Strategic Option Comparison Synthesizer

A strategic-option synthesizer is useful only when it scores options against criteria the client already agreed in writing. If you let the model invent a framework, it will score the firm's favorite option against the firm's favorite dimensions, and the steering committee will argue about the axes instead of the choice.

Lock the criteria first: names, definitions, and what "good" looks like on each, in a short note the sponsor signs or at least replies to. Weights can wait. They are a later argument. The lock is the list itself. Until that list exists, synthesis is a deck-writing shortcut, not a comparison.

Hidden criteria are the usual failure. The sheet says customer continuity, synergy speed, and implementation disruption. In the room, the real veto is that the CEO of the largest acquired brand will not accept a name change, or that the board already promised a footprint two of the options would reverse. If that veto is not on the sheet, the synthesizer looks analytical and still loses the meeting. Ask which criterion would make you reject an option that wins on everything else, and put it on the sheet, even if it is political.

If the client will not lock criteria, do not run the synthesizer. Write a decision-framing note instead: here are the options as we understand them, here is what we still need agreed.

Knowledge workspaces and spreadsheets in this class (Notion, Excel) are where the locked list usually lives. Put the signed criteria there, dated, with the sponsor on the page. Retrieval products in this class (Glean, Hebbia) will pull old comparison templates from other engagements. Do not paste those templates in as this client's starting grid.

The job is a grid: options as columns, locked criteria as rows, and under each cell a qualitative trade-off with a pointer to evidence from this engagement. Interview notes, operating facts the client supplied, and method precedent stripped of other-client names. The output should make the discussion about weights and residual risk, not about what the options even are.

The synthesizer must not pick a winner. Ranked options, "recommended path" banners, and a bolded top cell are a different artifact. Once the model names a winner, people argue with the ranking instead of with the cells. A partner who already has a preferred path can say so after the grid is on the table, not by asking the model to launder that preference as analysis.

Name each option's weakest cell in the same type size as the strong ones. An option that looks clean on four criteria and silent on the fifth is the one that blows up in steering. If a cell cannot be filled from this client's facts, write "not yet evidenced" rather than filling it from a similar-sounding analog. Method precedent can illustrate how a trade-off tends to play out, for example what usually breaks when you centralize pricing. It cannot substitute for this client's pricing power or talent map.

When the options are not actually comparable, stop. "Stay as three brands," "collapse to one brand," and "sell the company" are not the same kind of choice. Split the decision: first whether to stay independent, then how to operate if you stay.

Retrieved precedent is what distinguishes this from a blank matrix, and it is the fastest way to leak another client. A comparable past scope retriever can tell you the firm has compared similar operating-model choices. It must not return the other client's org chart, synergy model, named executives, or board exhibits.

What you may retrieve is method: how the firm structures a brand-architecture comparison, which failure modes showed up when centralizing a commercial team, what questions the last steering committee actually asked. Names, numbers, and identifiable anecdotes stay out. A reusable IP extraction agent is the place to separate method from client work product at close. If that separation never happened, do not retrieve the old engagement into this one.

Access follows the source system. A partner who could not open last year's portfolio-company folder should not see its comparison appendix in a retrieval snippet. Named-entity stripping is not enough: an identifiable anecdote is still a leak. If the corpus cannot enforce that split, run with no retrieved precedent. A grid from this engagement only is slower and safer than citing another account.

Take a PE-backed industrial distributor after two add-on acquisitions. Three operating options are on the table: keep independent regional brands with a thin shared back office; collapse into one national brand and a central commercial team; hybrid, with a national key-account team and regional brands for local trade.

The sponsor locked five criteria: customer continuity, speed of cost synergy, commercial talent retention, channel conflict between house brands, and disruption to the next two selling seasons. The team puts those in Notion, dated, with the sponsor copied. They do not ask the model for a framework.

What the synthesizer should return

A grid, not a ranking. Under customer continuity, independent brands are strongest and collapse is weakest, with pointers to interview notes. Under synergy speed, collapse is strongest, as an operational claim, not a modeled dollar range. Those ranges belong in the multi-scenario outcome modeler, using this client's cost and volume inputs. Method precedent, stripped: when commercial teams were centralized in similar roll-ups, local pricing discretion was the first thing that broke. No other client's name, no other client's synergy exhibit.

What usually goes wrong

Two failures show up together. First, a hidden criterion: the CEO of the largest acquired brand has already told the sponsor, off the sheet, that they will leave if their name comes off the trucks. Talent retention is on the grid; this veto is not. The synthesizer scores collapse as the cleanest synergy story and, if you let it rank, names collapse the winner. The meeting then becomes a fight with a ranking that never encoded the veto.

Second, retrieval. Glean or Hebbia, pointed at the firm's project archive, returns last year's similar PE roll-up, including the confidential brand-architecture appendix. A junior pastes three cells because they look like evidence. That is another client's exhibit, not the method note you wanted (local pricing broke first).

Fix both before the steering deck. Put the veto on the criteria list, even if it is political, and regenerate. Strip the retrieved appendix. Keep the grid unranked. The partner can still recommend hybrid in the room, with the weak cells named first.

This job is qualitative trade-offs on locked criteria. Quantitative ranges go to the multi-scenario outcome modeler. If you let the synthesizer invent a range because a cell felt empty, you have created a number nobody can defend in finance. Point the cell at the modeler, or leave it as a qualitative claim.

Once the client has chosen a direction, sequencing initiatives is a different artifact: the initiative prioritization matrix generator. Do not reuse the option grid as an initiative ranking. Options are mutually exclusive paths. Initiatives are a stack you might run under the path you picked.

Before the steering meeting, run the comparison, not a fake winner, through the recommendation adversarial stress-tester. The useful output is the objection that hits the weakest cell, and the answer you would need. The stress tester does not rewrite the grid.

Vendors will not enforce this split. Excel will hold a score, a made-up range, and a ranked list on one tab. Notion will store last year's named case next to this year's criteria. Glean and Hebbia will retrieve both unless you constrain the corpus. None of these is a ranking of products. Pick the store your knowledge managers will keep clean, and make the synthesizer read only the locked criteria, this engagement's evidence, and stripped method notes.

The system is doing its job when the steering committee argues about weights on a grid they recognize as their own criteria, every weak cell is visible, the numbers live in a model they can challenge, and nobody in the room is looking at another client's exhibit.