When "small team" is both a strength and a weakness

What evidence, shared inquiry and a reviewed next step add to SWOT in Harmonica.

In a synthetic test for Harmonica’s SWOT analysis, a volunteer-run facilitation team was considering paid workshops for the next quarter. One participant described the small team as a strength: it could tailor its work quickly. Another described a weakness: there was nobody to cover if a facilitator became unavailable.

Both observations could be true. If an AI groups them under a single “small team” label and chooses one quadrant, it removes a distinction the team needs for its decision. Fast adaptation might make a limited pilot attractive, while the lack of backup might make a larger commitment unwise. The analysis needs to preserve enough reasoning for people to examine that trade-off themselves.

We used this case while revisiting SWOT in Harmonica. Our checks of the existing implementation had found that some opposing factors could collapse together during merging, losing earlier evidence. A host’s correction could also remain on the original factor while a newly worded AI factor reintroduced the interpretation the host had corrected. These were problems in how the application maintained the analysis across changes, as well as in what the model was asked to produce.

Reviewing the matrix together is part of the FAO’s participatory SWOT guidance, which starts by identifying the subject and ends by working out responsibilities and timing for action. Its workshop variant also calls for agreement on the entries. In Harmonica, a host reviewing the matrix does not establish that participants agree, so differences must remain visible until people resolve them. FAO, Participatory policy development, Tool 9, p. 43.

The focal subject also changes how an observation should be understood. The small team’s capacity is internal to the team. A foundation’s possible grant next year is an external opportunity, and it should not become money the team already has. Current and future conditions can occur on either side of the internal/external distinction. In our test, the grant remained a future possibility, while one participant’s estimate of four months of runway remained an estimate rather than an agreed budget.

Keeping those qualifications requires a way back to what participants actually contributed. SWOT factors now reference eligible participant-authored messages; an AI-generated summary provides context, not additional participant evidence. We deliberately included an assistant-authored claim about a ten-person sales team in the test conversation, and it was excluded as a source for the factors. Evidence counts also remain counts of messages, not votes or measures of agreement.

A participant’s earlier warning about backup capacity should not disappear just because new contributions arrive. We separated older evidence outside the extractor’s recent-message window from contributions that have been deleted or excluded from the analysis. Refreshing the matrix can retain older eligible evidence while identifying factors whose support has changed.

Participants also need a way to examine the joint analysis. The results page now opens with the native SWOT matrix for eligible sessions, with the summary still available. People can inspect a finding’s reasoning and evidence-reference count within the session’s existing results permissions. Private participant quotes remain withheld from that shared view.

The shared-inquiry feature connects discussion to an individual finding. When enabled for a session, it lets people question that finding, challenge its classification or explicitly ask Monica about it. For the small team, a participant could ask whether the backup concern rules out a workshop or calls for a cover arrangement. Monica’s context distinguishes the permitted matrix from the summary, and references identify the source and version being discussed. Discussion can begin before host review is complete; questions and AI answers do not automatically correct the matrix, establish consensus or make a decision.

The host can inspect the permitted evidence and amend a factor’s wording, reasoning or quadrant. Those amendments are preserved through regeneration. For the small team, a useful correction might distinguish capacity for tailored pilots from capacity to deliver several workshops reliably. A question in the shared discussion can prompt that review, while responsibility for the interpretation remains with people.

Even a corrected matrix leaves the team’s launch decision open. We added a separate strategy handoff where the host can record options and who proposed them, who has decision authority, and whether the outcome is a decision, a deferral or a request for more evidence. It also records unresolved differences and the next step, including whether a proposed owner has accepted responsibility and when the matter should be reviewed. The host records that acceptance; entering somebody’s name alone does not establish it.

That handoff belongs to a particular reviewed version of the matrix. If the underlying factors or source basis change, the saved handoff is marked stale. An older editing session cannot silently overwrite a newer review. The record stays private to session editors, separate from the session summary; its copy function omits raw participant quotes. Sharing it onward still requires choosing an appropriate audience and checking what the host has written.

We kept the quick, single-session SWOT format and made the review and handoff explicit. A second contribution round can be useful when a group genuinely needs to consider options together, but it was not necessary for this change. TOWS offers a named way to develop strategies from relationships between internal and external factors; adopting that structure would be a deliberate variant. Weihrich’s original paper gives relationships between factors a central role in developing strategies. Heinz Weihrich, The TOWS Matrix: A Tool for Situational Analysis, 1982.

The Open Facilitation Library’s Conformance Model helps separate the judgments involved here. It distinguishes a method’s essentials from a product’s facilitation choices and runtime conventions, and treats participant outcomes separately. A method review follows those requirements through session structure, tools and artifacts. This SWOT workflow keeps one contribution session, with shared scrutiny where enabled, host review and a strategic handoff. That is our chosen product design, not a universal definition of SWOT.

Our earlier Delphi work needed different choices around rounds and controlled feedback. OFL’s portable method work can make such requirements and adaptations explicit, while Harmonica implements them and retains evidence of what it actually does. For SWOT, we worked from the published sources discussed above; we have not adopted a reviewed OFL SWOT profile or established a formal conformance claim. The implementation findings are concrete material that method work can examine.

The code is deployed. We reviewed two real model extractions against synthetic contributions, checked that host amendments survived regeneration, and exercised handoff review and concurrent-edit safeguards. The participant matrix was also inspected on an existing public results page with 18 findings, at phone and desktop widths. Session activation and live use of finding-linked discussion remain distinct from releasing its code; we have not established the quality of Monica’s answers in that workflow. Nor do these checks tell us whether real groups reach better strategies, whether hosts find the review worth their time, or whether people feel accurately represented. Those outcomes remain unmeasured.

For the small team in the example, a reasonable next step could be to investigate facilitator cover before committing to a launch. The handoff can preserve the case for a pilot alongside the reliability concern, identify the evidence still needed and record that no launch has yet been approved. That gives the next conversation a specific decision to return to.

Book a 30-minute call

Loading calendar…