Methodology

Many organizations use AI every day and have changed nothing about how they work. A few have rebuilt an important piece of work around it. The Navigator tells the two apart and says what to do next.

It applies the AI Adoption Framework, version 8.0.0, written by Jacek Siadkowski and Radosław Taraszka at Tech To The Rescue. The framework is published under an open licence, CC BY 4.0, that allows reuse with credit.

One invented organization runs through this page. Northgate Food Bank hands out food parcels and runs casework. For a year its AI has ranked the families who apply for parcels, as part of normal operations. Three people answered: the director, the operations coordinator and a caseworker.

A result has three parts:

• what kind of organization this is, and how far to take AI; this part sets the bar and never changes the stage;
• where the organization is today: its stage, the stage’s status, and what is in place and missing in seven capacity domains;
• what to do first: at most three actions, and the next step in the Adoption Loop, the framework’s seven-step cycle for putting AI to work.

A higher stage is not a better result. Northgate serves people in person and uses AI for part of that work. For such an organization the framework’s healthy aim is Systematizing or Innovating. For a tech-focused organization it is Innovating or Enabling.

The first part also says whether AI affects the people the organization serves. If it does, seven protections for them count from Systematizing on, such as a way to appeal an outcome and testing for bias. Northgate’s ranking decides when families receive food, so they count.

The stage says how far AI goes in the organization’s work, and whether the organization can carry it.

Each stage has two kinds of condition. The table above lists the deciding conditions: what places an organization at a stage, and what that stage also requires. If one is missing, the stage is not reached.

The second kind are readiness conditions, a longer checklist that shows whether the stage stands firmly. One of Systematizing’s is a baseline and an intended outcome for the priority workflows. Systematizing’s checklist has up to 24 conditions, Innovating’s up to 15 and Enabling’s up to 10. “Up to”, because some apply only in some cases.

From Systematizing on, a stage is:

• confirmed, when every condition is in place and every stage below is confirmed;
• conditional, when every deciding condition is in place, more than half of the readiness conditions that apply are, and every stage below is confirmed;
• not reached, otherwise.

The organization’s stage is its highest confirmed stage, or the stage just above it if that one is conditional. Exploring is reached by using AI; it has no readiness checklist.

A condition says in its wording when it applies only in some cases. A check of work built pro bono, for example, applies only where the AI work is built pro bono. Where a condition does not apply, it counts neither way. A condition nobody could confirm counts as not in place.

Northgate’s ranking runs in normal operations, and a week’s failure would harm families, so AI already goes as far as Innovating. Systematizing’s deciding conditions are in place: an AI owner with time and authority, and a team that shares what works. So are 19 of its 24 readiness conditions. That puts Northgate at Systematizing, conditional. Innovating is assessed once Systematizing is confirmed.

Apart from the conditions on how far AI goes, every condition belongs to one of seven capacity domains:

• Leadership and ownership
• People, workflows and change
• Data and organizational context
• Technology, implementation and resilience
• Governance, risk and accountability
• Evidence and learning
• Resources and sustainability

Each domain lists what is in place and what is missing toward the stage the organization is working toward. At Northgate, Governance lists a way for families to appeal the ranking as missing. The report marks each condition one of these ways:

• in place;
• missing;
• “Nobody could confirm it: counts as not in place”;
• “Not asked yet: counts as not in place”, where the survey has no question for it yet;
• “does not apply”, left out of the count.

Every condition that is missing, that nobody could confirm or that is not asked yet is a gap. The report turns at most three gaps into recommended actions.

Before the actions comes what the team must settle. Where people answered differently and the result depends on it, they agree first. When one person answered, the first step is to ask at least one more.

Support means a service, an expert, a partner or funding that can help close a gap, such as a data-readiness sprint with a data partner. Where support is listed, an action names its kind and the kind of provider. Otherwise it says “No support listed yet.”

The actions follow a fixed order. Gaps in the protections for the people served come first. Then gaps that support available now can help close, then the rest. Within each, what a workflow in normal operations still lacks comes first. Each action gives its expected outcome and the evidence of progress.

Northgate’s team first agrees on how it shares what works: two people said “regularly”, the caseworker “informally”. With no support listed, its three actions are, in order:

1. a way for families to question, correct or appeal an outcome;
2. a named owner of the ranking’s data, who checks its quality;
3. rules for human review, escalation and exceptions.

Last comes the next step in the Adoption Loop. For Northgate it is Step 5, redesign the workflow and roles, the step of its first action. If the team agrees and closes its gaps, Northgate’s next assessment confirms Systematizing and looks at Innovating.

The Navigator asks three or four people in named roles to answer the same questions alone. Nobody sees a result before their answers lock.

Their answers are combined by fixed rules. On how far the organization has got, the middle answer counts; with an even number of people, the more cautious of the two middle answers. On what is at stake, the most demanding answer counts. Northgate’s director said staff would feel a mistake first, and the coordinator said families, so the coordinator’s answer counts. “We don’t know” does not vote.

Confirming Systematizing or a stage above it takes at least two people who agree on the answers that decide it. If anyone’s answer would change a deciding condition, that stage is at most conditional until the team agrees. At Northgate, the caseworker’s “informally” is such an answer.

If only one person answers, their answers confirm no stage. The report then opens with “One person answered”, shows Resisting or Exploring without a status, and any higher stage as conditional at most.

The result has no score and no colour. A tool that drafts internal reports and a tool that decides who eats need very different capacity. One number would hide that. Organizations are compared on their stage and its status, and each against the healthy aim for its own kind.

Since 9 November 2026 the Navigator reads every assessment on framework 8.0.0, earlier ones included. The questions stay the same. These rules for reading the answers changed:

• A stage has a status: confirmed, conditional or not reached. In version 7 a missing readiness condition never mattered for the stage. Now any one keeps the stage from being confirmed, and half or more missing keep it from being reached.
• Using AI reaches Exploring. Naming an AI owner no longer does.
• Domains are no longer red. Each lists what is in place and what is missing.
• The seven protections for the people served became readiness conditions of Systematizing. In version 7 a missing one made governance red.
• The 23 warning signs of version 7 became conditions.
• Binding constraints became at most three recommended actions, with the kind of support instead of named programs.
• Systematizing and the stages above it are confirmed only when at least two people agree.

So a result first read on version 7 can read differently now.

Every result names the date it was read and the framework version. For now, every result is read again by the current rules each time it is opened. The answers people gave are never changed.

No language model decides a stage, a gap or an action. The rules are code, so the same answers, read by the same version of the rules, give the same result.

The framework changes only by a release its authors approve. Tech To The Rescue maintains the Navigator and this page. The Navigator relies on what people tell it, because it cannot read documents yet.