Gibbs reflection for managers: reviewing one decision in six steps

An order goes out three days late. The debrief runs for twenty minutes and ends with two actions: check machine schedules more carefully, tell the customer sooner. A fortnight later another order is late. The situation is invented, but that kind of meeting is what the Gibbs reflective cycle is aimed at. According to the summary in the site's library, Gibbs observes that debriefings often lurch from a superficial description to premature conclusions about what to do next (Gibbs, 1988). This post runs the six steps on one management decision, then states the limits.

What does each of Gibbs's six steps ask, and in what order?

Six questions, taken in turn: what happened, what were your reactions and feelings, what was good and bad about it, what sense can you make of it, what can you conclude, and what will you do differently next time Gibbs reflective cycle. The framework comes from Learning by doing (Gibbs, Further Education Unit, 1988) (Gibbs, 1988).

The order is what the framework imposes. Step 1 only describes; step 2 records feelings without analysing them; value judgements wait until step 3. According to the summary, Gibbs offers the steps as a structure for debriefing an experience, alone or in a group. With enough time, move through them in order and change the working format at each step (alone, in pairs, in fours, whole group) to mark the transition (Gibbs, 1988).

What does one decision look like when it goes through all six steps?

Pick a decision, not an outcome, then fill the six boxes in order. The table is an invented entry: Lan, the mid-sized packaging printer and all the numbers are my own creation and stand for no real business.

Hammond, Keeney and Raiffa (1998) write that bad decisions can often be traced to how they were made, such as the right information not being collected or costs and benefits not being accurately weighed (Hammond et al., 1998). That is why the description box records what Lan did, not only what happened. The situation: on the 2nd, sales asks Lan, the operations head, whether a rush order of 40,000 boxes can ship on the 12th.

Box 4 borrows two ideas from Hammond and colleagues: the anchoring trap gives disproportionate weight to the first information received, and the confirming-evidence trap makes us seek information that supports an existing view and discount the opposite (Hammond et al., 1998). The site's framework on decision traps summarises them. Labelling the entry this way is still a hypothesis; the machine log would show whether it holds. Box 6 is the only one with a checkpoint, which turns a conclusion into something measurable after 30 days.

Which two failures are most common, and what do they look like?

According to Gibbs, the first is leaping from a superficial description to premature conclusions about what to do next, skipping reflection and analysis. The second comes after a powerful experience: a group stays stuck on description or feelings and returns to them just when it should be discussing action plans (Gibbs, 1988).

The opening meeting is the first failure. The description box says only "order three days late", and the plan reads "check machine schedules more carefully, tell the customer sooner". Both would fit any late order, and neither touches the habit of fixing a date by phone. My test: if the plan still reads sensibly after you swap in a different late order, the analysis box was skipped.

The second failure is easy to picture if the customer threatens to cancel. The meeting spends the whole session on who said what in the call, then on who is angry with whom, and returns to those subjects when someone proposes talking about the plan. The summary only says to change the working format at each step to mark the transition; using that to escape this failure is my inference. My suggestion, not Gibbs's: each person writes the description alone, feelings are shared in pairs, evaluation is done in fours, and the whole group takes the last three steps.

How is Gibbs's cycle different from an after-action review?

An after-action review (AAR) is a structured discussion soon after an event asking what happened, why, and what to do differently, as the site's glossary puts it. Against that definition, Gibbs adds a separate feelings step and asks you to split what can be concluded in general from what applies to your own way of working.

The biggest difference is evidence. Keiser and Arthur (2021) pooled 61 studies (107 effect sizes, 915 teams and 3,499 individuals) and found that the AAR improves multiple training evaluation criteria, with an overall d of 0.79 (Keiser and Arthur, 2021). Two characteristics contributed consistently: whether the review is aligned to the individual or the team, and whether the performance review media are objective. High structure beat low structure in the military, but the two were comparable in healthcare (Keiser and Arthur, 2021).

As I read it, what was measured is the AAR, not the Gibbs cycle, so nothing here says how much any Gibbs step contributes. The finding on objective media does suggest a fix for the description box, which only asks for a description: bring in the machine log, the order schedule and the confirmation email, rather than letting Lan write from memory. That is my inference, not a recommendation from Keiser and Arthur.

Can reflection simply rationalise?

Yes. Six questions do not test the answers, so the writer can still fill in whichever boxes sound most plausible. The framework summary gives no figures showing that completing all six steps leads to better later decisions.

Two traps on Hammond and colleagues' list can reach the evaluation and analysis boxes. The confirming-evidence trap leads someone who already suspects sales of pushing the date to find only reasons that prove it. The recallability trap prompts us to give undue weight to recent, dramatic events (Hammond et al., 1998). One late order might lead Lan to write a rule for all rush orders, even though the previous ten shipped on time. Linking the two traps to those boxes, and both examples, are my inference.

My suggestions for reducing the risk: put data in the description box, have someone who was not on the call read the entry, and set a checkpoint in box 6. The site's four-control-mechanisms framework adds a warning sign: reviews that only assign blame are failing reviews. It is a synthesis, not a tested finding. That is why box 3 in the example judges how the date was fixed, not the sales team.

This method suits you if you are reviewing a decision of your own or of a small team, can describe it through concrete events, and have someone who will read the entry again after 30 days. It does not suit you if you need the true cause of an incident with many variables, because that takes data and investigation, or if the team fears being blamed. I suspect such a team will write its feelings box for safety rather than honesty; the condition is called psychological safety.

Going deeper: what did I read, and what could I not read?

The Gibbs cycle: I could not read Learning by doing. Open Library lists the 1988 edition at 129 pages, and the Internet Archive holds a borrow-only copy that I could not open. The six steps, the way of using them and the two common failures come from the framework summary in the site's library, so I have no page numbers.

Keiser and Arthur (2021): I read the publisher's abstract on Ovid, not the article. I do not know how the studies were selected or how many came from business settings. The abstract mentions military and healthcare, so applying it to management decisions is my own extrapolation.

Hammond, Keeney and Raiffa (1998): I read the summary and opening paragraph on the Harvard Business Review page, not the full article, so I have not read their advice on avoiding the traps.

This post does not show that six-step reflection leads to better decisions; it sets out a structure and an invented example.