The Two Types of Problems AI Actually Solves (And the Ones It Still Can't)

Why 74% of Companies See Nothing from AI

Here's a number worth sitting with: 74% of companies globally report they've seen no tangible value from AI, according to Deloitte's 2026 survey across 3,000 business leaders.

That's not a story about bad technology. The technology works. It's a story about the wrong problems getting the wrong tools. Most businesses are trying to use AI on problems it wasn't built for — while ignoring the problems it absolutely dominates.

Before you spend another dollar on AI tools, you need a map of which problems AI actually solves. There are two types that matter for any business your size. Then there's a third category where AI still can't help you, no matter what the vendors promise.

Effort Problems: Big, Not Hard

The first type is what you'd call an effort problem. These aren't intellectually difficult. They're just relentlessly large.

Think about what consumes the most hours in a typical service business. Appointment reminders that go out every day. Follow-up emails after every estimate or inquiry. Invoices that need to be generated, sent, and tracked. Data that has to move from one system into another. Every one of these tasks is straightforward. Any competent person can do any individual piece. The problem is sheer volume — sustained attention across thousands of touchpoints, week after week, without dropping anything.

AI was built for this. Not because it's smarter than your team, but because it doesn't get tired, doesn't forget, and doesn't need to prioritize which follow-up to skip when the day gets busy.

The clearest example of scale: Klarna's AI agent handled the equivalent work of 853 full-time employees. That's not a rounding error. That's what happens when you point an agent at the right problem type — high volume, repetitive, pattern-based — and let it run.

For a plumber, this looks like automated appointment confirmations and same-day job follow-ups. For a mortgage broker, it's status update emails at every stage of a file. For an HVAC company, it's seasonal maintenance reminders across your entire customer list. The work is the same every time. The only thing stopping it from happening consistently is human bandwidth.

What an Effort Problem Looks Like in Practice

A landscaping company with 300 active clients has to send spring startup reminders every year. In the past, someone on the team would spend three days manually emailing every client, tracking who responded, chasing the ones who didn't, and entering confirmations into the calendar. The work itself takes seconds per client. The total cost is three days of salary and the near-certainty that some clients get missed.

An agent handles the same job in minutes. It sends personalized reminders, logs responses in the CRM, flags non-responders for a second touch, and marks confirmed appointments automatically. The three days of effort work disappear. The outcome — clients scheduled, revenue secured — improves because nothing falls through.

That's the structure of every effort problem: the individual task is simple, the volume makes it expensive, and automation captures the full benefit without any sacrifice in quality.

Coordination Problems: Right Information, Right Place, Right Time

The second type is the coordination problem. This is where AI is doing something people underestimate.

Coordination problems aren't about thinking hard. They're about information flow. Customer submits a form → your CRM needs to be updated → someone needs to get notified → a follow-up needs to be scheduled → another system needs a record created. In most businesses, pieces of this chain get dropped every week. Not because anyone is incompetent, but because it requires a human to notice and act at each step.

AI agents can watch an entire chain like this and handle each step automatically, 24 hours a day, without needing someone to trigger it.

An event planner who automates her inquiry intake knows exactly what this looks like. A form submission used to sit in email until she got to it. Now an agent logs the inquiry, checks availability, sends the prospect a personalized response with pricing, and adds a task to her to-do list — before she's even seen the original message. She went from responding to leads in six hours to responding in under three minutes. The agent didn't think for her. It just never dropped the handoff.

This matters because coordination problems are expensive in ways that are hard to measure. A lead who doesn't hear back in an hour goes somewhere else. A client who wasn't notified about a schedule change becomes a problem call. A task that fell through the cracks becomes a refund conversation. Agents don't solve hard problems. They solve the problem of things consistently not getting done.

The Real Cost of Coordination Failures

Most businesses underestimate what dropped handoffs actually cost. Research by Harvard Business Review found that companies lose up to 20% of revenue annually from poor information flow across teams. For a $2 million service business, that's $400,000 a year in lost opportunities, refunds, and client churn — caused not by bad service delivery but by the gap between one step and the next.

When an agent owns the coordination layer, handoffs become automatic. A job gets marked complete → the agent triggers an invoice → logs a follow-up task → and sends the client a completion summary with a review request. Four steps that used to require four separate human actions happen in sequence without anyone touching a keyboard. This is what moving from Level 1 to Level 3 AI adoption actually looks like in practice.

Where AI Still Can't Help You

There's a third category, and this is where businesses get into trouble when they expect too much.

Emotional intelligence is still yours. Knowing when a long-time client who's been loyal for eight years needs a real conversation, not a template email — that requires reading someone. No agent can tell when frustration is about the invoice versus about something else entirely. No agent can sense that a customer who sounds fine is actually about to churn.

Judgment calls are still yours. Whether to bend policy for a specific situation, whether to fire a client who's draining your team, whether to take a job that's priced too low but opens a door you want — these depend on context that exists only in your head, built from years of running your business.

Ambiguity is still yours. When a customer can't articulate what they actually need, figuring out the real problem underneath the stated one is a deeply human skill. "I want better communication" usually means something specific and often something different for each person saying it. An agent can't decode that.

These aren't temporary limitations waiting for a better model. They're the areas where the problems aren't primarily cognitive — they're relational, contextual, and often require a kind of courage that isn't computable.

The Klarna case makes this concrete. Their AI agent resolved customer tickets in 2 minutes instead of 11 — a spectacular effort-problem win. But when the tickets required judgment, empathy, or the kind of nuance that comes from knowing a company's culture and values, the agent fell short. Klarna had to rehire human agents because the problem they pointed AI at was actually in the third category, not the first. You can read the full breakdown in Klarna Fired 700 People and Had to Hire Them Back.

How to Find Your Effort and Coordination Problems

The $85,000 per month that enterprises now spend on AI — up 36% year-over-year — isn't going toward solving ambiguity problems or making judgment calls. It's going toward exactly what we've described: eliminating the human hours spent on high-volume, repetitive work and keeping information flowing correctly across complex operations.

You don't need an enterprise budget to get the same results. You need to look at your business through the right lens.

Walk through a typical week and ask two questions. First: what tasks does my team do every day or every week that follow the same pattern every time? These are your effort problems. Second: what information fails to get to the right place at the right time — and what does it cost you when that happens? These are your coordination problems.

To make this concrete: write down every recurring task your team does in a week. For each one, ask whether a competent person who had never met your business before could do it with a clear checklist. If yes, it's a candidate for automation. If the answer is "it depends on knowing this client" or "it depends on reading the situation," that's judgment work — keep it human.

The businesses seeing real results from AI aren't the ones who bought a tool and hoped for the best. They're the ones who identified a specific effort or coordination problem, pointed an agent at it, and measured what changed.

Frequently Asked Questions

What's the easiest effort problem to automate first? Follow-up sequences are the most common starting point because the pattern is clear and the payoff is immediate. If your business sends any kind of follow-up after an estimate, a completed job, or an inquiry — and that follow-up sometimes doesn't happen because the day got busy — you have an effort problem ready for automation. The agent sends the message at the right time every time, regardless of what else is happening.

Can a small business with no technical staff actually build this? Yes. The barrier isn't technical — it's clarity about what you want the agent to do. The tools that connect to your CRM, email, and booking systems are designed to be configured without code. What takes time is specifying the logic: when does the follow-up go out, what does it say, and what happens if the client doesn't respond. That's a business problem, not a technology problem.

How do I know if a problem is an effort problem or a judgment problem? Ask whether a new hire with a good checklist could do it reliably in their first week. If yes, it's an effort problem. If the answer is "they'd need months of context to get it right," it's judgment work. The line isn't always clean, but in practice most problems sort themselves quickly when you ask that question honestly.

What happens when I automate an effort problem and it goes wrong? Every agent needs a clear escalation path. Before you launch anything, define the situations where the agent should stop and alert a human instead of continuing. An automated follow-up that gets sent to a client who just had a complaint resolved is a mistake. That's why escalation conditions matter — you build the exceptions in advance so the agent handles them correctly, not on the fly.

Does automating effort work actually free up my team, or does it just create other work? The honest answer is: it depends on whether you actively redirect the time. The team hours freed up from effort automation don't automatically fill with better work. You have to decide what you want your people doing instead — more client-facing work, more complex problem-solving, more business development. The automation creates the capacity. You have to deploy it.

Start With One Problem

The goal isn't to automate everything at once — it's to find the one effort or coordination problem that's costing you the most, build something that works, measure it, and go from there. The businesses that get real value from AI don't start with a technology roadmap. They start with a problem list and pick the most expensive item at the top.

Associates AI helps businesses run exactly this kind of audit — identifying the specific effort and coordination problems worth automating first, then building workflows that measure results. If you're thinking through AI for your business, book a call.