The best place to start with AI is usually not the task that sounds most futuristic. It is a repeated, valuable part of the business that has a clear owner, usable examples, safe boundaries, and a result you can measure.
That is the purpose of an AI readiness assessment. It helps you compare real work before you compare products. The output should be a decision: which workflow deserves attention first, what kind of system fits it, and what must be true before you build.
This matters because access to AI is no longer the main constraint. AI features are already built into email, customer systems, accounting tools, office software, and project platforms. The harder questions are operational:
- Which problem is worth solving?
- What company knowledge does the system need?
- Where should a person review the work?
- Who will own the process after launch?
- How will you know whether it helped?
Start with the business result
A weak assessment begins with, “Where can we use AI?” That question produces a long list of ideas and very little direction.
A stronger assessment begins with a visible business problem:
- Quotes take too long to prepare.
- Client handoffs lose important details.
- A senior employee answers the same internal questions every week.
- Follow-up depends on the owner remembering to do it.
- Reports require information from several systems and repeated checking.
Then connect the problem to a result. The result might be faster response, more completed work, less rework, better preparation, or fewer missed handoffs. Revenue can matter, but it is rarely the only useful measure. A workflow that returns scarce owner time or removes a capacity bottleneck can support growth even when it does not create revenue directly.
The NIST AI Risk Management Framework starts its Map function by asking organizations to define the business value and context of use. That is good practical advice, not just risk guidance. If the value and operating context are unclear, the technology decision is early.
Use this six-part opportunity scorecard
List three to five workflows that consume time, delay work, or create repeated friction. Score each one from 0 to 3 on the six factors below. This is a discussion tool, not a scientific formula. Its purpose is to make assumptions visible.
| Factor | 0 points | 1 point | 2 points | 3 points |
|---|---|---|---|---|
| Business value | No clear effect | Minor convenience | Useful capacity or quality gain | Direct link to an important goal or bottleneck |
| Repetition | Rare or unpredictable | Monthly | Weekly | Daily or high volume |
| Process clarity | Nobody agrees on the steps | Mostly tribal knowledge | Core path is understood | Steps, exceptions, and owner are clear |
| Knowledge readiness | Inputs are missing or unreliable | Information is scattered and stale | Most sources exist | Approved sources and examples are available |
| Safe boundaries | Consequences are hard to contain | Many sensitive decisions | Review points can contain risk | Work can begin read-only or draft-first |
| Measurability | No baseline or observable result | Only opinions | One useful measure exists | Time, quality, volume, or value can be compared |
Compare the totals, but do not simply select the highest number. First look for any zero in process clarity, knowledge readiness, safe boundaries, or measurability. A valuable workflow may still be a good future candidate, but a zero often identifies preparation that should happen before implementation.
For example, proposal preparation may score well because it repeats, uses known source material, and ends in a human review. Contract negotiation might have high value, but its exceptions and consequences make it a poor first assignment. Cleaning up customer records might be easy to automate, but low value if nobody uses those records to make decisions.
The right first project sits where meaningful value and practical readiness overlap.
Check five readiness gates before you build
A scorecard compares opportunities. These five gates determine whether the leading opportunity is ready to move forward.
1. A person owns the workflow
Someone must be able to explain what good work looks like, answer questions, and decide how exceptions should be handled. The owner does not need to be technical. They need to understand the job.
Without an owner, a specialized AI assistant inherits conflicting preferences from whoever happens to review its latest output. That creates rework instead of capacity.
2. Representative work exists
Gather several real inputs and completed outputs, including imperfect cases. Examples show the difference between the documented process and the process people actually follow.
Useful materials might include approved proposals, intake forms, meeting notes, checklists, reports, policies, and corrections from prior reviews. Do not upload everything by default. Start with information that is relevant, current, and approved for the intended use.
3. Access can be limited
Identify which systems the assistant needs to read and which, if any, it needs to change. A first version can often prepare work without sending, publishing, deleting, spending, or changing a business record.
Read-only access and draft-first operation make early testing easier to observe and easier to stop. More authority can be considered after the system has produced evidence in the real workflow.
4. Review boundaries are explicit
Write down what the assistant may complete, what always needs approval, and when it must ask for help. A boundary such as “prepare the weekly account brief from these approved sources, then send it to the account owner for review” is more useful than “help with client work.”
The OpenAI guide to building agents recommends human intervention when failure limits are reached and before sensitive, irreversible, or high-stakes actions. Those controls should be part of the workflow design, not added after an uncomfortable surprise.
5. A baseline can be recorded
Before changing the process, record how it works now. Depending on the job, that can include:
- Minutes of active work per completed item
- Total cycle time from request to completion
- Number of items completed in a week
- Review corrections or rework
- Missed deadlines or handoffs
- Response time to a customer or team member
Use measures the business can observe without creating a reporting project. The purpose is to learn whether the new approach is useful, not to manufacture an impressive case study.
Do not confuse tool access with readiness
The U.S. Small Business Administration advises businesses to start small, test whether a tool adds value, and have another person review AI-generated work. Its practical AI guidance also notes both the efficiency opportunity and the risks of using AI in daily operations.
Starting small does not mean choosing a trivial task. It means limiting the first responsibility enough that the team can teach it, review it, and measure it.
The distinction is important because business adoption is still uneven. A 2026 U.S. Census Bureau working paper found that 18 percent of firms used AI in a business function during the survey period. Among firms using AI, 57 percent used it in three or fewer functions. Those figures do not prove why any specific project succeeds or fails. They do show that access and isolated use are not the same as broad operational integration.
The OECD's 2025 report on AI adoption by small and medium-sized enterprises similarly identifies different adoption paths based on a firm's maturity and the complexity and scope of its use. A useful assessment should respect that context rather than prescribe the same AI stack to every company.
What to bring to a 15-minute AI assessment
You do not need a technology plan. Bring one repeated or frustrating part of the business and enough detail to discuss it:
- What starts the work?
- Who does it now?
- Which information and systems are involved?
- Where does judgment matter?
- What tends to slow down, break, or return to the owner?
- What would be observably better?
That conversation may lead to a simple software change, a fixed automation, a company knowledge project, a specialized AI assistant, or no AI project at all. A credible assessment should preserve all of those outcomes.
If the opportunity needs deeper analysis, an AI Opportunity Audit can map the workflow, compare options, define review boundaries, and recommend what to do first. You can also book a free 15-minute AI assessment and bring one bottleneck to work through.
The goal is not to leave with more AI ideas. It is to leave with a better decision.
Sources
- NIST, AI Risk Management Framework Core
- U.S. Small Business Administration, AI for small business, updated February 14, 2025
- OECD, AI Adoption by Small and Medium-Sized Enterprises, December 9, 2025
- U.S. Census Bureau, The Microstructure of AI Diffusion, 2026
- OpenAI, A Practical Guide to Building AI Agents