The strategy step you shouldn’t skip.

Most organizations move straight from ‘we have a problem’ to ‘let’s hire someone to fix it.’ The Premis™ Brief sits in between. It takes the problem you’re dealing with — the one where the team is split, or the initiative that keeps stalling, or the decision nobody wants to make — and runs it through a rigorous diagnostic before anyone spends budget or commits a direction. You get a structured read on what’s actually broken, three strategic options with one recommendation, and a 90-day execution plan — without a retainer.

Business leaders and strategy teams who need to move — without waiting months for a consulting engagement. If you’re dealing with a problem that keeps resisting straightforward solutions, preparing for a difficult board conversation, or facing a decision where every option has a credible argument, this is built for you.

The ones where reasonable people disagree on what’s actually wrong. An initiative that’s stalling despite the right team and resources. A market position that isn’t holding. A strategic decision where the data points in different directions. The Premis™ Brief is not designed for simple operational problems with clear answers — it’s built for problems where the framing itself is uncertain.

No — Premis scopes the work to the question. When your problem is bounded (a clear either/or, with enough on the table to settle it), it recommends a First Read: a 60-second committed verdict — the call, the reasoning, a measure of whether it’s working, and the condition that would tell you it’s wrong. When there’s more underneath, it recommends the Full Brief: the four-stage diagnosis, three strategic options with a recommendation, and a 90-day plan. Premis recommends the lane; you confirm with one click — and a First Read can always escalate to the full brief.

Most AI tools take your framing at face value and generate a response. The Premis™ Brief probes the framing of the problem first — running a structured four-stage diagnostic before any answer is produced. It classifies the problem type, builds a primary hypothesis grounded in what you’ve described, stress-tests it against alternatives, and designs the specific tests that would confirm or overturn it. Only then does it move to strategy. The output isn’t a generated answer. It’s a reasoned position with the logic visible.

Unlike a standard AI tool that responds to a single prompt, The Premis™ Brief runs a pipeline of specialized agents. Each has a distinct task, a structured input from the previous stage, and a defined output that feeds the next. An agent doesn’t chat — it receives a specific job, reasons through it, and hands a structured result forward.

After the diagnostic completes, three agents run in sequence:

Blueprint — converts the diagnosis into a scored problem blueprint: root causes, constraints, evidence gaps, and the fastest available test. Everything downstream reasons from this structure.

Strategy — generates three distinct strategic options, commits to one recommendation, names what it sacrifices, and states the single assumption it can’t survive being wrong about.

Execution — converts the chosen strategy into a 90-day task-level plan with action-first tasks, functional owners, binary success metrics, dependencies, and the decisions that need a human in the room.

The methodology governing each stage is built into the system. It doesn’t change with how the question is phrased.

It will tell you when it’s uncertain. Every diagnostic includes a confidence level and a list of what the analysis is working without — the specific evidence gaps that would sharpen or change the conclusion if filled. The diagnosis is a committed hypothesis to pressure-test, not a verdict to act on without review. The strategy names the single assumption it can’t survive being wrong about.

Your input is never used to train the underlying model, and every analysis starts fresh from the details you provide. What carries forward is context. Your organization context — standing constraints, what’s already been tried, current priorities — is saved and injected into every run, so the analysis reflects your situation, not a generic one. And once you’ve run before, your outcomes carry forward too: when you log how a past plan actually held at its check-in, that result is fed into your next diagnostic — so Premis reasons from what it told you last time and how it played out, not a blank slate.

Six stages. Three downloadable outputs.

Output	What’s in it
The diagnosis	What kind of problem this is and what decision needs to be made Primary hypothesis grounded in what you wrote — alternatives ranked by probability, falsifiability condition for each An issue tree built to pressure-test the diagnosis — each branch marked as supporting or challenging it Specific tests that would confirm or abandon it — and where analysis hands back to you
The strategy	Three meaningfully distinct strategic options One committed recommendation with constraint-aware justification The assumption it cannot survive being wrong about The leading indicator that a pivot is needed When Deep Research + Competitive Context is on, options are developed with awareness of the external market landscape — not just the internal problem frame.
The execution plan	90-day task-level plan with action-first tasks Functional role owners, binary success metrics, dependencies Failure modes with early signals Decisions that need a human in the room

The diagnosis and strategy download as PDFs — formatted for a leadership conversation. The execution plan downloads as a structured Excel workbook, organised by phase, ready to assign.

Treat it as a serious first position, not a final verdict. The diagnosis tells you its confidence level. The strategy names the assumption it can’t survive being wrong about, and the leading indicator that would signal it’s failing. The execution plan names the calls that need you in the room — the ones that depend on knowing your organization, your people, and what’s actually feasible. The system is honest about where it runs out.

Yes — it’s built to leave your screen. The diagnosis and strategy download as structured PDFs, formatted for a leadership conversation and designed to be presented directly. The diagnostic PDF includes a section marked Sensitive Analysis — the handoff note naming the political, relational, and contextual calls the analysis can’t make for you. It’s included by default, but can be toggled off the download if you’re sharing more broadly.

The execution plan downloads as an Excel workbook — one tab per 30-day phase, pre-filled with tasks, functional owners, success metrics, and dependencies — plus an import-ready tab that drops straight into the planner your team already runs on, from Asana and Linear to Jira, Monday, or Notion. It also names what it can’t resolve: the missing evidence, the single assumption the strategy can’t survive being wrong about, and the decisions that need a human in the room. Read both — the plan tells you what to do; the flags tell you where your judgment replaces the analysis.

It doesn’t end at the download. For signed-in users, every run is saved — a full history, with read-only replay of any past brief — and Premis schedules 30/60/90-day check-ins on the plan it gave you. At each one you mark how it actually held. The brief is the start of a loop, not a one-off document.

It runs on enterprise-grade infrastructure — encrypted connections and security headers on every response. We’re specific about what’s kept: for each run we retain the problem classification, the confidence level, the generated brief, and a verbatim excerpt of what you typed — the first ~1,000 characters (so for a short problem statement, that can be all of it). Signed-in runs are also saved to your account so you can revisit and replay them. None of it is used to train any model. Attached documents are handled differently — they’re used only for the run and never stored. Everything you submit is processed via the Anthropic API, so apply the same standard you would to any external AI tool (see below).

What you type goes to the Anthropic API to generate the brief; it isn’t used to train Anthropic’s models. On our side we retain the problem classification, the confidence level, the generated brief, and the first ~1,000 characters of your input, verbatim — for quality review, and (if you’re signed in) so your run history and replay work. Input beyond that length isn’t stored. Attached documents aren’t stored at all — they exist only for the run that uses them. If you’re signed in, your runs are tied to your email so you can find them again.

Yes. Alongside the problem you describe, you can attach up to three documents — a P&L, a term sheet, a data export (.txt, .csv, .md) — as evidence. The analysis reads them for that run: it cites specific figures where they sharpen the diagnosis and flags where they contradict your narrative. Their contents are treated strictly as data, never as instructions. And unlike your typed input, attached documents are never stored — they exist only for the run that uses them, then they’re gone.

Match the detail to the channel. The diagnostic works on the structure of a problem, so the free-text box rarely needs identifying specifics — and because a verbatim excerpt of it is retained, keep names, client data, and anything under a confidentiality or regulatory obligation out of it. When the analysis genuinely needs your real figures, attach them as documents instead: those are used only for the run and never stored. One rule covers both — everything you submit is processed via the Anthropic API, so material you’re contractually or legally barred from sharing with an external provider is best left out entirely.