Model the agentic organisation before you commit to it.

Model a business without claiming it is one.

A page full of headcount, released capacity and money reads as a statement of fact about a business. A what-if model and a real client organisation looked identical once they were open — and nothing on screen said whether that business exists. A checkbox on the organisation now records it as modelled rather than actual, and that label travels wherever its figures are read.

• A Simulation chip follows the organisation name in the header, a SIM badge sits beside it in the organisation list, and the Agentic and Target Operating Model menus open with the line Simulation — modelled figures, not actuals. The caveat sits where the modelled figures are read.
• New organisations arrive with the box already ticked. You record an organisation as real by unticking it, rather than by remembering to tick something — the safe default runs the right way round.
• Nothing else changes. No page is renamed, no generator behaves differently, no figure is calculated differently and nothing is hidden. It is a label — enough that a demonstration cannot be mistaken for the business itself on a board-facing page.
• Getting to a full twin is fast: upload the annual report and the AI extracts it, scans included, through OCR; Generate all pages then runs roughly fifty pages as one server-side job you can close the tab on.

What the organisation looks like on the other side.

Readiness scores say which departments are ready. They never said how much of the organisation that readiness represents — a department scoring 5 with three people read exactly as urgent as one scoring 3 with two hundred. The Future Agentic Organigram puts headcount behind the scores and draws both states over the same views.

• Current state and Future state are two tabs over one set of views: headcount per department, the organigram itself, the human-versus-agent mix, a current-to-future table, capacity and cost released, an impact-versus-readiness bubble chart, and the staffing gap.
• The target comes from the rubric, not from a generator: a department's readiness score maps to the share of work carried by agents — ten per cent at score one through ninety at score five — and any department can override its own share.
• The organigram hangs agents beneath their department. In future state those are the agents the target implies; in current state they are the agent workflows actually deployed — a real number, not a projection, on the same picture.
• Around it sits the rest of the target model: agentic departments, roles and teams with their diagrams, agentic governance where the four leadership rails are generated, the TOM roadmap, the KPI framework, and a Human tab on every role that describes the person rather than the work.

ROI you compute, not ROI you argue.

The Business Case reads the roles already assessed, takes the economics recorded on each role profile, and derives forward annual value and payback in months. There is no AI call anywhere on these pages — the same inputs always produce the same numbers, which is the difference between a figure you can defend and a figure you have to re-argue every meeting.

• Every figure is tagged From your data, Set by you, or Assumed default, and any of them can be overridden per role. The ROI table always says where its hourly rate came from.
• Freed capacity is split by an editable redeploy percentage — seventy per cent by default — and only the remaining reduction is costed, at the same hourly rate and working year the Business Case uses. The organigram and the business case cannot quote different money for the same organisation.
• The portfolio matrix plots agentic potential against effort and risk into Quick Wins, Strategic Bets, Fill-ins and Avoid — so the case arrives with a sequence, not just a total.
• Roles that cannot be calculated are listed separately under Needs input and excluded from the matrix and the totals, so the headline number is never quietly inflated.

Every number traces back to a line you can show.

A board overview is generated for you, and behind it sits a maturity model scored per readiness domain — base and target, in bar, radar and grid views — plus agentic-score snapshots that build a trend over quarters. What makes them usable is not the chart; it is that each figure can be traced back to a rubric line and the data someone typed in.

• A department is only projected when it has both a headcount and a score. Anything missing either is named as uncovered in a banner above every tab, with the headcount it represents — never counted as zero, which would quietly shrink the totals in your favour.
• An Estimate FTE button seeds empty departments from the employee count on the organisation profile and never overwrites a number already typed; where the departments that already have a headcount account for the whole company, it declines to seed at all and says so.
• Reports carrying AI-written content show an explicit AI-generated content marker with a tooltip naming Article 50(4) of the EU AI Act — so a deck can leave the building without someone adding the disclosure by hand.
• The simulation label is part of the buy-in, not an obstacle to it. A board that is told which figures are modelled trusts the ones that are not.

The simulation stops being one.

Once the case is funded, the model does not get thrown away and rebuilt in the real tenant. Untick the Simulation box, or move the company between deployments — the model travels, including the typed figures behind the business case. Then the first agents go out, and the projections start being replaced by what actually happened.

• Agent skills are generated per role, and the human-in-the-loop posture declared on that role is written into the deployed n8n workflow as a binding policy section, noted on the canvas as a sticky, and tagged on the fleet. A role marked Approve-each cannot deploy as an unattended agent.
• Three deployment paths — a single agent, an agentic team, or a one-button workforce. In a team each specialist carries its own posture and the manager applies the strictest across the team.
• The Agent Operations Center then shows the fleet live, with execution history, durations and failures, and realized-value tracking captures what each agent actually saved, month over month — the projection replaced by a measurement.
• A maturity snapshot each quarter closes the loop against the simulation: the modelled target on one side, the measured trend on the other, in the same units.

A loop that keeps turning, not a programme that ends.

Deploying agents is the easy half. Getting an organisation to absorb them is the work, and it does not finish — capabilities keep improving, so there is no refreeze to aim at. Crescith is the adoption framework that runs alongside: Surface, Pilot, Evaluate, Evolve, Diffuse, over four rails leadership owns.

• Growth gives consulting material its own home: upload methodology, frameworks, proposals and research as PDF, DOCX or TXT, and Growth Chat answers strictly from those files — separate from Company Chat, which answers from the client's own organisation model.
• Three chats with three scopes, deliberately not merged: Company Chat for the organisation, Growth Chat for the consulting library, and the Orgith Assistant for the tool itself.
• Named menu views let a consultant hand each stakeholder a surface that fits them — the board sees the board pages, a department head sees their own — without maintaining ten copies of the same permission set.
• The platform is multi-tenant and white-labelled, in four languages, with per-tenant data isolation — so a consultancy can run the whole route under its own brand.