By Jared Feder, REL8TION founder and business strategist.
The best place to add AI is rarely the place with the flashiest demo. It is the point in a real workflow where information repeatedly arrives, a decision must be made, and preventable friction slows the next useful action.
Before buying software or automating a process, run a focused workflow audit. These seven questions help separate a valuable operating improvement from an expensive experiment.
1. What measurable problem are we solving?
“Use AI” is not a business objective. Define the operational problem in plain language:
- Qualified inquiries wait too long for a response.
- Employees spend hours reconstructing conversations.
- Important follow-up tasks are not consistently created.
- Customer questions repeat but the answers are scattered.
- Managers cannot see where a process is getting stuck.
Attach a baseline. Measure response time, manual hours, error rate, completion rate, abandonment, or another outcome that matters. Without a baseline, a business cannot tell whether the automation helped.
2. What actually happens from start to finish?
Map the current workflow before designing the future one. Identify where information originates, which systems it enters, who reviews it, which decisions repeat, where approval is required, and what outcome completes the process.
This often reveals that the main problem is not a lack of AI. It may be an unclear owner, duplicate data entry, a broken handoff, inconsistent labels, or an unnecessary step. Fix simple process problems with simple tools.
3. Which parts require judgment?
Separate repetitive assistance from consequential decisions. AI can be useful for classifying an inquiry, drafting a response, summarizing a conversation, or suggesting the next task. It should not quietly make a consequential decision when accuracy, fairness, compliance, or customer trust requires accountable human judgment.
Define the approval boundary in advance. State who reviews the output, what evidence they can inspect, and when the system must stop and escalate.
4. What data does the workflow need?
List the minimum data required for the task. Then ask:
- Do we have permission to use it this way?
- Is any of it sensitive, regulated, or confidential?
- Where is it stored and for how long?
- Which vendors can access it?
- Can the workflow operate with less data?
- How will inaccurate or outdated information be corrected?
The National Institute of Standards and Technology’s AI Risk Management Framework is a useful voluntary reference for incorporating trustworthiness considerations into the design, use, and evaluation of AI systems.
5. What happens when the output is wrong?
Every automated workflow needs an error plan. Consider the cost of a false positive, a false negative, an invented detail, a missed exception, or a message sent to the wrong person.
Low-risk outputs may need a quick human review. Higher-risk workflows may require source links, structured validation, approval before action, detailed logs, and a clear way to reverse or correct the result.
The more consequential the outcome, the less appropriate it is to rely on an unverified answer.
6. Can we test one valuable loop first?
Do not begin with a vague company-wide “AI transformation.” Choose one bounded loop with a clear owner and a measurable result.
A practical first loop might be:
- An inquiry arrives through an approved channel.
- The system identifies the topic and drafts a summary.
- A person reviews the source and approves the next action.
- The correct task or response is created.
- The outcome is logged for measurement.
Run it with a small group, study the exceptions, and improve reliability before expanding.
7. How will people stay in control?
Useful automation should make responsibility clearer, not hide it. Users should know what the system did, what information influenced the result, what still requires approval, and how to override or correct the workflow.
Human control is not a decorative promise. It needs to appear in permissions, interface design, audit logs, escalation rules, and the everyday operating process.
A simple scorecard for the opportunity
Before implementation, score the proposed workflow from one to five on:
- Frequency of the problem
- Time or revenue currently lost
- Quality and availability of required data
- Ability to verify the output
- Clarity of human ownership
- Ease of measuring the result
- Cost and consequence of failure
The best early opportunities are frequent, measurable, reversible, and easy to supervise. High-consequence decisions with weak data and unclear ownership should not be the first experiment.
Build the operating discipline before the operating system
AI multiplies the process around it. If the process is clear, measurable, and accountable, that leverage can be valuable. If the process is confused, automation can make the confusion move faster.
For more examples, read Practical AI Automation for Real Estate and Mortgage Businesses, explore the direction behind REL8TION OS, or review my AI and business consulting.
This article provides general business and technology information. Appropriate controls depend on the workflow, data, industry, laws, contracts, and risks involved. Obtain qualified legal, compliance, security, or other professional advice when required.
