do better (Dell’Acqua et al., Organization Science 2026).
• More information is not automatically better strategy: in an experiment with 348 decision-makers, broader information led to broader but shallower mental models, with no significant gain in foresight (Kanis, Mann & Stumpf-Wollersheim, Strategy Science 2026).
• Automated classification remains error-prone: on more than 15,000 real complaints at a medical technology company, even the best model got about a quarter wrong (Oesterreich et al., 2021).

The consequence: analyst, option generator, devil’s advocate

Integrated AI can widen the decision space. It can just as easily produce a seemingly new but useless option, or a convincing false consensus, if categories, matches, cut-off dates or causal assumptions are wrong.

So the AI belongs in the role of analyst, option generator and devil’s advocate. People make the decision. Quantitative statements rest on traceable calculation or are labelled as estimates. And before such a system goes into live decisions, it is tested on historical cases with a frozen information cut-off, against a human team.

That is exactly how we build our Decision Sprint: one concrete management question, diagnosis from original sources, the known options plus the hidden one, full cost and risk, and at the end a human decision with a bounded test or a deliberate stop.