The AI Jobs to be Done Analyst skill for Claude

The AI Jobs to be Done Analyst is a free skill from our own library. A skill is an instruction file, SKILL.md, that gives an AI assistant a fixed method for one task. Not software you install, no subscription, no data connection: one text file of 1,633 words that tells Claude what a jobs to be done analysis should look like, which questions it has to ask first, in which order the analysis is built up and which words it may not use.

You download the zip at the top of this page, put it in your Claude environment, and from that moment on Claude thinks about customers in jobs instead of in audiences. Nothing else on your machine changes. What a SKILL.md is exactly and why a text file steers the behaviour of an AI is explained in what Claude skills are.

The core principle is in the first lines of the file and has deliberately been kept simple. People do not buy products or services. They hire them to get a specific job done. The job is the goal, the product is a means. A customer who buys a drill is not hiring a drill bit, they are hiring a hole in the wall. And that hole is itself a means to something else, for instance hanging up a family photo that creates a sense of connection. The task of the skill is to look through the surface at the real job and at the progress the customer is trying to make in their life.

It is written for business owners, marketers, product people and anyone who has to position something. The frontmatter names the openings on which Claude should pick the skill up: JTBD analysis, customer needs research, product positioning, work on a value proposition, summarising customer interviews, and the synthesis of a switch interview. Loose sentences count as triggers too, such as what is my customer hiring me for, what is the real job, or pull push anxiety habit. You will find it in the skill library that we make available for free from the AI and automation service, with no account and no sales email afterwards.

What you should know before you start: this skill is strict. It accepts no assumptions as input. Say that your customer wants to work faster and you get no analysis but a question back: why do you think that, on which observations does it rest, and have you ever done a switch interview to validate it. That feels awkward when you want something quickly, and it is exactly the reason the outcome is usable. Invent the job yourself and you have a persona with a different label on it.

The word persona does not appear in this skill, and that is not a style choice but a rule with a reason attached: personas are demographic, jobs are situational. Two people with identical demographics can have completely different jobs. And the other way round: two people who resemble each other in nothing can have exactly the same job, at exactly the same moment.

That difference is practical. A persona tells you who someone is: age, job title, company size, a stock photo and an invented first name. There is little you can do with that, because you cannot change someone's age and it tells you nothing about what they are trying to get done today. A job tells you why someone is moving right now. But that moment you can reach: with your message, with your onboarding, with what you do and do not promise.

So the question it asks is a different one. Not what the customer wants, but what they are trying to achieve and which progress they want to make in their life or work. It sits in the rules as well, as an explicit ban on the first form of that question. Asking what someone wants produces a wish list of features they happened to see somewhere. Asking what someone is trying to achieve produces a situation, and that situation holds all the information you need.

Features come last. There is another rule that belongs with this: the skill treats features as irrelevant until it knows the job, with the note that a feature without a job is guessing. That turns the usual order round. In most product discussions it starts at what we can build and ends at who is waiting for it. Here it starts at which progress someone wants to make and ends at what is needed for that.

And the skill writes from the customer. The last rule is the least conspicuous and perhaps the most useful: everything is formulated from the first person of the customer, not from the company. Not we want more conversion, but when I am on this page, I want to understand whether this is for me. In diff form that reversal looks like this.

A skill is only as good as its instructions, so we simply describe them here. The file opens with frontmatter: name, version 1.0.0, MIT licence, our own name as the author, and a description with the openings on which Claude picks the skill up. Then comes the role description, and it is sharply worded: you work strictly according to the framework of Christensen, Moesta and Ulwick, and your aim is to lay bare which job the customer wants solved, not which product they want to buy.

Three layers of analysis. Every input gets three layers laid over it. The first are the three job dimensions, which are always all three named, even if the customer names only one. The functional job is what practically has to be done, the measurable task. The emotional job is how the customer wants to feel during and after the job. The social job is how they want to come across to others. The second layer are the four forces of Bob Moesta, and the third layer is the job story structure from FIG.02 above.

A five-step workflow. For every question the skill follows the same path. Step 1 is job excavation: asking on until the job under the surface is visible, with at least five questions. What triggered the first time you thought about this, the first thought. What you tried before that, which solutions were rejected. When you started searching seriously, the active looking moment. What it had to be able to do at minimum before you would consider it, the bare minimum. And what would be a dealbreaker, the anxiety triggers.

Then it moves on. Step 2 is the job map. Step 3 are the outcome statements. Step 4 is the hire and fire analysis: which solution does the customer hire today and which have they fired, with the note that a fired solution gives more information than a hired one, because that is where the pain was. Step 5 is the synthesis.

Eight fixed output sections. Every analysis has the same build-up. The primary job in job story format. Then the job dimensions, in three lines. After that the map of eight steps, with the pain per step. The four forces as a two by two matrix, with push and pull on the pro side, anxiety and habit on the con side, and at least two concrete observations per force. Five to eight outcome statements. The competition, including indirect solutions. The untapped opportunities. And finally three test hypotheses, each measurable and falsifiable.

Outcome statements with a fixed grammar. Desired outcomes are written as minimise or maximise, plus a measurable metric, plus context. The file gives examples: minimise the time needed to send a quote after an intro call, and maximise the likelihood that a customer feels understood at the first point of contact. That form is directly usable in a backlog, and that is exactly the intention: it forces you to write down what has to get measurably better instead of which function you want to build.

Five rules that bind the skill. Never the word persona. Never ask what the customer wants. Features are irrelevant until the job is settled. Accept no assumptions as input. And always write from the first person of the customer. Those five are worked out in FIG.05 further on, with the response it gives per rule. The file closes with four sources and one sentence of accountability: the skill is based on those sources and applies the frameworks in a structured analysis workflow. Want to build such a file yourself? The whole structure, from frontmatter to rules, is described in writing a SKILL.md.

Jobs to be Done is not the method of one author but a school with a few clear voices. The SKILL.md names four of them, with the year, and adds in one sentence that the skill applies those frameworks in a structured analysis workflow. Which is useful, because you can read them and decide for yourself whether you agree.

Clayton Christensen, Competing Against Luck, 2016. This is where the heart of it comes from: people hire a product to make progress in a particular circumstance. The consequence is bigger than it looks. If the job is your starting point, who your competitor is changes, and that explains a large part of the surprises in markets. That gets made explicit in the competition analysis, with an example you do not forget quickly: a Netflix subscription competes with a glass of wine as evening relaxation.

Bob Moesta, Demand Side Sales 101, 2020. From him come the four forces in FIG.03 and the emphasis on the switch interview: the conversation in which you unpick what exactly happened in the days and weeks before someone switched. Moesta's contribution is that selling starts at the demand side, at the moment someone begins to move, and not at the offer. That connects directly to what the Mom Test Coach skill does with your question list: getting facts out of the past instead of opinions about the future.

Tony Ulwick, Jobs to be Done Theory to Practice, 2016. From him come the eight universal job steps in FIG.04 and the outcome-driven approach: writing desired outcomes down as measurable statements instead of as wishes. Ulwick makes the framework operational. Where Christensen shows why you have to look differently, Ulwick shows what you then write down.

Alan Klement, When Coffee and Kale Compete, 2016. Klement sharpens the idea of progress and works out the job story form this skill uses. His title sums the point up: coffee and kale resemble each other in nothing and still compete, as soon as you look at them as two ways of making the same progress.

You can do something with this without installing the skill as well. Write down one job story for your most important customer, in the form from FIG.02, and test it against the four forces. If you do not know what the push was, you do not yet know why they came to you. And if the honest answer is that every provider in your market solves that job in the same way, the Blue Ocean Strategy Coach is the skill that looks for the open water beside it. And that is usually the most interesting part of the conversation.

An analysis that ends in insight is a nice document. This skill ends in three hypotheses that can be knocked down, and that is the difference. The last three sections of the output are all aimed at what you do next: the competition, the untapped opportunities and the test hypotheses.

The competition is drawn more widely than you are used to. It asks which other solutions this customer hires for the same job, and stresses that direct competitors are only half of the picture. In the example those are spreadsheets, the accountant, other software, and doing nothing and hoping. That last one is the most important and at the same time the most ignored. Anyone who does not acknowledge their biggest competitor aims their message at the wrong comparison.

Untapped opportunities sit where high pain and poor solutions meet. It hunts for job steps with a lot of pain for which the market offers no good solution at all, because that is where the innovation sits. In the example that is Confirm: no solution gives real peace of mind before filing. That is a concrete opening, far more usable than the observation that the market is crowded.

Three hypotheses, each measurable and falsifiable. The example gives them including the way you test them. The first: sole traders with more than fifty invoices per quarter have daily VAT stress within two weeks of the deadline, to be validated with a diary study. The second: an explicit confirm function that says everything has been checked and looks right raises the willingness to switch, to be validated with a landing page test. The third: sole traders trust soft signals from other sole traders more than product features, to be validated by analysing reviews for emotional language.

Note what those three have in common. They can all be proved wrong. You can run a diary study and discover that the stress is not daily at all. That is exactly the intention: a hypothesis that cannot be knocked down is an opinion. So the skill delivers no conclusions but a research agenda, and that agenda is the most usable part of the whole document. And when you want to hold those conversations well afterwards, the questioning technique is a craft of its own, and that is what the Mom Test Coach is about.

The zip is 4.4 KB and contains one folder, ai-jobs-to-be-done-analyst, with the SKILL.md inside it. Installing is a matter of putting the file in the right place, and that place differs per environment. SKILL.md has been an open standard since December 2025, so the same skill also works in Codex, Cursor and Gemini CLI. You are therefore not downloading a Claude file but a working instruction that any modern AI assistant can read.

After that, use is simple: describe what you sell and to whom, and paste in as much real material as you can. Interview transcripts, support tickets, reviews, a write-up of a switch interview. The more real words from real customers, the less the analysis has to guess. If you work with assumptions only, you get those assumptions back in the form of questions, and that is not a bug but the rule from the file.

Everything here sits under the MIT licence, so you may open SKILL.md and adjust the triggers, the workflow or the examples to your own market. Should you get stuck on the installation itself, the full route is set out step by step in installing Claude skills. Want to learn to set this kind of work up with your team? Our AI training was built for exactly that.

Where it is strongest: when you already have material and need to bring structure into it. A pile of interview transcripts, a series of support conversations, a collection of reviews: that is exactly the kind of input on which an analysis workflow with eight fixed sections does its work. It is also strong on a switch interview, because the four forces are literally hidden in that conversation. And it helps with positioning questions, provided you are willing to accept that the answer is not about your product.

There are also situations in which you are better off not using it. The first is when you only have assumptions and want them confirmed. Then it mostly delivers questions back, and that is frustrating when you want something finished quickly. The second is when you are looking for a quantitative grounding. This framework is qualitative in nature: it explains why people do what they do, it does not measure how many people do it. This one solves that by ending in hypotheses instead of conclusions, but it does mean the real work still comes afterwards.

And then the most honest limit. No skill can talk to your customers for you. Everything it delivers is a structured working of what you supply, supplemented with the questions you have not yet asked. That helps you, because most people skip exactly those questions. But anyone who reads the analysis as evidence instead of as a hypothesis is further from the truth than someone who has called two customers. The outcome is a research agenda and not a research result.

What logically follows depends on where you end up. Are you stuck on which features are expectations and which really set you apart? Then the Kano Model Analyst skill is your next step. Do you want to lay the job beside your offer instead? Then it is the Value Proposition Canvas skill you need. And if the outcome touches your whole model, the conversation belongs with the Business Model Canvas Coach or the Lean Canvas Strategist.

Knowing which job your customer has you do makes your message sharper, but there does have to be somebody to read it. If too little traffic comes in, the real work sits before that: being found by the people who are searching at exactly that moment. That moment, the active looking moment from step 1 of the skill, is incidentally exactly what search engines turn on. The free SEO scan shows in a few seconds where your site stands. And if you want to talk further about what AI can do for your business, from separate skills to full automation, we simply do that in a conversation.