The engine
Everyone else tells you what happened. We tell you what they'll do — and what it's worth to you.
Competitive intelligence stops at the feed. PYRAMYD models the next move as an imperfect-information game: a probability over what your competitor does, a score for how much of that you can actually take, best responses ranked by expected value, and an honest band around the outcome.
The engine below is the real one. Drag anything.
Run it yourself
A competitor, four predicted moves, and one belief they hold that the evidence contradicts.
These formulas are the production two-sided synthesis engine, running in your browser. The scenario is illustrative — a composite competitor, not a customer's data — but every number you see is computed, not drawn.
What we think they'll do
Four predicted moves, each with a confidence the engine derives from signals. Drag one and every number below re-computes — the probabilities form a simplex, so raising one lowers the rest.
What they believe that isn't true
A Four-Corner blind spot: a management assumption the evidence contradicts. The wider the contradiction, the more exploitable every move premised on it — because they will keep committing to it.
They assume: Mid-market churn is a price problem
Evidence says: Their own reviews name onboarding time, not cost, as the reason for leaving
A price cut spends margin on the wrong cause — and leaves the real one intact
Posterior distribution
Exploitability
40%
moderate
Worth a considered response
- From the blind spot
- +4%
- Top contributor
- price-cut
Payoff distribution
4,000 seeded draws over the joint move × response tree. Responds to the move priors, not to divergence — a false belief changes what an opening is worth, not how the branches pay out.
- P5 (bad case)
- -0.44
- Median
- +0.16
- P95 (good case)
- +0.62
- Mean
- +0.11
- Probability of a net loss
- 39%
Prescriptions, ranked by expected exploitable value
- 01
Publish an onboarding-time benchmark and a 14-day migration guarantee
Moves the fight off price and onto the thing their own reviews complain about.
0.179
Experience
- 02
Pre-announce your agent tier and anchor the category on cited answers
Sets the evaluation criterion before they get to define it.
0.121
Capabilities
- 03
Bundle data quality into the base tier before their deal closes
Removes the gap they are buying, and de-risks the agent story at the same time.
0.110
Economics
- 04
Lock three EMEA public-sector references behind a compliance pack
Reference density is the real entry barrier in public-sector procurement.
0.068
Trust
Probabilistic decision-support: a best-response and exploitability estimate under imperfect information, not a certainty. Treat probabilities as calibrated priors to update as indicators fire.
How it works
Four ideas, none of which are a dashboard.
A posterior, not a guess
Each predicted move starts from a confidence the engine derives from signals, then gets reweighted by a Porter Four-Corner Driver prior — what management is actually incentivised and organised to do. The result is a probability simplex: the moves compete for one unit of belief, so raising one lowers the others. A prior can multiply a move by 4× or divide it by 4, and no further.
Exploitability, not just likelihood
Knowing a competitor will cut price tells you nothing about whether you should care. Exploitability asks a different question: how much better off are you if you respond well, versus doing nothing? It is the best-responder's expected gain over the status quo, weighted by the posterior — so a near-certain move you cannot profit from scores near zero, and that is the correct answer.
The blind-spot amplifier
This is the part with no equivalent elsewhere. When a competitor's move rests on a belief their own evidence contradicts — they think churn is about price; their reviews say onboarding — that move becomes more exploitable, not less, because they will keep committing to it. The amplifier is 0.5: at full divergence an affected move is up to 1.5× as exploitable. The premium is reported separately, so you can always see how much of the opening is theirs and how much is the mistake.
A band, not a point
A single expected payoff hides the thing that decides whether you act: how badly this can go. Four thousand seeded draws over the joint move × response tree give a P5, a median, a P95 and an explicit probability of a net loss. The mean converges on the enumerated expectation, which doubles as a cross-check that the tree and the simulation agree.
The difference between a competitive feed and a competitive strategy is whether the output tells you what to do on Monday. A prediction you can't price is a headline. A prediction with an expected value, a downside and a named counter-move is a decision.
What it refuses to do
The abstentions are the credible part.
Any model can produce a number for everything. The useful ones say when they don't know — and these rules are enforced in the engine, not in a policy document.
No data means no opportunity
A move with no assessed payoff contributes exactly zero to exploitability. It does not get an optimistic default. The engine would rather report a smaller opening than invent one.
The disclaimer ships with the output
Every two-sided synthesis carries its epistemics as a constant in the code, not a footnote in the UI: this is a best-response estimate under imperfect information, not a certainty.
Predictions are written down before they resolve
Each call goes to a ledger with the leading indicators that would confirm it, and is scored when it resolves. A forecast you cannot grade later is entertainment.
Where it shows up
The engine isn't a page in the product. It's underneath the pages you already use.
Competitor 360
Predictions sit above the evidence on every dimension — score, then forward look, then the signals that produced it.
See Competitor 360 →Battlecards
The prescribed counter-move rides on the card your reps carry, so the strategy reaches the deal instead of a slide.
See battlecards →Analyst forecasting
The same two-sided logic predicts where vendors land in the next Magic Quadrant, Wave or MarketScape — and prescribes how you move.
See the platform →Point it at your competitor.
The scenario above is a composite. Yours isn't. Bring a competitor you actually meet in deals and we'll run the real one.