White Glove LabsFrançaisBook a call

Free tool

Score a candidate AI workflow before you buy anything

This is the scoring instrument the Opportunity and Governance Sprint runs, published in full with its weights and its anchors. Twenty minutes, one workflow, a number you can defend to a sponsor. Print it, run it with your own team, and use it on any supplier, including us.

Go to the sheetHave us score it with you

How to run it

Twenty minutes, three people, one workflow

The instrument answers one question: is this workflow worth governing and shipping at all. It is deliberately answered before any control design, architecture or vendor selection, because each of those is expensive to undo.

  1. Pick one workflow, described as it runs today. Not a capability, not a department, not “AI for policy”. A workflow has a trigger, a set of steps, a person who does them and an output someone waits for.
  2. Get three people in a room. You need the person who does the work, the person who owns the outcome, and someone who knows where the data lives and who may see it. Without the third person, you guess the Data score instead of knowing it.
  3. Score each dimension one to five using the anchors below, and write the evidence in the next column. A score without evidence beside it cannot be defended to a sponsor later.
  4. Multiply each score by its weight and total the results. The weights are fixed, so you can compare two workflows scored by two different people.
  5. Apply the three hard gates before you look at the total, because they override it.

Note

Score what is true today, not what is planned

The single failure mode of this sheet is scoring the intended state. If you plan to clean up entitlements next quarter, Data scores what it is now, and the cleanup becomes a prerequisite with a date on it. A workflow that scores well only against a planned state will stall if the plan slips.

Score it now

Score your workflow in five rows

The quick form of the sheet below: five dimensions, five described lines in each, a total out of 25. Score the line that is closest to true today, not the one you hope for.

Pick one line per row. 1 is the weakest description, 5 the strongest.

ValueIs there a measurable baseline today, and an outcome someone has agreed to be judged on?
OwnershipIs there a sponsor who can fund a decision, and an operating owner who will hold this after go-live?
DataDo approved, current, entitled sources exist, and can access be granted inside the pilot window?
RiskIs the decision reviewable by a human who was already in the process, and is the impact containable?
EvaluationCan you assemble representative cases with known-good answers, and agree a threshold before you build?

Your answers stay in your browser until you fill all five rows. Then we receive the five numbers and a random session code, and nothing else.

0 of 25

0 of 5 rows answered

Pick one line in each row to see your verdict.

0 of 5 rows answered

On each card, 1 is the weakest description and 5 the strongest. Write down the number of the line closest to true today, then add the five scores for a total out of 25.

Value

Is there a measurable baseline today, and an outcome someone has agreed to be judged on?

  1. The benefit is described only in general terms. Nobody can say what the workflow costs today, and no system of record would show a change if it improved.
  2. A baseline is guessed rather than estimated, and nobody has agreed an outcome.
  3. A baseline exists, estimated rather than measured, and the team agrees the target outcome informally.
  4. The baseline comes from a system of record and the team has agreed a target, but nobody has yet signed up to be judged against it.
  5. A system of record measures volume, cycle time and rework, a named person has agreed to be judged against a target, and the measurement would survive audit.
Ownership

Is there a sponsor who can fund a decision, and an operating owner who will hold this after go-live?

  1. Interest sits with an innovation or transformation function with no operating owner behind it.
  2. A sponsor likes the idea but cannot fund a pilot yet, and nobody has agreed to run it.
  3. A sponsor can fund a pilot, but who runs it after go-live is unresolved.
  4. A sponsor holds the budget and an operating owner is named, but neither has confirmed that this becomes normal work.
  5. The sponsor holds the budget, the operating owner is named, and both have confirmed that this becomes part of the unit’s normal work.
Data

Do approved, current, entitled sources exist, and can access be granted inside the pilot window?

  1. The authoritative content is spread across repositories nobody has classified, and entitlements exist only in people’s heads.
  2. Sources are known, but most are out of date or unclassified, and nobody has asked who may see them.
  3. Sources are identified and mostly current, but access for a pilot requires a decision that nobody has made yet.
  4. The team has named and classified current sources and agreed access in principle, but nobody has granted it yet.
  5. Sources are named, classified, current, owned by someone who can approve access, and entitlements resolve against a directory.
Risk

Is the decision reviewable by a human who was already in the process, and is the impact containable?

  1. The output directly affects a person’s rights, benefits or eligibility, and there is no natural point at which a human already checks it.
  2. A review step is discussed but nobody owns it, and an error could reach the person affected.
  3. A review step exists but would need to be created or resourced for the volume involved.
  4. A qualified human reviews this output today, but the review is not recorded, and an error could be caught late.
  5. A qualified human already reviews this output as part of the job, the review is recorded, and an error is recoverable before it reaches anyone outside.
Evaluation

Can you assemble representative cases with known-good answers, and agree a threshold before you build?

  1. There is no source of truth, and correctness can only be adjudicated by one expert whose judgement is not written down anywhere.
  2. Past cases exist in scattered form, with no agreed right answers and nobody to settle them.
  3. Past cases exist and could become a test set with effort, but nobody has discussed a threshold.
  4. Representative cases with known-good answers can be assembled, but the pass threshold is not yet agreed with whoever signs off.
  5. You can pull a set of representative cases with known-good answers this month, and you know who would sign off on a pass threshold.

How to read the total

Worth building first
Every gate row holds and the total reaches 20. The work ahead is control design and evaluation, not persuasion.
Fix the data first
The total is under 20, or Value, Ownership or Risk is under 3. Start with the lowest rows, then score again.

The sheet

The seven dimensions

Print this section, or copy the table into whatever you already use. Weighted score is the score out of five multiplied by the weight, so a five on Value contributes 1.00 and the maximum total is 5.00.

Show the detail

Workflow scorecard. Score each dimension one to five against the anchors, record the evidence, and multiply by the weight to get the weighted score.
DimensionWeightThe question you are answeringScore1 to 5Evidence
Value20%Is there a measurable baseline today, and an outcome someone has agreed to be judged on?Score, to be completed by youEvidence, to be completed by you
Ownership15%Is there a sponsor who can fund a decision, and an operating owner who will hold this after go-live?Score, to be completed by youEvidence, to be completed by you
Data15%Do approved, current, entitled sources exist, and can access be granted inside the pilot window?Score, to be completed by youEvidence, to be completed by you
Risk15%Is the decision reviewable by a human who was already in the process, and is the impact containable?Score, to be completed by youEvidence, to be completed by you
Evaluation15%Can you assemble representative cases with known-good answers, and agree a threshold before you build?Score, to be completed by youEvidence, to be completed by you
Integration10%Can this reach the sources and the surfaces people already work in, without a new program of work?Score, to be completed by youEvidence, to be completed by you
Adoption10%Will the people doing the work use it, and is there a named change owner and a feedback loop?Score, to be completed by youEvidence, to be completed by you
Weighted total100%Sum of score multiplied by weight. Maximum 5.00.Total, to be completed by youNotes, to be completed by you

The sheet prints on one page from your browser. It submits nothing and stores nothing.

Anchors

What a one looks like, and what a five looks like

Anchors keep scoring consistent between people. Score against the described condition, not against how promising the workflow feels.

Show the detail

Value · 20%

Measurable baseline and an agreed outcome

Score 1
The benefit is described only in general terms. Nobody can say what the workflow costs today, and no system of record would show a change if it improved.
Score 3
A baseline exists, estimated rather than measured, and the team agrees the target outcome informally.
Score 5
A system of record measures volume, cycle time and rework, a named person has agreed to be judged against a target, and the measurement would survive audit.
Ownership · 15%

A sponsor who can fund and an owner who will hold it

Score 1
Interest sits with an innovation or transformation function with no operating owner behind it.
Score 3
A sponsor can fund a pilot, but who runs it after go-live is unresolved.
Score 5
The sponsor holds the budget, the operating owner is named, and both have confirmed that this becomes part of the unit’s normal work.
Data · 15%

Approved, current, entitled sources

Score 1
The authoritative content is spread across repositories nobody has classified, and entitlements exist only in people’s heads.
Score 3
Sources are identified and mostly current, but access for a pilot requires a decision that nobody has made yet.
Score 5
Sources are named, classified, current, owned by someone who can approve access, and entitlements resolve against a directory.
Risk · 15%

A human review step that already exists

Score 1
The output directly affects a person’s rights, benefits or eligibility, and there is no natural point at which a human already checks it.
Score 3
A review step exists but would need to be created or resourced for the volume involved.
Score 5
A qualified human already reviews this output as part of the job, the review is recorded, and an error is recoverable before it reaches anyone outside.
Evaluation · 15%

Known-good answers and an agreed threshold

Score 1
There is no source of truth, and correctness can only be adjudicated by one expert whose judgement is not written down anywhere.
Score 3
Past cases exist and could become a test set with effort, but nobody has discussed a threshold.
Score 5
You can pull a set of representative cases with known-good answers this month, and you know who would sign off on a pass threshold.
Integration · 10%

Reachable sources, familiar surfaces

Score 1
It requires writing back into a system of record, or depends on another team’s roadmap landing first.
Score 3
Read access is feasible, but users would work in a new place rather than where they are today.
Score 5
Read access to approved sources is available now, and the output lands in a surface people already have open.
Adoption · 10%

A named change owner and a feedback loop

Score 1
The user group was not consulted before the decision, and no one owns the change.
Score 3
Users are supportive in principle, and change management is assumed rather than assigned.
Score 5
A change owner is named, users helped shape the scope, and there is a route by which a complaint reaches someone who can act on it.

Note

Where the instrument comes from

The seven dimensions and their weights are White Glove Labs’ own instrument, not a published standard and not a regulatory requirement. They are our statement of what decides whether a knowledge workflow can be governed and delivered. We publish the weights so you can adapt or challenge them.

Reading it

What your total means, and the three gates that override it

Read the gates first. A workflow can total well and still be one we would decline, and the gate is the reason.

Show the detail

Warning

The three hard gates

A score below three on Value, Ownership or Risk stops the workflow regardless of the weighted total. No measurable baseline means you cannot tell afterwards whether it worked. No operating owner means it has no home after the pilot team leaves. No existing human review step on a decision that affects someone means the control problem is larger than the delivery problem. Solve that problem before any agent is involved.

4.0 and above
A strong candidate. The work ahead is control design and evaluation, not persuasion. This is a workflow that can reasonably be scoped to a bounded pilot with an acceptance threshold.
3.0 to 3.9
A candidate with named prerequisites. One or two dimensions are dragging, and they are fixable on a known timeline. Fix them first, then re-score; do not start a build against the intention to fix them.
2.0 to 2.9
Not yet. Something structural is missing, such as data classification or an owner. The remediation is real work and deserves its own decision, separate from any AI decision.
Below 2.0
Choose a different workflow. The exercise has ruled out a build that would have failed later, and the dimensions show what a stronger first candidate would look like.

If two candidate workflows are close, score both and compare the dimensions rather than the totals. The workflow with a five on Data and a two on Adoption is a different project from its mirror image, even at the same total.

On the other side of the table

Run the same sheet on whoever is selling to you

The dimensions are also a supplier test. A supplier who cannot answer these about your workflow is proposing a build without knowing whether it should exist.

Show the detail

The same seven dimensions as supplier questions, with an example of a weak answer.
DimensionAsk themA weak answer sounds like
ValueWhat baseline would you measure against, and where does that number come from?A percentage improvement from another client, quoted without a method or a date.
OwnershipWho owns the system on the day your engagement ends, and what is the handover artifact?A support retainer, with no handover artifact named.
DataHow are entitlements enforced at query time, and what happens when access is revoked? Describe the test that proves a restricted document cannot appear in a citation.An assurance that the model “only sees what it is given”, with no test named.
RiskWhich decisions stay with a human, and how is that enforced in the system rather than in the training? Does the Directive on Automated Decision-Making apply here?Reassurance in place of reasoning, or an answer that treats the Directive as a legal question for someone else.
EvaluationWhat is the acceptance threshold, who signs it, and what does the system do when it is not met after go-live?Nothing specific, which means you are being asked to accept the output on trust.
IntegrationWhat has to be true in our environment before week one? In writing.A verbal “we will work with whatever you have”. No prerequisites named in writing.
AdoptionWhat happens if the users do not use it, and is that a reason to stop?No stopping condition, so the engagement is a rollout rather than a pilot.

Note

Including us

Ask us the same seven. Our answers are on the site rather than in a meeting. We publish the sprint deliverables and scoring method, and how we accept or decline a pilot against a threshold. We also publish the control domains that the design follows. The supplier evidence pack sets out what we can and cannot currently evidence, and we send it to a contracting authority on request. Where we hold no certification or clearance, that pack says so.

Next step

If you want a second opinion on your score

You do not need us to use this. If you want the score pressure-tested, the call is thirty minutes and the notes are yours either way.

Bring the completed sheet to a 30-minute call. We will score the same workflow independently, show you where our score differs from yours and why, and give you a plain answer on whether it is worth governing.

Paste your scores into the workflow field on the form, or say that you have a completed scorecard and we will ask for it. It is not a prerequisite: a described workflow is enough.

Book a 30-minute callSee the full ten-day sprint

If you have not chosen a workflow yet, start with five knowledge workflows Canadian public institutions can pilot safely. It starts from where the Directive draws its line rather than from what the technology can do.