Many Brussels operations teams can see where AI might help, but they cannot yet tell whether a process is ready to automate. A customer inbox is busy, finance is overloaded, sales follow-up is inconsistent, or managers keep asking the same reporting questions. The temptation is to test a tool immediately. The safer question is more practical: is this workflow ready for AI, ready for conventional automation, or not ready for either?
This scorecard gives operations managers and transformation sponsors a simple way to assess one workflow before committing budget. It is designed for Belgian SMEs and Brussels-based teams that need useful implementation guidance, not a generic AI checklist. The goal is to make the decision explicit: proceed to a controlled pilot, fix specific gaps first, or redesign the workflow before adding automation.
How to use the scorecard
Score one workflow at a time. Do not score a department, a tool, or an ambition such as "use AI for customer service." A workflow has a trigger, inputs, decisions, handoffs, systems, exceptions, and a final record. "Handle inbound support emails for Belgian customers until the ticket is resolved or escalated" is a workflow. "Improve service" is not.
Use six categories worth up to 100 points: volume and value, process clarity, data quality, integration readiness, risk controls, and ownership. A score of 75 to 100 means the workflow may be ready for a controlled pilot. A score of 50 to 74 means there is potential, but the gaps should be fixed before build. A score below 50 usually means process redesign, data cleanup, or ownership work should come first.
The score is not a legal approval or a guarantee that AI is the right tool. It is an operating decision aid. NIST's AI Risk Management Framework is useful here because it treats AI risk as something to map, measure, manage, and govern. For a Brussels SME, that means looking at the process, the data, the people, and the controls before choosing a model or automation platform.
Category 1: volume and value
Give up to 20 points for volume and value. A workflow scores highly when it happens often enough to matter, consumes meaningful staff time, delays customers or revenue, or creates avoidable rework. It scores lower when it is rare, strategic, highly bespoke, or already handled efficiently by a small number of experts.
Good evidence includes monthly transaction count, average handling time, waiting time, number of escalations, missed follow-ups, cost of delay, and rework. Brussels teams often start with a visible pain point, but visible pain is not always the highest-value workflow. A manager may feel the reporting process is annoying, while the larger operational cost is hidden in invoice exceptions or customer handoffs. Measure before ranking.
A practical scoring rule is simple. Award 15 to 20 points when the workflow is frequent, measurable, and connected to cost, revenue, risk, or customer experience. Award 8 to 14 when the value is plausible but not yet measured. Award 0 to 7 when the workflow is too rare or unclear to justify an implementation.
Category 2: process clarity
Give up to 15 points for process clarity. A workflow is ready only when people can describe how work moves today. That includes the trigger, the source of truth, required decisions, approval points, handoffs, exceptions, and final outcome. If the process changes depending on who is working, AI will inherit that ambiguity.
Use a short process map. List every system involved, including email, spreadsheets, CRM, ERP, finance tools, shared drives, and chat channels. Then mark where a human decision is required. The most useful automation opportunities are not always fully automatic. Many strong first pilots are human-in-the-loop workflows where AI drafts, extracts, classifies, or recommends, while a named person approves the final action.
Score 12 to 15 when the workflow is documented and exceptions are known. Score 6 to 11 when the main path is clear but exceptions or ownership are not. Score 0 to 5 when the team cannot agree how the process currently works.
Category 3: data quality and access
Give up to 20 points for data quality and access. AI workflows depend on reliable inputs: documents, records, messages, product data, customer data, knowledge articles, or historical tickets. The question is not whether data exists. The question is whether the data is current, accessible to the right roles, structured enough to use, and safe to process.
For a knowledge assistant, check whether permissions in SharePoint, Google Drive, or another source system reflect real access rules. For invoice intake, check whether supplier names, purchase orders, approval rules, and accounting fields are consistent. For sales follow-up, check whether CRM stages, owners, consent, and previous interactions are reliable. If the workflow uses personal data, include privacy review and retention rules in the readiness work.
Score 15 to 20 when the source of truth is known, permissions are clean, and representative samples are available for testing. Score 8 to 14 when the data is usable but needs cleanup. Score 0 to 7 when critical inputs are scattered, outdated, or over-permissioned.
Category 4: integration readiness
Give up to 15 points for integration readiness. A workflow that only drafts a response has different requirements from one that writes to a CRM, updates an accounting system, sends an email, and logs an approval. Each integration brings authentication, field mapping, rate limits, error handling, and rollback questions.
Check whether the required systems have APIs or supported connectors, whether those connectors expose the objects and fields you need, and whether sandbox testing is possible. Also decide what happens when a system is unavailable. Production workflows need retries, error queues, and clear ownership for failed actions.
Score 12 to 15 when the needed integrations are available, documented, and testable. Score 6 to 11 when some integrations are possible but mapping or permissions are uncertain. Score 0 to 5 when the workflow depends on manual exports, brittle screen scraping, or unsupported access.
Category 5: risk controls
Give up to 20 points for risk controls. The EU AI Act and broader governance expectations make one point clear for operational teams: AI should be deployed with attention to purpose, transparency, data, oversight, and accountability. A Brussels SME does not need a heavyweight policy for every small pilot, but it does need proportionate controls.
Name the decisions the system may make, the decisions it may only recommend, and the decisions it must escalate. Define the human approval gate, the audit log, the refusal rules, and the monitoring owner. For customer-facing workflows, decide how users are informed and how they reach a person. For internal workflows, decide how employees report wrong answers or unsafe suggestions.
Score 15 to 20 when approval gates, logs, privacy boundaries, and escalation rules are clear. Score 8 to 14 when risks are known but controls are incomplete. Score 0 to 7 when the workflow could affect customers, money, legal obligations, or personal data without named oversight.
Category 6: ownership and measurement
Give up to 10 points for ownership and measurement. Every AI workflow needs an accountable business owner, a technical owner, and a review rhythm. It also needs a baseline. Without those, the pilot becomes a tool experiment instead of an operating improvement.
Define one primary metric and a few guardrails. For a support workflow, the primary metric might be time to first useful response; guardrails could include escalation accuracy, customer satisfaction, and correction rate. For a finance workflow, the primary metric might be invoice cycle time; guardrails could include duplicate detection, approval compliance, and exception rate.
Score 8 to 10 when ownership, baseline, and review cadence are defined. Score 4 to 7 when a manager supports the idea but measurement is weak. Score 0 to 3 when no one is accountable after launch.
Interpreting the result
If the workflow scores 75 or above, move to a small production pilot. Keep the scope narrow, include real users, and build the control path from the start. If the score is 50 to 74, resist buying a tool immediately. Fix the lowest-scoring categories first. That may mean cleaning CRM stages, documenting exceptions, reviewing permissions, or confirming API access. If the score is below 50, the right next step is process redesign. Automating too early will make the mess faster and harder to diagnose.
Google's people-first content guidance is about publishing useful, reliable material for people rather than search engines. The same principle applies to internal AI work. A readiness scorecard is useful when it helps a team make a better operating decision. It should not become paperwork that justifies a decision already made.
Intyb uses this kind of assessment before building workflow automation and custom AI solutions for Brussels teams. For the cautionary side of the same decision, read why bad process automation costs more, or contact Intyb's Brussels implementation team to review one workflow.