AI business solutions: find the one use case that pays first

Most AI business solutions so far have cost more time and money than they returned, usually because they started broad, with no baseline and no owner. Start with one narrow, repetitive job you already measure, where the data is in decent shape and a person checks the output. Prove the return there, then decide on the next one.

Why do so many AI projects fail to pay off?

A lot of AI spending so far has been a waste of money and time. That's not a fringe view. The research points the same way from several directions.

One widely cited study, MIT NANDA's "The GenAI Divide: State of AI in Business 2025," put it bluntly: "Despite $30–40 billion in enterprise investment into GenAI… 95% of organizations are getting zero return." Treat that number with care. The report isn't peer-reviewed, the sample is small (52 interviews, 153 leader surveys and a review of 300+ public implementations), and critics have questioned its method. It defined success strictly, as deployment beyond pilot with measurable KPIs six months on. Still, the direction matches other research:

• McKinsey's State of AI 2025 found only 39% of respondents report any enterprise-level EBIT impact from AI, and only about 6% qualify as "high performers" with more than 5% of EBIT coming from AI.
• BCG's "The Widening AI Value Gap" (September 2025) found 60% of companies report hardly any material value from AI; only 5% are getting value at scale.
• Gartner predicted in 2024 that 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, and in 2025 that over 40% of agentic AI projects will be cancelled by the end of 2027 because of rising costs, unclear value or weak risk controls.

The common thread isn't that the technology doesn't work. It's that many AI business solutions were aimed at the wrong problem, or at too many problems at once.

Where is the ROI in AI business solutions?

Often not where the budget goes. MIT NANDA found about half of generative AI budgets going to sales and marketing, while back-office automation often delivered better ROI. The glamorous use case gets the money; the dull one pays.

BCG's data shows why getting it right matters: the leaders see five times the revenue gains and three times the cost reductions of the laggards. And McKinsey found that redesigning workflows, rather than bolting AI onto the existing ones, had the strongest link to EBIT impact. Buying a tool is easy. Changing how the work gets done is where the return comes from.

In practice, the first use cases that pay tend to be the repetitive work your team does between the real work: intake, routing, follow-up, reconciliation. Nobody writes a press release about them, but they're measurable, frequent and easy to check.

How do you pick your first AI use case?

Pick one. The AI business solutions that pay first are rarely the broad ones. Not a roadmap of twelve, not an "AI strategy" deck. One job, with a number attached. A good first candidate passes most of these tests:

• It happens often. Hundreds or thousands of times a month, not a few times a quarter. Volume is what turns small savings into real ones.
• You already measure it. You know how long it takes, what it costs or how often it goes wrong today. Without a baseline, you can't show a return.
• The inputs are consistent. Forms, emails, documents or records that look roughly the same each time.
• The data is usable. Gartner predicts that through 2026, organizations will abandon 60% of AI projects that lack AI-ready data. If the information lives in five systems that don't talk to each other, fix that first or pick another job.
• A person checks the output. Especially at the start. A reviewer catches errors and builds the trust you'll need for the next project.
• Someone owns it. One named manager whose team does the work today and who will say whether it's better.

What that looks like in insurance and professional services

Insurers are already well into this. The NAIC's May 2025 survey of 93 health insurers found 84% using AI or machine learning, most commonly for utilization management (71%) and prior authorization (68%). Those are high-volume, rules-heavy, document-heavy jobs: the profile of a good first use case. About 92% of those insurers also had AI governance principles in place, which is worth copying.

In professional services, the same logic points to intake and paperwork: sorting incoming requests, drafting first-pass documents from templates, reconciling time and billing records. Each is frequent, measurable and checkable by someone who already does the work.

Here is what a good first use case looks like in numbers. A growing Metro Detroit recruiting team spent about four hours researching candidates for each search. Decypher built them an AI research agent that searches and cross-checks 25 public sources and returns a list of candidates, each with the source and the reason it was flagged. The same work now takes about 15 minutes. It is one repetitive job, measured before and after, with a person checking every result: exactly the profile above.

And sometimes the first step isn't AI at all. In an insurance marketing analytics project, Decypher Corp standardized raw marketing data and attached an ROI figure to new-business marketing spend. That's plain data work, not AI, but it's the kind of groundwork that makes a later AI project measurable. You can see it on our results page.

Should you build AI in-house or use artificial intelligence services?

For most mid-market companies, getting outside help for the first project is the safer bet. MIT NANDA found external partnerships reached deployment about 67% of the time, against about 33% for internal builds (self-reported). Your team knows the work; a builder who has shipped AI automation before knows where these projects break.

The same study found that only 40% of companies had bought an official LLM subscription, while workers at over 90% of them use LLMs anyway. Your people are already experimenting. That's useful: ask them which tasks they're using it for. Your first use case may already be sitting in someone's browser tab, just without the governance, security or measurement.

How to start: a simple plan

1. List the repetitive work. Ask each department head for the three most repetitive tasks their team does. Include volume and time per task.
2. Score each against the tests above. Volume, baseline, consistent inputs, usable data, human review, an owner.
3. Choose one. The one with the clearest number, not the most exciting demo.
4. See it working before you commit. Ask for a working prototype on your own data. Decypher, for example, covers the first three steps (an initial call, a deep-dive discovery and a working prototype) at its own cost, and aims to solve the core problem in two to four weeks, so you see the result on your real work early.
5. Measure against the baseline. If it doesn't beat the old way, stop. That's a cheap lesson, not a failure.
6. Then pick the next one. The second project is easier, because the data, the governance and the trust are already partly in place.

If you'd like help finding that first use case, our custom software and AI automation page explains how we scope it. We help you decide what's worth automating and what isn't; if a build makes sense, the work goes to Decypher, which pays Salter Growth when an engagement goes ahead. You pay Salter Growth nothing.

Score your own candidates with the free AI use case scorecard, built on the six tests above.

Sources

1. Decypher Corp, "From hours to minutes: AI-powered sales talent sourcing" case study (decyphercorp.com, accessed September 2026)
2. MIT NANDA, "The GenAI Divide: State of AI in Business 2025" (July 2025)
3. Futuriom, "Why We Don't Believe MIT NANDA's Weird AI Study" (August 2025)
4. McKinsey & Company, "The State of AI" (November 2025)
5. BCG, "The Widening AI Value Gap" (September 2025)
6. Gartner, "Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025" (July 2024)
7. Gartner, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027" (June 2025)
8. Gartner, "Lack of AI-Ready Data Puts AI Projects at Risk" (February 2025)
9. NAIC, "NAIC Survey Reveals Majority of Health Insurers Embrace AI" (May 2025)