Does My Company Need AI? A Readiness Test

AI becomes a sensible business decision when a specific process has repeatable work, accessible data, a measurable outcome and a person who owns the result. A company-wide ambition is too broad for a useful first decision.

Choose one process such as invoice intake, support triage, reporting or quote preparation. Answer each question for that process. A yes earns one point; an uncertain answer earns zero until the missing evidence is available.

Process signals

1. Can you name the process as one unit of work?

Write the input, the main steps and the output in one paragraph. “Improve operations” is a theme; “classify inbound support cases and route them” is a process.

2. Do the main steps repeat?

Look for recurring inputs, fields and decisions. A process that changes shape every time needs mapping before automation.

3. Can the rules and exceptions be shown?

Ask the person doing the work to describe ordinary cases and the cases that require escalation. Record the boundary in plain language.

4. Is there enough activity to observe?

Use your own queue or work records. You need enough recent cases to see normal variation, errors and exceptions; a rare task may be better served by a checklist.

Value signals

5. Can you measure the current effort?

Record time spent, rework, waiting, handoffs or another operational measure. A rough internal baseline is more useful than an industry benchmark.

6. Can you name the outcome that should improve?

Choose one primary measure and a guardrail. For example, time to first useful action paired with factual-error review, or routing time paired with escalation quality.

7. Can the team compare the new flow with the current one?

Keep a sample of current cases and agree how reviewers will judge the result. A comparison plan makes the decision observable.

Ownership signals

8. Is there a process owner?

Name the person who can change the rules, approve a release and accept an escalation. Technology ownership alone is insufficient.

9. Can the required data be accessed lawfully and technically?

List the systems, fields, permissions, retention rules and data steward. A missing permission belongs in discovery work before prompt design.

10. Will someone maintain the workflow?

Assign responsibility for source changes, evaluation, incidents, costs and user feedback. Production work needs an owner after launch.

Read the score as a triage signal

The score is a routing aid. A high score still needs a feasibility review, and a low score can improve after the process is documented. Use the weakest block to choose the next piece of preparation.

When a manual workflow is the right choice

Keep the work manual or use a checklist when the task is rare, the rules are unsettled, the source data is inaccessible or the cost of an error exceeds the value of faster handling. A small process may benefit more from clearer ownership than from a new system.

Pause a project when no one can approve the scope, reviewers cannot access representative cases or the team has no path for exceptions. Those are decision conditions to fix before implementation.

Prepare one evidence pack

Bring these items to a discovery call:

• a short process map and the named owner;
• a de-identified case sample with common exceptions;
• current effort and quality measures;
• systems, permissions and data-retention constraints;
• the proposed outcome measure and guardrail;
• the action the workflow may perform and the action requiring approval.

The result should be a mechanism choice: a checklist, fixed automation, retrieval assistant, agent workflow or a decision to wait. The test helps you enter that conversation with evidence.

Discuss one process with the Syntalith team or review AI process audits.