AI Readiness Assessment: Is Your Business Ready for Agents?

An AI readiness assessment tells you whether your business can put an AI agent to work now, or what has to be fixed first. For agents, readiness comes down to seven things: a specific job with real volume, rules that are written down, data the agent can reach through a system rather than someone's inbox, data clean enough to act on, a named owner, clear limits on what the agent may do alone, and a baseline to measure against. Score each from 0 to 2. A total of 10 or more out of 14 means you can start building a first agent this quarter. Between 6 and 9, you are four to eight weeks of preparation away. Below 6, fix the foundations before you spend money on an agent. The checklist below takes about 30 minutes with the person who runs the job.

This post is part of our guide to rolling out agents, which starts at wizeb.com/blog/how-to-implement-ai-agents-in-business. Run this check before you choose a first agent with wizeb.com/blog/first-ai-agent-for-business, and before the 90-day plan in wizeb.com/blog/ai-agent-implementation-roadmap.

The Problem: Agent Projects That Stall Before They Start

Most AI agent projects don't fail because the model is weak. They stall in week three, when the build team asks where the pricing rules are written down and the answer is "Sarah knows". Or when the customer data turns out to live in three spreadsheets and a CRM nobody updates. Or when legal sees the plan for the first time and asks who approves what the agent sends to customers.

Each of those problems is cheap to find before a project and expensive to find during one. An agent build that stops to clean up data or write missing rules typically runs 30 to 60% over budget, and the delay costs the sponsor's patience, which is harder to win back than money. Generic AI readiness surveys don't catch this because they ask about strategy, culture and "data maturity" in the abstract. Agents need something more concrete: one job, the information that job depends on, and the permission to act.

The 7-Point AI Readiness Checklist

Pick one candidate job, ideally one your team does every week, and score it with the person who actually does it. Score 0 if the statement is false, 1 if it is partly true, and 2 if it is fully true today. Be strict: "we could do that" scores 0.

1. One job with clear volume

You can name the job in one sentence ("match supplier invoices to purchase orders") and say how often it happens from a system count, not a guess. A job that happens 200 times a month is a candidate. "Improve customer experience" is a goal, not a job, and scores 0.

2. The rules are written down

Someone new could do the job from a document: the steps, the exceptions, when to escalate and who to ask. Rules that live in one experienced person's head are the most common gap we see. An agent can't follow knowledge nobody has written down.

3. The data is reachable

The information the job needs sits in systems with an API, an export or at least a structured shared location: your CRM, accounting software, help desk, order system. Data in personal inboxes, paper files or screenshots scores 0 until it is moved. Our guide at wizeb.com/blog/train-ai-agent-on-company-data covers what to connect first.

4. The data is good enough to act on

Not perfect, just reliable for this job. Take 20 recent records and check them: are customer names, prices, product codes and statuses correct and current? If more than two or three are wrong, the agent will repeat those mistakes faster and more confidently than a person would.

5. An owner and a sponsor

One person owns the job day to day and will review the agent's work, answer its edge cases and sign off on changes. A senior sponsor will clear blockers and decide at day 90 whether to scale. Without an owner, nobody tunes the agent and its quality drifts. Without a sponsor, the project loses priority the first time it hits a snag.

6. Clear limits on what the agent may do

You can say which actions the agent may take alone, which need a person's approval, and which it must never take. For example: draft replies freely, send standard replies alone, never issue refunds above $200. If nobody can answer that yet, it isn't a technology gap; it's a decision someone senior needs to make. Our guide at wizeb.com/blog/ai-agent-security-guide covers permissions in detail.

7. A baseline to measure against

You know, or can measure in two weeks, how long the job takes today, what it costs and how often it goes wrong. Without that before-number, you can't prove the agent works, and the project will be judged on opinions. The method is in wizeb.com/blog/how-to-measure-ai-agent-roi.

How to Score Your AI Readiness Assessment

Add up the seven scores for a total out of 14, then read it against these bands.

• 12 to 14 — Ready now. Start scoping. A first agent for this job can usually be in production within six to ten weeks.
• 10 to 11 — Ready, with one gap. Close the gap during the build, typically by writing the missing rules or setting the approval limits in the first two weeks.
• 6 to 9 — Nearly ready. Spend four to eight weeks fixing the zeros first. This is normal; most businesses we assess for the first time land here.
• 0 to 5 — Not ready for this job. Either pick a different job that scores higher, or invest in the foundations (documented processes and connected systems) before any agent work.

Score two or three candidate jobs, not one. Readiness often differs more between jobs inside the same business than between businesses. A company with messy sales data may have a perfectly clean accounts-payable process, and that is where its first agent belongs.

What to Fix First When You Score Low

The gaps cluster into a few patterns, and each has a fairly predictable fix.

• Rules in people's heads (item 2): record the job owner talking through 10 real cases, then turn the recording into a written procedure. One to two weeks, and useful even if you never build an agent.
• Data in inboxes and spreadsheets (items 3 and 4): move the job's data into one system of record and clean the fields the job depends on. Two to six weeks, depending on volume.
• No owner (item 5): assign the person who already does the job and protect two to three hours a week of their time for the first three months.
• No agreed limits (item 6): a one-hour meeting with the sponsor, the owner and whoever carries the risk (finance, legal or compliance) to agree the act-alone, approve and never lists.
• No baseline (item 7): two weeks of timing real work in a shared spreadsheet, as described in our ROI guide.

None of these fixes need AI expertise. They need a decision and a few weeks of focused effort, which is why it pays to do them before you hire anyone to build.

Beyond One Job: Company-Wide AI Readiness

The checklist above answers "can we build this agent?". Leadership teams usually also want to know "how ready is the business as a whole?". That depends on six broader areas: how well your processes and knowledge are documented, how much your people already use AI day to day, whether engineering uses agents or just autocomplete, how far operations are automated, whether sales and marketing run at machine scale, and whether you can trace what agents do.

Our free AI-Native Index at wizeb.com/ai-native-index scores exactly those six areas with 30 questions and gives you a score out of 100 with a breakdown per area. It takes about ten minutes, and it is a useful way to see which department to start with before you run the job-level checklist.

Case Study: From a Score of 6 to a Live Agent in 11 Weeks

Consider an 80-person property management company that wanted an agent to handle tenant maintenance requests, about 900 a month arriving by email, phone and WhatsApp. Its first readiness assessment scored 6 out of 14. Volume was clear (2) and there was an owner (2), but the triage rules lived with two coordinators (0), requests were scattered across three inboxes (0), contractor records were half out of date (1), nobody had agreed which repairs the agent could book alone (0), and there was a rough baseline from the help desk (1).

The team spent five weeks on preparation. The coordinators recorded themselves triaging 40 real requests, which became a four-page procedure. All channels were routed into one help desk. The contractor list was cleaned. The operations director agreed that the agent could book routine repairs under $300 alone, while anything involving gas, water damage or a cost above that went to a coordinator. A re-score came out at 12. The agent went live six weeks later, handling 55% of requests end to end by day 60 and cutting average time to book a repair from 26 hours to under 4. This is an illustrative example, but the pattern is typical: the preparation took almost as long as the build, and it was the reason the build went smoothly.

How WizeB Runs an AI Readiness Assessment

WizeB starts every AI agent engagement with this assessment. In a free scoping call we score your two or three best candidate jobs with the people who run them, flag the blockers, and tell you plainly whether to build now, prepare first, or pick a different job. When you are ready, we build the agent on your own systems with the limits and review steps you agreed, and you own everything we build. See how we work at wizeb.com/services/ai-agents.