Software people love jargon the way lawyers love Latin — it sounds impressive and keeps you from asking follow-up questions. That's a problem when you're the one signing the check. Here are the twenty-three terms you'll actually hear when buying software — or AI — for a small business, defined in plain English, with no quiz at the end.
Custom software
Custom software is an application built specifically for one business, matching its exact workflow instead of forcing the business to adapt to a generic tool. Some developers hand over the code and leave hosting and upkeep to you; others, including us, keep it hosted, patched, and fixed and license it to you monthly. Either way, your data should stay yours — get the export terms in writing. It used to be a big-company luxury; AI-accelerated development has pushed the cost into small-business territory.
Off-the-shelf software
Off-the-shelf software is a ready-made product built for thousands of businesses at once, typically sold as a monthly subscription. It's cheap to start and fast to adopt, which makes it the right answer surprisingly often. The catch: it fits everyone approximately and nobody exactly, so businesses end up bending their process around the tool — or bolting three more tools on to fill the gaps.
Web app
A web app is software that runs in a browser and does something interactive — storing data, taking bookings, running calculations — rather than just displaying pages. A website is a brochure; a web app is a tool. Because it lives in the browser, there's nothing to install and it works on any device. Not sure which one your business needs? That distinction drives the price more than anything else.
AI-accelerated development
AI-accelerated development is building software with experienced developers using AI tools to work dramatically faster, cutting the cost and timeline of custom software without cutting quality. The AI writes and refactors code at speed; the human decides what to build and takes responsibility for it, and nothing ships without a human approving it. The result: software that used to take months to build can now ship in weeks.
Automation
Automation is having software perform a repetitive task — sending reminders, moving data, generating invoices — automatically instead of a person doing it by hand. Good candidates are tasks done the same way every time, more than a few times a week. The payoff isn't just saved hours; it's that software doesn't forget, mistype, or call in sick. Our automation guide covers where to start.
Integration (API)
An integration connects two software tools through an API so they share data automatically, eliminating manual copy-paste between systems. An API (application programming interface) is simply the doorway a tool exposes so other software can talk to it. When someone says "does it integrate with QuickBooks?", they're asking whether the two systems can pass data through that doorway without a human in the middle.
CRM
A CRM — customer relationship management system — is a database of your customers and every interaction with them: calls, quotes, jobs, follow-ups. Its real job is making sure nothing lives only in the owner's head or a sales rep's notebook. Plenty of businesses run one in a spreadsheet until the spreadsheet starts fighting back; that's usually the moment to upgrade.
Dashboard
A dashboard is a single screen that pulls your key business numbers — sales, jobs in progress, cash position, whatever you steer by — together in real time. Instead of assembling a report from five tools every Monday, you glance at one page and know how the business is doing. Here's what a good one shows and what it costs.
SaaS (and per-seat pricing)
SaaS — software as a service — is software you subscribe to and use over the internet, running on the vendor's servers instead of yours. Many SaaS products use per-seat pricing: a monthly fee for each person who logs in, so the bill grows with your team. Per-seat pricing is common and reasonable — we charge $49 a month per staff login ourselves — so the useful comparison is what each seat gets you: how closely the tool fits your workflow, whether hosting, support, and fixes are included, and how you get your data out if you leave. Our cost guide walks through the math.
Data migration
Data migration is moving your existing records — customers, jobs, history — from an old system or spreadsheet into a new one, cleaned and intact. It's the unglamorous step that decides whether a new system starts useful or starts empty. Done well, it also fixes years of duplicates and typos on the way in. Always ask who's doing it and what it costs. It's often priced separately (ours is, in the written quote); it's only a problem when nobody mentions it until the invoice.
MVP (minimum viable product)
An MVP — minimum viable product — is the smallest version of a piece of software that solves the core problem, shipped first so you get value quickly. Instead of guessing every feature up front and paying for all of it, you launch the essential 20%, use it, and let real experience decide what gets built next. "Minimum" describes the scope, not the quality.
Hosting
Hosting is the service that keeps your website or app running on a server connected to the internet, twenty-four hours a day. The raw server for a small-business app typically runs $5–50 per month. What costs more is everything around it — monitoring, backups, security updates, and someone on the hook when it breaks — which is why a managed plan (ours included) costs more than the server itself. Whoever builds your software, ask where it's hosted, whose name the account is in, and what the monthly fee covers. The answers should be short.
Scope creep
Scope creep is when a software project slowly grows beyond what was originally agreed — one "quick addition" at a time — inflating the cost and timeline. It's a common reason projects blow their budgets, and it thrives on vague agreements. The defenses are a written scope, written approval before each change starts, and a parking lot for good ideas that can wait for version two.
Technical debt
Technical debt is the future cost of quick-and-dirty shortcuts in software: code that works today but gets harder, slower, and more expensive to change over time. Like financial debt, a little is a reasonable trade for speed — an MVP takes some on deliberately. Too much, and every small change starts costing large money, which is how a "simple update" turns into a rebuild.
Large language model (LLM)
A large language model (LLM) is an AI system trained on enormous amounts of text so it can read, summarize, draft, and answer questions in plain language. The models behind ChatGPT, Claude, and Gemini are LLMs. They're very good with words and patterns — reading an invoice, drafting a reply, sorting a messy note — and not dependable for arithmetic or for facts they can't see in front of them. That's why the software we build lets the model read and draft, and lets ordinary code do the math. Our practical guide to AI for small business covers where that line falls.
Prompt
A prompt is the instructions and information you give an AI model — the question you type into ChatGPT, or the instructions a piece of software sends along with a document. Clear prompts get better results: say what you want, show an example, and say what to do when the answer isn't there. In custom software, the prompt is usually written and tested once by the builder, so staff press a button instead of learning to phrase requests. With online AI tools, anything in a prompt is sent to the provider, so read whether it's safe to put customer data into ChatGPT or Claude before you paste in someone's details.
Tokens
Tokens are the chunks of text an AI model reads and writes — often a short word or part of a longer one — and they're how AI usage is priced. Anthropic's rough estimate is about three-quarters of an English word per token, and providers charge separately for tokens in (what you send) and tokens out (what the model writes). As of September 25, 2026, Anthropic's published API prices per million tokens were $1 in / $5 out for Claude Haiku 4.5, $2 / $10 for Claude Sonnet 5.5, and $4 / $20 for Claude Opus 5.5. So reading a one-page document — say 2,000 tokens in, counting instructions, and 500 tokens out — on Sonnet 5.5 costs (2,000 × $2 + 500 × $10) ÷ 1,000,000 = $0.009, under a penny. The same job is (2,000 × $1 + 500 × $5) ÷ 1,000,000 = $0.0045 on Haiku 4.5 and (2,000 × $4 + 500 × $20) ÷ 1,000,000 = $0.018 on Opus 5.5. More in what AI actually costs a small business.
Hallucination
A hallucination is when an AI model states something false with the same confidence it uses for things that are true — a made-up figure, a misread number, a policy that doesn't exist. It isn't lying; the model produces the most plausible-looking answer, and plausible isn't the same as correct. On WillIFit, Dan's own site that checks parking-garage clearance heights for RVs and trucks, an early $5 test using only the cheapest model got 4 of 11 sign readings wrong — including reading a speed-limit sign as a clearance height. The fix: the cheap model is now only trusted to spot whether a sign is there, a stronger model reads it, and a check rejects any number that doesn't match the model's own transcription. That's the pattern behind how we build AI into small business software.
Human-in-the-loop
Human-in-the-loop means a person reviews and approves an AI's work before it counts — before the expense is entered, the email is sent, or the change goes live. The AI does the reading and drafting; a person makes the call. The rule we build by: the model transcribes the numbers, ordinary code does the arithmetic, a person approves anything involving money, and customer-facing AI answers only from text you've approved. Where AI does act on its own — like a short automatic reply to a crew member who answered a booking text off-script — it's logged, and anything unclear goes to a person. In our own client portal, AI sorts incoming bug reports and an AI agent drafts the fix, but nothing deploys until a person approves it; for a nonprofit's board, we built a tool to draft a 150–250-word monthly financial summary for the accountant to edit and approve. More in AI reads the paperwork, you approve it.
AI agent
An AI agent is an AI model set up to take actions on its own — search, click, fill in forms, send messages, update records — working step by step toward a goal instead of answering a single question. A chatbot answers; an agent does. That makes agents useful for narrow jobs with a clear finish line and easy-to-check results, and risky anywhere a wrong step costs money or reaches a customer. Our rule is that an agent can gather, sort, and draft, and a person approves before anything is paid or sent out as paperwork; the few steps that run on their own are small and logged. Here's what AI can do for a small business today, and what it shouldn't.
OCR (optical character recognition)
OCR (optical character recognition) is technology that turns a picture of text, like a scanned page or a photo of a receipt, into text a computer can search, copy, and use. Classic OCR reads the characters but doesn't know which number is the total; AI models can go a step further and pull out the vendor, date, and amount for a person to check. OCR still earns its keep on its own: the payroll tool in our nonprofit payroll case study sends scanned payroll registers through OCR that runs on the tool's own server — no outside service sees the register — with no AI involved, and a scanned import must pass the same tie-out as any other file before the tool continues.
AI Overviews
AI Overviews are the AI-generated answers Google shows at the top of some search results, summarizing a topic with links to supporting pages. Google pitches them as a fast way to get the gist of a question, with links to dig further. For a local business, that summary may be the first thing a customer reads about you, which makes accurate, consistent facts on your own site matter more. As of October 2026, Google's Search Central documentation says there are no extra requirements or special optimizations to appear in AI Overviews and no special structured data to add: a page needs to be indexed and eligible to show in Search with a snippet, and normal SEO best practices still apply. More in how customers find local businesses in ChatGPT and Google's AI answers.
Generative engine optimization (GEO)
Generative engine optimization (GEO) is the work of making sure AI tools like ChatGPT and Google's AI Overviews find, understand, and accurately describe your business. Honestly, it's mostly the same good practice as SEO: readable pages that answer real questions, the same accurate facts — hours, services, service area — everywhere your business is listed, and structured data that spells those facts out for machines. Google says its AI features need no special markup, so structured data is ordinary good practice, not a secret switch. Google sells labeled ads inside some AI Overviews, but nobody can buy the unpaid answer or recommendation itself, and no one can guarantee one — be wary of anyone selling that. More in how customers find local businesses in ChatGPT and Google's AI answers.