AI assistants: when should a project be stopped?
A disappointing assistant often prompts another instruction, another test or an additional feature. For managers, the precise question is this: how can they decide whether to continue a project on the basis of its future usefulness rather than the work already invested?
What the literature establishes
Hal R. Arkes and Catherine Blumer (1985, Organizational Behavior and Human Decision Processes) show that previous investments of money, effort or time can encourage people to continue an activity. These sunk costs then influence a decision that should concern future costs and benefits. Applied to an AI assistant, this finding calls for distinguishing between “we have worked hard on it” and “it deserves further resources”.
When project owners evaluate their own choices
Barry M. Staw (1976, Organizational Behavior and Human Performance) shows, in a resource-allocation experiment, that personal responsibility for an initial choice can strengthen reinvestment when outcomes are unfavourable. The mechanism therefore goes beyond difficulty accepting a loss: continuing can also serve to defend an earlier decision. For a manager who has built a no-code assistant themselves, this possibility deserves consideration, without assuming that it explains every adjustment.
The premature conclusion
One might conclude that an assistant should be abandoned as soon as its initial results disappoint. These studies do not support that conclusion: a modification may have genuine value if it addresses an identified, testable problem. The useful distinction is between a new hypothesis accompanied by a test and continued investment justified mainly by past expenditure. (our executive and employee training programmes)
What these findings cannot determine
These studies concern neither generative AI nor no-code development platforms; they establish no universal stopping threshold. Nor do they demonstrate that an external evaluator or a stopping rule is sufficient to eliminate escalation of commitment. In practice, a successful demonstration alone does not establish whether the next stage deserves funding.
A practical check in Fribourg
As part of the SHR programme “Creating your own AI assistants and agents without coding”, a pilot in Fribourg could focus on preparing meeting requests with the university sector for an agri-food or industrial company working in French and German. Before the trial, the team would set a time budget for the next adaptation and a continuation criterion: reduce total preparation and correction time compared with the usual process, without increasing omissions against a predefined checklist of mandatory elements. Using comparable cases in both languages, the team would record times and omissions, then have the results reviewed by someone who did not design the assistant. The decision to continue, modify or stop would be documented against this criterion; any departure would require a testable hypothesis, rather than an appeal to time already spent. To go further: explore the Building AI Assistants and Agents Without Coding training in Fribourg in the canton of Fribourg, or browse our executive and employee training programmes in Switzerland.
In pictures: Building AI Assistants and Agents Without Coding in Fribourg
- building ai assistants and agents without coding
- Fribourg
- research
- training fribourg