What problems will people keep paying to solve?

Why is it getting harder to know what people will still pay for?

The safest problems to build around today are becoming clearer: people keep paying when ignoring the problem costs them money, creates risk, blocks something essential or leaves them personally responsible for a bad outcome.

AI has made the distinction much sharper. Writing a competent email, summarizing a contract, producing a basic image, generating code or explaining a tax concept can already cost almost nothing. Yet spending around accounting, payroll, cybersecurity, healthcare administration, payments and physical repair remains enormous.

Recent numbers make the contrast difficult to dismiss. Intuit's latest full-year results showed Global Business Solutions revenue rising 16% to $12.9 billion, while QuickBooks Online Accounting grew 23%. Gartner currently expects worldwide information-security spending to reach roughly $244 billion this year. And the American Medical Association's latest physician survey still found practices spending about 13 hours every week handling prior authorizations.

Technology has improved dramatically in all three areas. The underlying problems survived because somebody still has to keep the books correct, prevent the breach and get the treatment approved.

That is a tougher test than asking whether a problem is common. We need to know what happens when the customer leaves it unsolved.

Does making a task cheap eventually destroy the business around it?

Cheap AI will crush the price of many tasks, but businesses built around consequential outcomes can keep charging even when much of the work underneath becomes automated.

Tax preparation gives us a particularly useful example. Intuit reported that total U.S. TurboTax units fell 2% in its latest fiscal year. At the same time, TurboTax revenue rose 7% to $5.3 billion, while TurboTax Live revenue jumped 37% and reached 53% of total TurboTax revenue.

Customers had access to better software and increasingly capable AI, yet more revenue moved toward assisted tax products.

Why? Because producing an answer and being confident enough to file that answer with the tax authorities are different problems. A taxpayer may happily use AI to explain a deduction and still pay when the return becomes complicated enough that a mistake could cost thousands of dollars.

The same pattern applies elsewhere. AI can draft a legal clause without making contractual liability disappear. It can classify an expense without closing the books. It can describe a suspicious login without containing an intrusion.

So the pressure from AI depends heavily on where a product sits in the chain. Software that produces an intermediate artifact faces brutal price pressure. Software that completes the transaction, verifies the result or carries responsibility for the outcome has much more room to charge.

What the customer buys	Durability as AI improves
A generated output	Low
Faster execution	Medium
A completed workflow	Higher
A verified result	High
Prevention of a costly failure	Very high

Will businesses always pay for more customers and revenue?

Businesses will keep paying heavily to make more money, although they are becoming much less interested in paying for marketing activity that cannot be tied to sales.

That difference is already visible. AI can produce huge volumes of advertising copy, landing pages, sales emails and social posts. None of that has made customer acquisition easy. Companies still fight over distribution, qualified leads, conversion rates and retention because those determine whether revenue appears at the end.

The durable opportunity sits close to that outcome. A tool that produces 50 cold-email variants competes with increasingly cheap models. A product that identifies the right prospects, contacts them, books meetings, updates the CRM and can show how much pipeline it created has a much stronger reason to exist.

Payments make the same point from another angle. Toast finished last year with roughly 164,000 restaurant locations using its platform and about $195 billion in trailing-12-month gross payment volume. Restaurants use Toast for far more than generating information: orders turn into payments and operational records inside the same system.

That is the kind of proximity to money we should look for.

Lead qualification, appointment filling, conversion improvement, repeat purchasing, cart recovery, sales follow-up and revenue collection should all remain valuable problems. Generic marketing production looks considerably less protected because the supply of decent content is exploding.

Will companies keep paying software to cut costs?

Companies will keep paying to remove large recurring costs whenever the savings can be measured clearly enough to make the purchase almost self-justifying.

Technology spending alone shows how large that opportunity can become. Gartner currently expects worldwide IT expenditure to exceed $6 trillion this year, with software and IT services accounting for trillions of dollars between them. The FinOps Foundation's recent research covers organizations managing more than $80 billion in annual cloud expenditure, while FinOps teams are increasingly taking responsibility for SaaS, licensing, private cloud and AI costs too.

Once a spending category reaches that scale, small percentage savings become large dollar amounts.

Imagine a company spending $20 million annually across cloud infrastructure, SaaS subscriptions and AI models. Finding and permanently removing 5% of genuine waste creates $1 million in annual savings. A product capturing $100,000 of that value can still look cheap.

The same logic works well outside software. Logistics companies pay to reduce empty miles. Retailers pay to lower inventory waste. Factories pay to reduce downtime and energy use. Restaurants pay to schedule labor more efficiently. Finance teams pay to catch duplicate or erroneous payments.

The strongest products increasingly close the loop themselves. A dashboard saying "you could save $600,000" is useful. A system that safely removes the waste and then proves the saving on the next bill is much harder to cut from the budget.

Will people keep paying to stop fraud and cybercrime?

Fraud and cybersecurity look more durable today than they did before the AI boom because attackers are improving too, and the amount of money being lost is still rising.

The FBI's latest Internet Crime Report recorded more than one million complaints and nearly $21 billion in reported losses from cyber-enabled crime, up 26% from the previous year. Cyber-enabled fraud accounted for more than $17.7 billion of those losses. Americans over 60 alone reported around $7.7 billion.

The Federal Trade Commission found the same direction from a different dataset. Consumers reported about $16 billion in fraud losses last year, roughly 25% more than the year before. Imposter scams accounted for $3.5 billion and have nearly tripled in reported losses since 2020.

Businesses are spending accordingly. Gartner currently expects information-security expenditure to reach about $244 billion, an 11.6% increase in constant currency.

AI complicates this market in an unusual way. Better models help defenders detect attacks, investigate events and automate response. They also make phishing, impersonation, fake identities, social engineering and malicious automation cheaper.

That arms race keeps regenerating the problem. A company cannot decide that cybersecurity has become cheap enough to ignore, because the adversary has an economic incentive to return tomorrow.

Will taxes, payroll and compliance ever become too automated to sell?

Taxes, payroll and compliance should remain excellent paid problems even as most of the manual work disappears, because the obligation keeps coming back and mistakes still have consequences.

Payroll is probably the cleanest commercial example. ADP generated roughly $21.9 billion in its latest fiscal year. Its Employer Services client revenue retention was 92.1%, and based on those retention levels ADP estimates an average Employer Services relationship of about 13 years.

Thirteen years is an extraordinary relationship for business software. Payroll earns that persistence because companies have to run it again next week or next month. Employees expect the right amount. Governments expect taxes and filings. Benefits, deductions and employment changes all have to be reflected correctly.

Tax behaves similarly. As seen above, Intuit's latest results showed customers moving further toward higher-value assisted tax services even while total TurboTax filing units declined.

Regulation also keeps creating new work. Privacy rules, employment law, financial reporting, healthcare requirements, customs procedures and industry-specific licensing continually change what businesses must record and prove.

Automation will shrink the hours required to complete these processes. The payment survives because the customer still needs a correct result by a deadline.

Recurring obligation	What the customer ultimately needs
Payroll	Everyone paid correctly and on time
Tax	Correct filing and defensible treatment
Financial reporting	Accurate records that survive scrutiny
Privacy compliance	Required controls and evidence
Licensing	Permission to keep operating

Will healthcare remain worth paying to fix when AI gets much better?

Healthcare will remain one of the largest pools of paid problems because better information does not remove illness, treatment decisions, insurance friction or the need to deliver care.

U.S. health expenditure already exceeds $5 trillion annually. That scale matters because even tiny inefficiencies support enormous businesses.

Prior authorization shows how stubborn those inefficiencies can be. In the American Medical Association's latest survey of 1,000 practicing physicians, doctors reported handling an average of 40 prior authorizations per week. Physicians and staff spent around 13 hours a week on them, and 40% of physicians employed staff dedicated exclusively to the process.

There has barely been a breakthrough despite years of digitization. The same AMA research found that 74% of doctors said denials had increased over the previous five years. Nearly one-third said requests were often or always denied.

Those figures describe a problem with a clear buyer, measurable labor cost and direct effect on whether a patient receives treatment. AI can help automate the paperwork, but somebody still has to move the request through the insurer's rules and resolve an adverse decision.

Other healthcare problems have the same shape: claims, scheduling, medication adherence, patient monitoring, billing, clinical documentation and care coordination.

Generic medical information is becoming abundant. Getting a real patient through a real healthcare system remains difficult.

Is elder care one of the safest long-term problems to build around?

Elder care is unusually durable because demand is rising quickly and much of the actual work still requires a human being to show up.

The Bureau of Labor Statistics has just updated its long-term projections. Employment of home health and personal care aides is expected to grow 18% between 2025 and 2035, compared with about 3% for all occupations. That would add roughly 847,000 jobs. Once worker replacement is included, BLS expects about 760,500 openings every year.

Very few large labor markets are growing at that speed.

Demographics explain much of it, but there is another force underneath: more long-term care is moving from institutions into homes and community settings. That spreads care across thousands of fragmented locations and makes coordination harder.

AI can help families schedule visits, monitor changes, manage medications, communicate with caregivers and handle paperwork. Robotics may gradually automate some physical assistance. Neither development removes the need for dependable care in the foreseeable future.

There are therefore several durable layers around the same problem: finding caregivers, verifying credentials, filling shifts, coordinating families, managing payments, arranging transportation, adapting homes and monitoring vulnerable people.

Aging itself continually generates the demand. That gives elder care a structural advantage most software categories do not have.

Will people always pay to repair homes and other physical things?

Physical repair should remain highly durable because buildings, vehicles and equipment keep breaking regardless of how capable software becomes.

Harvard's Joint Center for Housing Studies currently expects annual U.S. spending on homeowner improvements and repairs to remain around $519 billion through mid-2027. Growth is slowing sharply, yet the spending level remains above half a trillion dollars.

That is a good example of why market growth can mislead us. Home repair does not need to grow 20% annually to support strong businesses. Roofs age. Pipes leak. Air conditioners fail. Electrical systems need upgrades. None of those failures waits for favorable venture-capital conditions.

AI should change how the work is sold and managed. It can diagnose symptoms, quote jobs, answer customers, optimize routes and help technicians find the right procedure. The final repair still happens in a physical location.

Software built around those operators can therefore inherit some of that durability. Lead management, quoting, dispatch, field-service scheduling, payments, financing, parts ordering and follow-up all connect directly to work that somebody eventually has to perform.

Physical entropy is a very dependable source of repeat business.

Will businesses keep paying to find workers when AI can do more jobs?

Businesses will keep paying to find reliable people in occupations where work still has to happen in the physical world, although generic recruiting software is becoming much easier to replace.

The labor market has softened from its post-pandemic extremes, so we should be careful with sweeping "worker shortage" claims. The more durable shortage is concentrated in specific jobs and skills.

The latest Bureau of Labor Statistics projections still show major growth in home-care workers, registered nurses and several hands-on occupations. Small-business surveys also continue to show employers struggling to fill some skilled positions even as overall hiring demand cools.

AI creates a strange split here. Generating a job description, screening keywords or writing a candidate email is becoming trivial. At the same time, employers still need to know whether an electrician is licensed, a nurse is credentialed, a caregiver is trustworthy, a technician is available and a restaurant worker will actually turn up for the Friday-night shift.

That makes verification, matching, credentialing, scheduling and attendance more attractive than resume generation.

The commercial opportunity gets stronger when the software stays involved after the hire. Payroll, time tracking, scheduling and workforce management continue every week, creating much better retention than a one-off recruiting transaction.

Will businesses always pay to get paid?

Businesses will keep paying to turn completed work into cash because accounts receivable, billing and collections sit directly between reported revenue and money in the bank.

Intuit's latest annual filing offers a useful clue. Growth in its money products was driven partly by payments, with increases coming from more payments customers, greater payment volume per customer and higher revenue yield. Toast's roughly $195 billion of annualized restaurant payment volume shows the same economic pull from another industry.

Payments become particularly sticky because the software participates in the transaction itself.

There are many variations of this problem. Contractors chase invoices. Healthcare providers chase claims. SaaS companies recover failed subscription payments. Marketplaces distribute money between buyers and sellers. Finance teams reconcile incoming payments against invoices. Lenders collect repayments.

AI can automate almost every administrative step surrounding collection. That should make these products better rather than remove the customer's reason to buy them.

Cash still has to arrive.

Are boring back-office workflows actually safer than flashy AI apps?

Boring back-office software is currently one of the safer places to build when it holds important business records or runs a process customers cannot casually abandon.

ADP's estimated 13-year Employer Services relationship gives us the strongest example. QuickBooks Online Accounting growing 23% in Intuit's latest fiscal year gives us another. These businesses live inside recurring processes where historical data and tomorrow's transaction depend on the same system.

Switching is therefore more painful than cancelling a standalone productivity app. A company moving payroll systems has to migrate employees, tax settings, benefits, deductions, historical records and integrations. Replacing an accounting system means migrating transactions, reconciliations, invoices and reporting structures.

The same advantage can appear in inventory, field service, property management, healthcare administration, logistics and vertical software.

Integrations strengthen that position when they carry mission-critical data. Simple "connect app A to app B" features will become cheap. Reconciliation, permissions, identity matching, audit trails and monitoring remain harder because bad data can trigger bad actions across several systems.

AI agents may actually increase this need. Humans often notice that a number looks wrong before acting on it. Automated workflows can move incorrect data at machine speed.

For durable software, owning the reliable operational record looks increasingly valuable.

Will people still pay just to save time?

People will keep paying to save time when the time is expensive, repeated frequently or directly limits how much work someone can do.

The prior-authorization example makes this concrete. Thirteen hours of physician and staff time every week is not a minor convenience problem. Removing even part of that workload can free meaningful clinical capacity.

The same arithmetic works for a field-service company wasting technician hours on routing, an accounting team reconciling transactions before every close or a freight operator manually planning hundreds of movements.

A standalone email-writing tool faces a much weaker case. Saving four minutes on an occasional email certainly has value, but users increasingly get that ability bundled into software they already pay for.

Frequency and hourly value decide the economics.

A product saving a surgeon ten hours a month has room to charge far more than a product saving a student ten hours a year. A system freeing one delivery vehicle for an extra job every day can show its value in revenue. A meeting-summary app often has a harder time proving what the saved minutes were worth.

So "save time" remains a valid business proposition, but we should immediately ask whose time, how much of it, how often and what the customer can do with the capacity they recover.

Is trust becoming more valuable because AI can fake more things?

Trust is becoming a much bigger paid problem because generating something convincing is getting cheaper while checking whether it is real still carries consequences.

Fraud data are already moving fast. The FTC recorded roughly $16 billion in reported consumer fraud losses last year, the highest level in its dataset and around 25% above the previous year. Imposter scams alone caused $3.5 billion of reported losses. The agency says those losses have nearly tripled since 2020.

The FBI's cybercrime figures point in the same direction, with reported losses approaching $21 billion.

Generative AI lowers the cost of producing credible phishing messages, fake documents, synthetic voices, manipulated images and large volumes of personalized outreach. That gives verification software more situations in which it can earn money.

The strongest opportunities appear right before consequential decisions. Banks need to know whether a transfer request is genuine. Employers need to verify workers. Insurers need to detect fabricated claims. Marketplaces need to know who their sellers are. Companies need to decide whether an invoice actually came from a supplier.

People rarely pay much for abstract "trust." They do pay to avoid wiring $250,000 to an impersonator.

That difference makes identity, authentication, provenance, fraud detection and transaction verification unusually interesting these days.

Which popular problems are becoming bad businesses?

Generic content creation, basic information retrieval and shallow productivity are becoming much tougher standalone businesses because users still want the output but increasingly expect it to be cheap or included elsewhere.

Writing social posts is useful. Summarizing documents is useful. Generating basic images is useful. Producing first-draft presentations is useful. Basic coding assistance is enormously useful.

The problem is abundance.

When dozens of models can deliver an acceptable result and large platforms can bundle the same feature into software customers already use, differentiation collapses quickly. A good product can still win through distribution, brand, proprietary data or a much better experience, but the underlying problem no longer provides much protection.

This is why the move from generation toward execution matters so much. A support tool that drafts an answer faces intense competition. A system that identifies the issue, checks account data, takes the permitted action, records what happened and closes the customer case owns far more of the outcome.

The same progression appears across many categories.

Increasingly commoditized	More durable problem
Write sales emails	Produce qualified meetings
Summarize invoices	Reconcile the books correctly
Explain tax rules	Complete a defensible filing
Draft support replies	Resolve customer cases
Generate security alerts	Contain real attacks
Suggest staff schedules	Keep every required shift covered
Explain an insurance denial	Get the claim processed

What problems will people keep paying to solve?

People will keep paying most reliably to make money, get paid, avoid large losses, satisfy unavoidable obligations, receive care, find essential workers and keep physical things functioning.

After looking across the evidence, those categories separate themselves quite clearly from generic digital productivity.

Revenue generation survives because companies always want more profitable customers. Payments and collections survive because booked revenue is useless until cash arrives. Cybersecurity and fraud prevention survive because attackers keep returning and reported losses are rising. Taxes, payroll and compliance survive because deadlines and legal responsibility remain. Healthcare and elder care survive because real people still need treatment and assistance. Physical repair survives because buildings and equipment continue to deteriorate.

Back-office operational software belongs close behind when it becomes embedded deeply enough that customers depend on it every week. ADP retaining Employer Services clients for an estimated 13 years is far more informative about durability than the latest fashionable software category growing 200% from a tiny base.

The weakest problems are increasingly those where the customer wants an artifact and faces little consequence if the output is merely adequate. Generic writing, images, summaries, simple research and lightweight productivity all sit under growing pressure because AI keeps expanding the supply.

A simple question cuts through most of the ambiguity: what happens if the customer does nothing?

Losing money, missing payroll, violating a rule, leaving a medical claim unresolved, suffering fraud, failing to cover a critical shift or living with a broken air conditioner usually forces action. People reach for their wallets because the alternative hurts.

If doing nothing merely means spending a few extra minutes or accepting a slightly worse draft, the pricing power is much weaker today.

That is where the durable opportunity has moved. The best businesses increasingly take responsibility for something the customer needs to happen, while the easiest tasks around that outcome become cheaper and cheaper.