All articles
5 min readentrepreneurship / focus

Told to quit, 29% kept spending: write the kill criteria first

I have watched a team keep an AI pilot running for six months after it stopped producing anything anyone used. Nobody defended it in the meeting. They had spent the budget, signed the vendor, and closing it would have made the original call look wrong. Mark Keil and colleagues put a number on how common the pattern is: 30% to 40% of information systems projects show some degree of escalation 4.

What the research shows.

Hal Arkes and Catherine Blumer ran the tidiest field test in 1985. The first 60 people buying a season ticket to the Ohio University Theater got a random price: full price at $15, $2 off, or $7 off. Same seats, same plays, same information. Over the first five plays, the full-price group used 4.11 tickets. The two discount groups used 3.32 and 3.29. In the second half of the season the gap closed 1.

Barry Staw had shown the organizational version nine years earlier. 240 business students played a financial vice president allocating R&D money between two divisions. Some made the first allocation themselves, some inherited another manager's choice. Everyone then read the same disappointing results. The people who had made the original call put more money into the failing division 2.

Across the literature the effect holds at a moderate size. Dustin Sleesman and co-authors meta-analyzed the determinants of escalation and found sunk costs correlated .24 with continued commitment, corrected for artifacts, across 34 independent samples and 5,524 participants 3.

The field version costs more than theatre tickets. Thomas Åstebro's team followed inventors who had submitted their invention to Canada's Inventor's Assistance Program and received a diagnostic evaluation telling them to stop. 29% kept spending money on the project. 51% kept spending time. How much they had already spent predicted how much more they spent 5.

Itamar Simonson and Barry Staw tested de-escalation techniques head to head in 1992. Three worked: making the negative outcome less threatening to the decision maker, evaluating people on their decision process rather than the outcome, and setting minimum target levels in advance that trigger a change of policy when missed. The advance target produced the largest reduction 6. Holding people accountable for outcomes pushed escalation up.

The rule you write yourself is the weak half of the fix.

William Boulding, Ruskin Morgan and Richard Staelin tested five decision aids on senior executives running a new product launch. Better market information barely moved them. Precommitting at launch to a rule the executives wrote themselves barely moved them either. Two things worked: precommitment to a predetermined decision rule, and introducing a new decision maker at the stop-or-continue moment, the second one marginally stronger. None of the five aids cleared the bias against the control condition 7.

The same group later complicated the person swap. Eyal Biyalogorsky, Boulding and Staelin found that involvement in the initial decision is not a requirement for escalation. Managers who inherited a failing launch stayed committed too, carried by the beliefs they had formed about the product rather than by any choice of their own 8.

Field evidence is thinner than the lab. Staw and Ha Hoang reported in 1995 that NBA teams gave more playing time to high draft picks than on-court performance warranted 10. Daniel Leeds, Michael Leeds and Akira Motomura revisited the question with a regression discontinuity around draft thresholds and found teams allocate no more time to highly drafted players 9. Treat escalation as a tendency that appears under some conditions with a moderate effect, rather than a law of organizations.

The protocol.

  1. Write the kill criteria before the first euro leaves. One metric, one threshold, one date. "If fewer than 40 of the 120 trained users run a generation in week 6, we stop." Advance minimum targets beat every other technique Simonson and Staw tested 6.
  2. Let someone outside the project set the threshold. Rules the executives wrote for themselves did close to nothing, predetermined ones worked 7. Draft your number, send it to a peer, use the number they send back.
  3. Name who makes the stop call, and pick someone who was not in the original decision. That swap helped most in the experiment 7 and will not carry the load on its own 8, so hand that person the criteria in writing.
  4. Review on the calendar date. Book week 6 at kickoff. A review triggered by a bad feeling arrives after the money is gone.
  5. Judge the decision process at that review, not the result. Ask whether the team measured the threshold the way you all defined it at kickoff, and whether they read the evidence as written. Outcome accountability raised escalation in Simonson and Staw's tests 6.
  6. Decide where the people go before you decide to stop. In my trainings, the teams that close a project cleanly are the ones who already know which two engineers move to what. No study behind this step; it comes from the sessions I run.

Pick the project you are least sure about. Open its document and add three lines: metric, threshold, date. Send those three lines to one person who does not work on it.

Sources.

  1. Arkes, H. R., & Blumer, C. (1985). The psychology of sunk cost. Organizational Behavior and Human Decision Processes, 35(1), 124-140. sciencedirect.com
  2. Staw, B. M. (1976). Knee-deep in the big muddy: A study of escalating commitment to a chosen course of action. Organizational Behavior and Human Performance, 16(1), 27-44. sciencedirect.com
  3. Sleesman, D. J., Conlon, D. E., McNamara, G., & Miles, J. E. (2012). Cleaning up the big muddy: A meta-analytic review of the determinants of escalation of commitment. Academy of Management Journal, 55(3), 541-562. doi.org/10.5465/amj.2010.0696
  4. Keil, M., Mann, J., & Rai, A. (2000). Why software projects escalate: An empirical analysis and test of four theoretical models. MIS Quarterly, 24(4). aisel.aisnet.org
  5. Åstebro, T., Jeffrey, S. A., & Adomdza, G. K. (2007). Inventor perseverance after being told to quit: The role of cognitive biases. Journal of Behavioral Decision Making, 20(3), 253-272. doi.org/10.1002/bdm.554
  6. Simonson, I., & Staw, B. M. (1992). Deescalation strategies: A comparison of techniques for reducing commitment to losing courses of action. Journal of Applied Psychology, 77(4), 419-426. doi.org/10.1037/0021-9010.77.4.419
  7. Boulding, W., Morgan, R., & Staelin, R. (1997). Pulling the plug to stop the new product drain. Journal of Marketing Research, 34(1), 164-176. doi.org/10.1177/002224379703400114
  8. Biyalogorsky, E., Boulding, W., & Staelin, R. (2006). Stuck in the past: Why managers persist with new product failures. Journal of Marketing, 70(2), 108-121. doi.org/10.1509/jmkg.70.2.108
  9. Leeds, D. M., Leeds, M. A., & Motomura, A. (2015). Are sunk costs irrelevant? Evidence from playing time in the National Basketball Association. Economic Inquiry. doi.org/10.1111/ecin.12190
  10. Staw, B. M., & Hoang, H. (1995). Sunk costs in the NBA: Why draft order affects playing time and survival in professional basketball. Administrative Science Quarterly, 40(3), 474-494. semanticscholar.org

Keep reading