The _MARKET-PROBLEM-MAP says where a tool plays (coverage 0–100 across 21 areas). This lens says whether and how WE attack it — scored, comparable, and tied to one specific thesis. Every dossier carries an Opportunity block scored on this lens. Roll-up = the #cross-competitor-opportunity-matrix. Dashboard: Dashboard · method: _RESEARCH-METHOD.
0. The play this lens serves (read first — the scoring only makes sense against it)
We are not replacing incumbents. We capture niches using two structural tailwinds incumbents can’t ride:
- Software is ~free to build now. Custom vertical software is no longer a moat to be bought; it’s a cost we don’t carry. So a “feature” an incumbent gates behind a $X/seat platform, we can ship standalone.
- AI inference is ~free for LLM-shaped work. DeepSeek-v4-class models do extraction / drafting / summarization / classification for cents. Wrapped in agent skills + SOPs, whole admin workflows collapse. So “the labor the incumbent’s software organizes” — we automate, not organize.
The wedge: a free, narrow entry point that delivers standalone value, so all value accrues up-funnel; then land-and-expand. We win where the incumbent is trapped (can’t copy us without self-harm) and lethargic (slow, mid-management-heavy, checked-out). We lose where the incumbent owns distribution and will bundle our AI as a free feature (e.g. Procore is already shipping “Procore AI agents” — do not fight there head-on).
Source frameworks (researched,
dossiers/raw_lens/): Helmer 7 Powers / counter-positioning, Christensen low-end & new-market disruption, PLG wedge / land-and-expand, and the LLM-shaped-workflow map for construction admin.
1. The three lenses + the prize gate + the kill-switch
Score each 0–100. Same axes every tool → comparable. One-line evidence per score (no absence-claims without proof — see _RESEARCH-METHOD).
Lens A — Counter-position: can the incumbent NOT follow?
High = reacting to us cannibalizes their own core. This is the difference between a soft target and a trap for us. Signals that raise the score (they’re trapped):
- Per-seat ARR trap — usage-based / free AI pricing would cannibalize the seat revenue their valuation rests on.
- Sales-led + services-revenue trap — high-touch motion and implementation/PS revenue can’t match self-serve.
- Channel conflict — resellers/partners block direct digital-first distribution.
- PE-owned / over-leveraged / public-market story — management optimizes legacy margin over the threat.
- Lethargy markers (from reviews) — low NPS, slow release cadence, stale UX, “support went downhill”, mid-management bloat, checked-out.
Signals that LOWER the score (don’t fight head-on): incumbent is modular, fast-moving, owns the distribution platform → they bundle your feature for free.
Momentum & chatter — the talk-vs-ship gap (key lethargy read): what is the vendor announcing / demoing / talking about on AI (keynotes, press, conference booths, LinkedIn) vs what’s actually shipped and in users’ hands? Today everyone talks AI; incumbents are structurally slow to ship it (legacy architecture, committees, channel/PS protection). A wide talk-vs-ship gap = a wide-open window — they’ve told the market the direction but can’t move fast enough to hold it. Capture: (i) their AI roadmap signals (announced/beta/GA?), (ii) the broader industry & conference chatter on this problem area (where the puck is going), (iii) whether their AI is real product or a slide. Big talk + thin ship + lethargy markers → raises Counter-position (they can’t defend the ground they’ve claimed). Real shipped AI + fast cadence → lowers it (they’ll bundle).
Lens B — AI-reimagination: can cheap AI eat the job (not just assist it)?
High = the value is LLM-shaped labor, not structured plumbing. Inference cost is NOT a constraint (assume ~free); the constraints are (i) is the job LLM-shaped and (ii) can we get the data (see kill-switch). Signals:
- Document-heavy / generation / extraction density — RFI & submittal drafting, daily-report synthesis, change-order/claim narratives, closeout assembly, contract review, doc-Q&A. (Construction’s most AI-amenable layer.)
- Manual hours removable — how much human admin time the AI deletes (not decorates).
- Agent + SOP leverage — can we collapse a multi-step / multi-role workflow into an agent skill + SOP?
- Compounding data loop — does running it accrue proprietary data that gets better over time (historical cost, claim outcomes, productivity benchmarks)? → durable moat. (Low = the value is structured data entry + integration plumbing, e.g. time→payroll. AI doesn’t reimagine that; integrations own it.)
Lens C — Wedge: is there a free, narrow entry that lands & expands?
High = a single acute pain, one persona, standalone, below procurement, with a clear expansion axis. Signals:
- Standalone single-persona pain — useful day one without buying the platform (“system of love, usable without integration”).
- Free / below-procurement entry — adopts without a buying committee.
- Value accrues up-funnel — the free thing creates the artifact/data the paid thing needs.
- Value-scaling axis + modular expansion — natural “I need more” at a seat/record/transaction limit, into an adjacent high-value module.
Prize gate — is the niche worth winning? (multiplier, not additive)
- Budget / P&L line strength — money-recovery (claims/variations) > payroll automation > risk > “nicer reports.” Who signs?
- Niche size & capturability — is there a real underserved 20% slice (segment / geography / trade / use-case) we can own?
Kill-switch flags (any one means re-think the angle)
The term is shorthand for “what would make us walk away.” In the brief itself, write these as a plain sentence, not a flag.
- A distribution-owner bundles you — a platform incumbent ships the same AI as a free feature (Procore-class).
- System-of-record moat — deep integrations and switching cost; the data loop is entrenched, so ride or flank rather than fight.
- Closed or no API — the data cannot be reached, so the AI layer cannot be built on top; this forces a full replacement and raises cost.
- Greenfield-only — only winnable on brand-new projects or firms with no lock-in; acceptable, but it narrows the addressable market and must be stated.
2. The verdict: one of four archetypes
The scores resolve to one archetype per competitor — this is the classification.
| Archetype | When | Move |
|---|---|---|
| A · Wedge-and-expand | High Counter-position + High Wedge; incumbent trapped & lethargic | GO. Free narrow entry on their turf → land & expand. |
| B · Reimagine-the-job | High AI-reimagination + the job is LLM-shaped end-to-end; often greenfield | BUILD-NEW. Don’t mimic their product; re-conceive the workflow AI-native. |
| C · Build-on-top / ride-along | Open API + strong adjacent whitespace; incumbent owns the capture, not the value | PARTNER/LAYER. Sit above as the AI layer on data they already hold. |
| D · Avoid head-on | Kill-switch fires (distribution-owner / SoR moat / closed) | DON’T FIGHT. Niche-flank only, or skip. |
Composite read (informal): Opportunity ≈ (A × B × C) × Prize, minus kill-switch flags. Resolve to GO / FLANK / AVOID + a one-line wedge hypothesis.
3. How this lens surfaces in the report
This lens is a thinking tool, not a section to paste in. The dossier never opens with archetypes or scores; the reader meets them only after they understand the product. The lens shows up in three places, always translated into plain language:
- Section 9 (the vendor’s own AI) carries the talk-versus-ship reading and the confidence score on how far the vendor can take their AI — this is where the counter-position and lethargy judgement lands in readable form.
- Section 11 (our read: can we enter and win?) is the conclusion. Write the archetype as a sentence (“we should build alongside them and flank the commercial gap”, not “Archetype C”), the wedge as a described entry point and expansion path, and the kill-switch as the “what would make us walk away” paragraph. Close the section with a compact assessment table whose rows are plain questions (where they are strong and off-limits; the verified gap; how hard it is for them to follow; how much AI can take over; the cheap way in; what would make us walk away; overall).
- The four-way archetype and the four scores (counter-position, AI-reimagination, wedge, prize) stay in our working notes so the cross-competitor matrix can be built, but they are not printed as a jargon block in the brief.
4. Cross-competitor opportunity matrix
Sibling to the coverage heatmap, kept in working notes. Rows are competitors; columns are the four scores plus the archetype and the walk-away risk. Sorted, it becomes the target board — who to attack first, who to ride, who to avoid — and it is built once the dossier set is populated.
5. Discipline notes
- Judge for our play, not in the abstract. A good company is not the same as a good target; a strong product with a trapped incumbent and an open API is a better target than a mediocre one that owns distribution.
- No absence-claims without proof (carried over from _RESEARCH-METHOD). “No API” must be verified on the developer docs, not assumed.
- The walk-away test is the honesty check: it is what stops us entering a fight a distribution-owner wins by default.
- The AI-reimagination judgement assumes inference is effectively free. Never mark a job down because it would be expensive to run; mark it down only if the work is not language-model-shaped or the data cannot be reached.