Problems Worth Paying For

The wrong question is "what AI business should I start?"

It starts from a capability and searches for a use, which is how people spend nine months building something nobody was going to buy.

The right question starts from money that is already moving.

Somewhere near you, a company writes a cheque every month for something tedious, repetitive, and important. The cheque is proof of demand: no persuasion required, no new budget to create, no behaviour to change.

Your job is to find that cheque, understand why it is written, and work out whether there is a place beside it where a small owner can stand.

This lesson is a process, not a list of ideas. By the time an idea is on a list, its price has already fallen.

Step One: Map Existing Spend

Start from spending you can observe, not from spending you imagine.

Look at three sources. The first is your own work history: every invoice you have sent or approved, every vendor your employer pays, every line in a budget you have seen.

The second is any business you know from the inside: who is hired, what agencies are retained, which tasks cause overtime, what gets outsourced at quarter end.

The third is public evidence: job postings (a company hiring three people into the same repetitive role is publishing its pain), regulatory deadlines, and procurement notices.

Write twenty lines, each in this form: who pays, for what, how often, roughly how much.

"Mid-size machinery exporters pay a documentation agency roughly $30,000 to $80,000 a year for translated manuals and conformity documents, per market." That is a line. "Businesses need better content" is not.

Twenty lines is an evening's work if you know an industry, and a week of reading if you do not.

Step Two: Find the Compressible Bottleneck

For each line, ask what specifically takes the time and the money.

Inside almost every expensive recurring process, one step eats most of the hours: gathering data, producing a first draft, reconciling two systems that do not talk, chasing people for responses, formatting to someone else's standard, or checking work against rules.

Those are the steps automated systems compress most. Compression alone is not a business, because anyone can compress; the business exists where compression meets something that stays scarce.

Ask three follow-ups.

1. After the bottleneck is compressed, what remains that still requires judgment, context, or a name on the result?
2. Who is currently accountable when the output is wrong, and would they pay to move that risk?
3. Can the customer do the compressed version themselves in an afternoon, and if so, why have they not?

The third question kills most candidate businesses, and it should kill them now rather than in month nine.

Step Three: Score the Opportunity

Now rate each surviving candidate against six criteria, from 0 to 5.

Existing spend

How much real money already flows to this problem each year from one typical buyer? Score 0 for nothing identifiable, 3 for roughly $10,000 to $50,000, 5 for more than $100,000.

Do not score market size. Score one buyer's cheque, because that is what you receive.

Frequency

Is this monthly or weekly (5), quarterly (3), or once every few years (1)? Recurring work becomes retainers, and retainers become something that resembles an asset.

Visible pain

Does the buyer complain about this without prompting, work weekends because of it, or staff it with people they resent paying? Score 5 when the pain is named by the buyer in their own words, 2 when you inferred it, 0 when you assumed it.

Verifiability

Can the result be checked against an objective standard: a regulation, a reconciliation, a rule set, a measurable outcome? Score 5 when correct and incorrect are clearly defined, 1 when quality is a matter of taste.

High verifiability is what lets you promise something, price against risk, and prove you delivered.

Liability

Who bears the consequence when the output is wrong, and how expensive is that consequence? Score 5 when the consequence is large, specific, and currently sitting with the buyer, who would rather it sat with a competent vendor.

Liability is uncomfortable, which is why it is not competed away. It is also why you need advice on contracts and insurance in your own country before signing anything.

Access to buyers

Can you reach ten of these buyers directly within a month, by name, without paying a platform for access? Score 5 for existing relationships, 3 for a reachable and well-defined group, 0 for "I would need ads".

Access is the criterion people fudge, and it is the one that most often decides the outcome.

Reading the total

The maximum is 30.

A total of 24 or more is a strong candidate: go and talk to buyers this week.

Between 18 and 23 is worth testing, but identify the weakest criterion and fix it before building anything.

Below 18, stop. The arithmetic is telling you that either the money is not there or you cannot reach it, and enthusiasm will not change either.

One more rule: any candidate scoring 0 or 1 on access or existing spend is dead regardless of its total, because those two are not compensable by strength elsewhere.

Step Four: Test the Score Against Reality

A score built from your own assumptions is a guess with arithmetic on top.

So talk to eight buyers before building anything, and ask about the past rather than the future: what you did about this last quarter, who did it, how long it took, what it cost, what went wrong. People predict their future behaviour badly and remember past spending reasonably well.

Adjust the scores afterwards. If pain drops from 5 to 2 across eight conversations, you saved six months for the price of eight phone calls.

Then run one small paid test: one buyer, one narrow deliverable, a real invoice. A person who pays $400 tells you more than fifty people who say the idea sounds good.

Step Five: Write the One-Page Thesis

If a candidate survives, write it down in six lines, because anything you cannot write this plainly is not yet understood.

1. Buyer: who exactly, in one sentence, including size and situation.
2. Problem: the recurring task, with its current cost and frequency.
3. Bottleneck: the step that consumes the time, and how it gets compressed.
4. What stays scarce: the judgment, accountability, context, or relationship you hold.
5. Offer and price: what you deliver, on what cadence, for how much.
6. Proof: the evidence you will produce in 90 days that this is real, including the first paid invoice.

Review the page quarterly. Most of the value is in how quickly a written thesis lets you abandon something that is not working.

Where These Problems Usually Live

The categories that score well are unglamorous and stable: compliance documentation, reconciliation and reporting, lead qualification, scheduling and dispatch, collections, onboarding material, claims and warranty processing, permitting paperwork, supplier verification.

None of these are exciting, which is a feature, because exciting categories attract capital and talent and the price falls.

They share an existing budget, a defined standard of correctness, a consequence for being wrong, and a buyer who can be named. That combination is what the rubric detects.

What the Three Readers Do

Tom

Tom's twenty lines are easy to write, because he has invoiced most of them.

His best candidate: machinery exporters paying $30,000 to $80,000 a year for conformity documentation per market, with a compressible bottleneck (translation and formatting) and a stubborn residue (clause-level regulatory checking and a signature).

He scores it: existing spend 5, frequency 4, visible pain 4, verifiability 5 (a regulation defines correct), liability 5, access 4 (eleven former clients). Total 27.

He then calls eight of those clients to ask what they spent last year and what went wrong. Two answers surprise him: the pain is not the translation, it is that nobody internally will sign off on machine-produced documents, so releases sit for weeks.

That reframes his offer from "review service" to "release sign-off within 72 hours", which is the difference between a $1,200 job and a $4,500 one.

Maya

Maya is not leaving her job, so she runs the rubric on problems in her own industry that she could serve on evenings and eventually as a business.

Her candidate: claim substantiation for mid-size consumer brands, where marketing claims must be traceable to evidence before publication.

Scores: existing spend 3 (agencies and legal review, but scattered budget), frequency 5, pain 4, verifiability 4, liability 4, access 5. Total 25, with existing spend as the weak criterion.

So her next step is not to build anything. It is to find out what four brands actually spent on claim review last year.

Leo

Leo's twenty lines come from his logistics employer, and his first candidates are things he does every week.

The strongest: exception handling for small freight forwarders, where a missed exception costs a customer relationship and the work is done by hand every Friday.

Scores: existing spend 2 (absorbed into salaries rather than invoiced), frequency 5, pain 5, verifiability 3, liability 2, access 3. Total 20, with a 2 on existing spend, which is disqualifying on its own.

So Leo looks for the version of the same problem where money already leaves the building: forwarders who pay a contractor to do it. That is a smaller set of buyers, and a real one.

He is 24 with $2,000 saved, so he keeps the salary, runs the test on evenings, and lets the invoices decide.

Worksheet

1. Write twenty lines of observed spend, each naming who pays, for what, how often, and roughly how much.
2. Strike every line where you cannot name a specific buyer or estimate a real number.
3. For each surviving line, name the single step that consumes most of the time and money.
4. For each, write what remains scarce after that step is compressed: judgment, accountability, context, or relationship.
5. Score your top five candidates on the six criteria, 0 to 5, and total them.
6. Discard everything below 18, and everything scoring 0 or 1 on access or existing spend.
7. For your highest scorer, list eight buyers you will contact within 21 days, with their names or roles.
8. Write the five questions you will ask, all of them about the past quarter rather than the future.
9. After the conversations, re-score without flattery and write the one-page thesis for whatever survives.
10. Define the smallest paid test you can run within 60 days, and the invoice amount that would count as proof.

Common Mistakes

Starting from the tool

"What can this system do?" is a hobby. "Who is already paying for this and why does it hurt?" is a business.

The first question produces demonstrations, the second produces invoices.

Confusing a problem you have with one people pay for

Plenty of real irritations have no budget attached, because the sufferer absorbs them. The existing-spend criterion exists to separate irritation from demand.

Choosing unverifiable work

If nobody can say objectively whether the output is right, you cannot promise an outcome or price against risk, and you will be compared on price alone. Prefer problems where a rule, a reconciliation, or a number defines correct.

Avoiding liability

The instinct to say "we take no responsibility for outcomes" is understandable, and it removes the main reason a buyer would pay you rather than run the tools themselves.

Take the responsibility you can understand and cover, in writing, with advice from a professional in your jurisdiction, and price accordingly.

Assuming access

Most plans quietly assume buyers can be found later, at a cost in time and money the plan never budgeted. If you cannot name ten buyers today, your first project is access, not product.

Scoring optimistically

An inflated rubric tells you what you already wanted to hear. Score pain 2 when you inferred it, access 3 when you know the group but not the people, and let conversations move the numbers up.

The RW Finance Perspective

This process is the same one a careful investor uses on a business, turned inward.

When we assess a company, we want to know who its customers are, why they keep paying, and whether anything protects the margin from competitors with the same tools.

This rubric asks a small owner those questions before the business exists, which is cheaper than asking afterwards.

The criteria map onto the Seven Scarcities. Existing spend and access are distribution. Liability is accountable judgment. Verifiability makes trust provable. Frequency turns a transaction into a relationship, and relationships are where proprietary data accumulates.

Businesses built on all six look, in miniature, like the companies whose quality and financial strength we admire in the Screener: recurring revenue, a defined product, and a reason customers cannot easily switch.

Businesses built on none of them grow for two years and then discover they were selling something anyone could sell.

The discipline is identical whether you are buying a share of a business or building one: understand the business before you argue about the price.

Lesson 19, Pricing When Your Competitors Also Have AI, takes the surviving candidate and answers the question that decides whether it is worth doing at all, which is what to charge when production costs almost nothing for everyone.

Key Takeaways

• Start from money that is already moving rather than from a capability searching for a use.
• Write twenty observed lines of spend naming who pays, for what, how often, and how much, and discard anything vague.
• Inside every expensive recurring process there is one step that consumes most of the hours, and that is what automation compresses.
• Compression alone is not a business, because anyone can compress; the business is where compression meets something that stays scarce.
• Score candidates 0 to 5 on existing spend, frequency, pain, verifiability, liability, and access, for a maximum of 30.
• Act on 24 and above, test and repair between 18 and 23, and stop below 18 regardless of how appealing the idea feels.
• A 0 or 1 on access or existing spend is disqualifying on its own, because neither can be compensated for by strength elsewhere.
• Test scores with eight conversations about the past quarter, then with one small paid invoice, before building anything.
• The categories that score well are unglamorous and stable, which is precisely why their prices hold.
• Write a one-page thesis with buyer, problem, bottleneck, what stays scarce, offer and price, and the proof you will have in 90 days.