The short answer
Pick one workflow that's costing you real money every week, in lost sales, missed charges, slow cash or hours, measure it, fix it with AI, and check the result after 30 days. Then do the next one. Companies that get real value from AI almost never start with a big transformation. They start with the quote that takes three days to send, the charges that never make it onto the invoice, or the invoices AP still keys by hand.
The rest of this guide explains how to find those tasks in your own business, which ones are worth automating, what they cost, and how to tell whether the money came back.
Start with the work, not the technology
The most common reason AI projects disappoint is that they start with a tool ("we should use ChatGPT for something") instead of a problem. A better starting question is: where do our teams spend time on work that follows the same pattern every time?
Walk through a normal week and look for tasks that are:
- Repetitive: the same steps, done many times a week.
- Text- or document-heavy: reading emails, PDFs, forms, notes or invoices and moving information somewhere else.
- Rule-based enough to explain: if you could teach a new hire to do it in an afternoon, AI can usually help.
- Annoying: the work your best people complain about is often the best place to start, because they'll actually use the fix.
The hard part is seeing them. Leadership teams consistently underestimate how much time their AP clerks, dispatchers, coordinators and office staff spend copying information between systems, chasing answers and rebuilding the same documents, because that work rarely shows up in a report.
Where AI actually pays off
Hours saved are the easiest benefit to see, but rarely the biggest. Almost every useful AI project pays off in one of four ways, and knowing which one you're after makes it much easier to decide what to build and how to measure it.
Sales you currently miss: slow quotes that lose bids, leads nobody follows up, empty appointment slots, customers who quietly stop ordering.
extra jobs won × average value × marginMoney already earned but never collected: parts and hours left off invoices, pricing errors, rework, missed rebates and penalties.
charges caught × average valueFaster invoicing and collections, so money owed to you sits in your account instead of on a receivables report.
days faster × daily billingsThe same team handles more volume, so growth doesn't mean another hire in every department.
hires avoided × loaded salaryA contractor or distributor who quotes the same day instead of three days later doesn't just save hours. They win work that used to go to whoever answered first, and at a larger company that one change can be worth more than every hour saved combined.
The most common quick wins, by department
These are the projects that show up again and again in established companies, from a single office to many locations. Most can be built in weeks, not months, and most work with the email, documents and software you already use.
| Department | Quick win | What changes |
|---|---|---|
| Front office | Inbox triage | Incoming email is sorted by type and urgency, routine replies are drafted, and requests are logged automatically. |
| Front office | Scheduling and reminders | Appointments are confirmed, reminded and rescheduled without phone tag. |
| Sales | Quote drafting | Notes from a call or site visit become a draft quote in your template within minutes. |
| Sales | Lead follow-up | Every new lead gets a fast, personal first response and timely follow-ups, so none go cold. |
| Operations | Status updates | Customers are told automatically when an order ships, a crew is on the way or a job is done. |
| Operations | Document intake | Forms, work orders and delivery tickets are read and their details entered where they belong. |
| Finance | Invoice intake | Vendor invoices are read, coded and routed for approval in your accounting system or ERP, with amounts and due dates flagged. |
| Finance | Collections reminders | Polite, escalating reminders go out on overdue invoices without anyone having to remember. |
| Finance | Missed-charge checks | Every job's parts and hours are compared with the invoice before it goes out. |
| Customer service | Answer common questions | A trained assistant answers routine questions from your own policies, and hands off anything unusual to a person. |
| Leadership | Weekly operating summary | Key numbers from every department and location, stuck items and what needs attention arrive in one summary every Monday. |
| Leadership | Contract and deadline tracking | Agreements are read as they come in, and renewals and deadlines surface before they're missed. |
Bigger projects come later: prediction (who's likely to cancel, which deals will stall, what demand looks like next month), connecting an ERP, CRM and field systems that don't talk to each other, and custom software built around how your business runs. They're worth doing, but they're rarely the right first step.
Every business has a dozen candidates on this list. The value is in knowing what to build first, and how to build it so it pays off. Book your analysis →
How to tell a good AI project from a money pit
Good first projects have a few things in common: the task happens often, someone can quickly check the result, and a mistake is an inconvenience rather than a lawsuit. They also have an owner, one person who uses it daily and speaks up when it's wrong.
Warning signs: the project needs data you don't have, nobody can describe the current process, success is described as "being more innovative," or the plan is to replace a whole department in one step.
What AI costs for a business
Costs vary widely, but they fall into a few recognizable bands. These are typical ranges we see for established companies, not quotes.
| Type | Typical cost | Good for |
|---|---|---|
| Off-the-shelf AI tools | Tens of dollars per user per month | Writing help, meeting notes, general chat assistants |
| Custom quick win | A few thousand dollars for a single-team task; more when it spans departments, locations or an ERP. Plus a monthly fee to run | One specific task, built around your process and tools |
| Bigger custom project | Tens of thousands and up | Prediction, connecting ERP, CRM and other systems, custom platforms |
| Ongoing support | Monthly | Keeping systems running, improving them, finding the next win |
The right question isn't "what does AI cost?" but "what is this specific problem costing me now?" A company billing $20 million a year that lets 1% of billable parts and hours slip off its invoices is losing $200,000 a year. A distributor that cuts quote turnaround from three days to the same day can win more bids in a month than a year of saved hours is worth. A 20-person company sees the same thing on a smaller number, and a build in the low thousands still pays for itself within months.
Not sure what your own problems are costing you? We price every project against that number, and you only pay if it hits the impact we agree on. Book your analysis →
How to measure the ROI of AI
Measuring AI is simpler than it sounds, as long as you measure before you change anything. Pick the number that matters for that project, whether it's win rate, missed charges, days to collect or hours, agree on the target before anything is built, and check the same number again 30 days after launch.
Monthly value = revenue gained + leakage stopped + cost of cash freed + hiring avoided
Not everything needs a spreadsheet. Sometimes the measure is simply "leadership stopped building the Monday report by hand." What matters is agreeing on it up front, so nobody argues about it later.
Buy a tool, build custom, or bring in help?
Buy a tool when…
the task is common, like meeting notes or writing help, and an existing product already fits how you work.
Build custom when…
the task is specific to your business, it crosses several systems (email to your ERP or accounting system to your job or field-service software), or you've tried tools and none of them fit. Custom doesn't have to mean expensive: many quick wins are small, focused builds.
Bring in outside help when…
you're not sure where AI would pay off, nobody in-house has time to build and maintain it, or the project touches sensitive data. Look for people who will measure results, build inside your systems, and tell you honestly when something isn't worth doing.
Five mistakes that sink AI projects
- Starting too big. "Automate operations" fails. "Read vendor invoices into the accounting system" succeeds.
- No baseline. Without a before, there's no way to prove the after, and the project loses support.
- Leaving out the people who do the work. If they weren't consulted, they won't use it.
- No human check at the start. Let people review the AI's output until it has earned trust.
- Ignoring upkeep. Formats change, vendors update software, and someone has to keep things running.
Data, privacy and security basics
- Use business-grade tools that commit to not training on your data, and turn off training wherever it's an option.
- Keep sensitive data in the right systems. Health, financial and personal information should stay in tools and setups that meet the rules for your industry, with agreements such as HIPAA business associate agreements in place where required.
- Control access. AI systems should see only what they need, and access should follow the same permissions your team already has.
- Avoid free consumer chat tools for customer data. They're great for drafting and brainstorming, not for client records.
- Keep a person accountable for anything that goes to a customer, especially early on.
Getting your team to actually use it
The best AI system fails if nobody uses it. Adoption comes down to a few habits:
- Build inside the tools people already use (their inbox, the ERP or accounting system, the CRM, the job tracker) instead of adding another app.
- Start with the task people hate, so the change feels like relief, not surveillance.
- Be clear about what it's for. In most companies, AI takes over the repetitive parts of jobs, not the jobs themselves, and the team should hear that early.
A 30/90/180-day plan to start
One or two quick wins, live, with a person checking the output.
Keep what worked, drop what didn't, add the next wins.
The bigger project: prediction, connected systems or a custom platform.
The order matters more than the timeline. The right sequence depends on where your hours and money actually go, and it changes as the business changes.
Common questions
What is applied AI?
Applied AI means putting AI to work on a specific business task, such as reading invoices, drafting quotes or answering routine customer questions, rather than researching or building AI models. The goal is a measurable result: hours saved, revenue gained or mistakes avoided.
What is the best way to start using AI in a small business?
Start with one workflow that costs real money every week, in lost sales, missed charges, slow collections or hours, and has a clear before-and-after you can measure, such as quoting, invoicing or entering vendor invoices. Measure it today, fix that one workflow with AI, and check the result after 30 days before moving to the next.
How much does it cost to implement AI in a business?
Off-the-shelf AI tools typically cost tens of dollars per user per month. A focused custom automation usually costs a few thousand dollars to build plus a small monthly fee. Larger custom systems typically start in the tens of thousands. The right number depends on how much time or money the project saves.
How do you measure the ROI of AI?
Measure the task before you change anything, then again after 30 days. Most projects pay off through revenue gained, leakage stopped, cash collected sooner or capacity to grow without hiring. Divide the cost by the monthly value to get the payback period.
What tasks can AI automate in a business?
The most common are reading and entering invoices and documents, drafting quotes, sorting and answering routine email, scheduling and reminders, customer status updates, follow-ups on leads and unpaid invoices, weekly summaries, and flagging problems like likely no-shows or stalled deals.
Is using AI different for larger companies?
The approach is the same; the numbers are bigger. When a workflow repeats across several departments, locations or an ERP, a single quick win can be worth six figures a year. It also takes more coordination: more systems to connect, more teams to bring along and tighter access controls. Smaller companies benefit the same way, on a smaller number.
Will AI replace my employees?
In most small and mid-sized businesses, AI removes the repetitive parts of jobs rather than whole jobs. The usual result is that the same team handles more work, responds faster and spends more time on customers and judgment calls.
Do I need clean data or new software before using AI?
Usually not. Most quick wins work with the email, documents and software you already use. Bigger projects like forecasting need a few months of reasonably consistent records, and a good roadmap will tell you honestly whether your data is ready.
Is it safe to put business data into AI tools?
It can be, with business-grade tools that don't train on your data, proper access controls, and sensitive information kept in systems that meet your industry's rules. Avoid pasting customer data into free consumer chat tools.
Should I buy an AI tool, build something custom or hire a consultant?
Buy when a tool already fits how you work. Build custom when the task is specific to your business or crosses several systems. Bring in outside help when you're not sure where AI would pay off, or nobody in-house can build and maintain it.
Want this mapped for your business?
Foundry NX is an applied AI firm in Scottsdale, Arizona. A guide can tell you what's possible. Our analysis tells you what's worth doing in your business, what it will cost and what it should return. Then we build the quick wins first, and you only pay if they hit the impact we agree on up front. We work with established companies across departments, locations and systems, and the approach scales down too: if one workflow is costing you real money every week, it's worth a look.
Book your analysis →