The AI Readiness Audit: 12 Questions That Find Your Highest-ROI Automations

The best audit questions never mention AI, because asking people about AI makes them pitch use cases, while asking about annoying rote tasks makes them name real problems. Tenex, a US AI transformation firm run by Morning Brew co-founder Alex Lieberman and former Google engineer Arman Hezarkhani, learned this across hundreds of company audits and published the method on their show Human in the Loop. Reviewing their own employee survey, Hezarkhani noted: "I don't think the word AI is mentioned in a single question." The reason is behavioural. "People immediately jump to brainstorming AI solutions," says Lieberman. "So much of the work we do is pushing people to talk about problems, because at the end of the day, AI is just a tool for problem solving." Left alone, staff treat AI like a hammer and every task like a nail, and you end up with a wish list of tools rather than a list of problems worth solving (Tenex, 2026). There is a second, practical reason. Most automation is not AI at all: in Tenex's experience, nine of the ten steps in a typical AI solution are traditional code and integrations, with maybe one step of actual AI (Tenex, 2026). A problem-first audit finds the work worth doing. The tool choice comes later. Hezarkhani's instruction to executives in their stakeholder interviews is to forget about AI entirely and ask: "What keeps you up at night about work? What are your goals and what's keeping you from those goals?"

You run the audit in two parts: a 15 minute written survey sent to every employee, then 30 minute interviews split evenly between executives and staff whose survey answers flagged real problems. Here is the one-week plan. 1. Monday: send the survey to everyone. Cap it at 15 minutes to complete. Tenex found people will give you at most 15 minutes, so every question has to earn its place (Tenex, 2026). A 20 person business should expect 12 to 18 responses. 2. Wednesday: close the survey and read every answer. Highlight any task named by two or more people, and anyone who described a specific, expensive problem in detail. 3. Thursday and Friday: run the 30 minute interviews. Half with owners and executives, half with the staff you highlighted. Run them online with an AI notetaker so nothing is lost. 4. Following Monday: score every task using the formula in the scoring section below, and pick your first three automations. Tenex open-sources this method. "You can steal it if you want," says Lieberman. "You can take all the questions, but we spent a lot of time thinking through what are the questions we should be asking executives and employees in this survey to reveal the biggest problems and bottlenecks." What follows is that method, adapted for Australian trades, allied health and professional services businesses.

These six questions go to every employee, and each is designed to be answered in under 15 minutes with no knowledge of AI required. 1. How long have you been with the business, and what are your main weekly tasks? List them line by line with rough hours for each. Tenure reveals who has the social currency to champion change, and the task list forces people out of autopilot. 2. What percentage of your week is repetitive, follow-the-same-steps work, versus work that needs your judgement? 3. What are the two most annoying, time-consuming rote tasks in your job? This is the question the whole audit hangs on. In Tenex's words, these are the tasks that, if automated, "would make you way happier in your job." 4. If each of those tasks disappeared tomorrow, what would you get back? Estimate it in hours per week or dollars per month. 5. Have you already tried to fix any of these tasks? What did you try, and why didn't it stick? Failed attempts tell you where the traps are. 6. Do you currently use any AI tools at work, which ones, and how often? This is the only question that mentions AI. It exists to set a baseline, not to gather ideas. Notice what is missing: nowhere do you ask "where could we use AI?" That question belongs to you and your scoring sheet, not to your staff.

These six questions are for owners and executives, and they point the audit at the profit and loss rather than individual annoyances. 1. What are the business's three biggest goals for the next 12 months, and what is keeping you from them? This is the "what keeps you up at night about work" question, and it frames everything else. 2. If you had to name one function that most holds the business back right now (quoting, scheduling, invoicing, client intake, compliance, marketing), which is it and why? Tenex calls this the compass question: it tells you where to dig first. 3. What are the biggest cost centres in the business, and which of them are growing faster than revenue? 4. What are the biggest revenue drivers, and what admin work wraps around every dollar earned? In a trades business that might be quoting and invoice chasing. In allied health, intake forms and rescheduling. In professional services, engagement letters and document collection. 5. How is your team split across functions, and where would you hire next if you could? The role you cannot fill or cannot afford is often an automation candidate wearing a job ad. 6. What have you already tried with AI or automation, what did it cost, and what happened? Dead pilots are data. Then apply the pairing rule: start where staff pain and executive priorities overlap. Staff answers reveal processes that technology can fix. Executive answers reveal which of those processes actually moves the profit and loss (Tenex, 2026).

Rank every task the audit surfaces with one formula: hours per week multiplied by the hourly cost of the person doing the work, multiplied by how automatable the task is on a 1 to 5 scale. Step 1: cost the task. Annual cost equals hours per week times hourly cost times 46 working weeks. Cost the person, not the task: an hour of $25-an-hour admin done by someone on a $100-an-hour salary is a $100-an-hour hour. Step 2: score automatability from 1 to 5. A 5 is digital in, digital out, rule-based and high volume, such as copying intake form answers into your practice software. A 1 is physical or judgement work, such as diagnosing a fault or advising a client. Step 3: multiply. Priority score equals annual cost times automatability, divided by 5. Worked example for a 12 person allied health practice, all figures in Australian dollars: - Chasing unpaid invoices: 5 hours per week at $90 per hour fully loaded equals $20,700 per year. Automatability 4. Priority score: 16,560. - Re-keying intake forms: 3 hours per week at $65 per hour equals $8,970 per year. Automatability 5. Priority score: 8,970. - Rescheduling appointments: 4 hours per week at $70 per hour equals $12,880 per year. Automatability 4. Priority score: 10,304. Build the top three, in order. Everything below the line goes on a list you revisit in six months.

A scored task list tells you where to look, not what to build. Before anything gets built, take the top-ranked item and break it into the individual tasks that make it up, then decide task by task which ones a machine does and which stay with a person. The reason is that automatability is not a property of a workflow. It is a property of each task inside it. Chasing unpaid invoices scores well as a whole, but inside it are tasks with very different answers: pulling the list of overdue accounts is pure rule following, drafting the reminder is close to it, deciding whether to chase a client who has just complained about a bill is not. Score the workflow at 4 out of 5 and you will build the judgement step by accident. Expect the breakdown to be contested. The people who run a workflow often run it differently from one another, and asking two of them to describe the same process will usually produce two versions with different steps and different exceptions. That is not an obstacle to the work, it is part of the work. The reconciliation is the first time anyone has had to agree what the process actually is, and skipping it means the build encodes one person's version and quietly breaks for everyone else. Write the tasks in order, in plain words, past tense, describing what happened the last time rather than what should happen. Mark each one as machine, person, or machine drafts and person approves. Breaking a workflow into tasks covers how to do that properly, including what to do with the steps nobody can state a rule for.

The output of an audit is not a score. It is a short document, one or two pages per workflow, naming five things: the workflow, its trigger, the task breakdown with the human steps marked, the named owner, and what a wrong output would cost. Those five items exist for a reason. The trigger tells a builder when the thing runs. The task breakdown tells them what to build and, more importantly, what not to build. The marked human steps are the boundary of the system. The owner is the person who maintains it after launch. And the cost of a wrong output is what decides the shape of the build rather than whether it happens. That last item is the one most audits leave out. Anything whose wrong output is expensive and hard to catch gets built as draft and review, with a person approving before it leaves the building, rather than left running unattended. A reminder sent to the wrong client is recoverable. A quote sent with the wrong price is not, and human approval before an automated send is the difference between the two. The artefact should also be portable. If the only people who can read it are the agency that wrote it, it is a sales document rather than a scope. A real scope can be handed to a different builder, or to your own staff, and produce roughly the same system. Test it by asking someone who was not in the interviews to read it and tell you what gets built. If they cannot, the audit is not finished.

Most AI pilots fail because nobody inside the business owns the result after the consultant, or the initial enthusiasm, leaves. The headline number: 95 percent of AI pilots deliver no measurable return (MIT, 2025). The cause is rarely the technology. Pilots die when they start from a tool rather than a problem, when no one owns them after launch, and when the person who championed them moves on. An audit fixes the first failure by forcing problem-first thinking. It cannot fix the second. So before you build anything from your top three, assign each automation a named internal owner: the person who feels the pain, helped surface it in the survey, and will maintain the system once it exists. In a 10 to 50 person business this is usually an office manager, practice manager or senior administrator, not the owner and not an outside agency. This is also the honest limitation of a DIY audit. Finding the problems is the easy part. Building automations that survive staff turnover, software updates and edge cases is where most small businesses stall. Budget time for the build, or get help with it.

Build the first automation within four weeks of finishing the audit, or the findings go stale and the document joins the pilot graveyard. Your checklist from here: 1. Finish scoring and pick your top three tasks. 2. Name an owner for each. 3. Build the first one. A focused build on a single rote task typically takes two to four weeks. 4. Measure the hours saved in the first month and feed the result back to the team. Visible wins fund the next build. If you want a hand, this is exactly what we do. At 13Labs we run this audit with you in week one of buildAutomation, a 6 to 12 week program where we train your own people to build and own the automations the audit finds. The capability stays in your business when we leave. Read more at 13labs.au/buildAutomation.

How long does an AI readiness audit take to run? About one to two weeks end to end for a business of 5 to 50 people. The survey takes each person 15 minutes, interviews run 30 minutes each, and scoring the answers takes about half a day. The longest part is usually booking the interviews. Do employees need to understand AI to answer the survey? No, and that is deliberate. Only one of the twelve questions mentions AI at all. Tenex found that removing AI from the questions produces better problem data, because people describe their actual work instead of pitching technology they have read about (Tenex, 2026). What if every task sounds important? That is what the scoring sheet is for. Multiplying hours per week by hourly cost by automatability ranks tasks objectively, and the executive compass question breaks ties in favour of whatever most affects the profit and loss. If two tasks score equally, build the one with the clearer owner. Can a sole trader or very small business use this audit? Yes. Answer the six owner questions yourself, and ask any staff or regular contractors the six staff questions. Even three to five responses will usually surface a first automation worth building, and the scoring formula works at any scale. Why not just ask everyone where AI could help? Because people answer with use cases, not problems, and use-case-first projects are how businesses end up in the 95 percent of AI pilots that deliver no measurable return (MIT, 2025). A problem-first audit picks fewer targets and gives each one a measurable payoff before you build.