How to Find and Prioritize AI Opportunities in Your Business

3. Describe each candidate as a workflow

An idea such as “use AI for customer service” is too vague to evaluate. Rewrite every candidate as a workflow with the same parts, so you can compare them fairly:

• Trigger: what starts the work, such as a form submission, an email, a call, a new order, or a scheduled date

• Inputs: the information needed, where it lives, and whether it is structured data or free-form language

• Decisions: which steps follow fixed rules and which require judgment

• Owner: the person accountable for the outcome and for exceptions

• Output: what done looks like, and which system records it

• Systems: the tools the workflow touches, and whether they can share data

This step often reveals that the real problem is not intelligence at all. It is a missing owner, a missing field in a form, or two systems that do not talk to each other. Those are still valuable findings, and they are usually cheaper to fix.

4. Score candidates on value, feasibility, and risk

Use one simple scorecard for every candidate. A 1 to 5 scale is enough. The point is consistent comparison and a written rationale, not false precision. A reasonable set of criteria:

• Value: how directly the workflow affects revenue, cost, speed, error rates, or customer experience

• Volume: how often it happens, because frequent work compounds small improvements

• Data readiness: whether the inputs exist, are accessible, and are reasonably clean

• Process stability: whether the steps are agreed and repeatable, or still changing every month

• Integration effort: how many systems must connect, and whether they offer supported integrations

• Risk: what a wrong answer or a missed step would cost, and who could be affected

• Measurability: whether you can capture a baseline now and compare results later

Weight value and risk more heavily than novelty. A modest workflow that runs hundreds of times a month, with clean inputs and low error cost, is often a better first project than an ambitious idea that depends on data you do not yet have.

5. Match the opportunity to the right kind of solution

Not every opportunity needs AI. Choosing the simplest approach that works keeps cost, maintenance, and risk down. As a rough guide:

• Fixed steps and structured inputs usually suit conventional workflow automation and system integration

• Unstructured language, such as emails, documents, chats, or call notes, is where AI models can add value by classifying, extracting, summarizing, or drafting

• Repeated conversations with clear boundaries may suit an AI agent, with defined handoffs to a person

• Problems caused by disconnected tools are often solved by integration first, before any AI is added

• Unclear goals, ownership, or data are a strategy problem, and building software will not fix them

Many useful projects combine these. For example, an integration moves an inquiry into the right system, an AI step extracts the key details, and a rule assigns it to the owner with a deadline. Keep a person responsible for anything that involves judgment, commitments, pricing, or regulated advice.

6. Set guardrails before the pilot, not after

The NIST AI Risk Management Framework is intended for voluntary use. Its core organizes AI risk management into four functions (govern, map, measure, and manage), and several of its outcomes translate directly into practical questions for a first project (NIST AI RMF Core):

• Is the business value or context of use clearly defined? (Map 1.4)

• Have the potential costs of errors, including non-monetary costs, been examined and documented? (Map 3.2)

• Are processes for human oversight defined? (Map 3.5)

• Is there a decision on whether the system achieves its intended purpose and whether deployment should proceed? (Manage 1.1)

In practice, that means writing down what the system may and may not do, what data it can access, who reviews its output, how a person can override it, and when you will switch it off. Keep personal and sensitive information to the minimum the workflow needs, and confirm how any vendor stores and uses your data before you connect it.