Most small and midsize businesses are not short on AI. They are short on a plan for it.

A few patterns are common. A team pays for Microsoft 365 or Google Workspace, and the AI features in those plans are switched on but barely used. Someone signs up for a standalone AI writing tool on a personal card. Someone else is pasting customer emails into a free chatbot. A sales rep pitches a new platform that promises to "add AI to your workflow", for another monthly fee per seat.

Each decision makes sense on its own. Together they produce a stack nobody planned: overlapping subscriptions, company data in accounts the business doesn't control, and AI that sits next to the work instead of inside it.

We use a simple order to avoid that. It is the same order we follow on every project.

  1. Built-in AI first. Use the AI already inside the software you pay for.
  2. Enterprise AI second. Where the built-in AI falls short, build automations that call enterprise AI models under business terms, in accounts you own.
  3. New software last. Add a new tool only when nothing you already own can do the job, and know its full cost before you sign.

The rest of this article explains why that order works and how to apply it.

Why the order matters

The order is not about being frugal for its own sake. It follows where the work and the data already live.

Your customer records are in your CRM. Your documents are in SharePoint, Google Drive or WorkDrive. Your team's day happens in Outlook or Gmail, Teams or Google Chat. Every new tool you add has to be connected to those systems, given access to that data, secured, paid for and learned by your staff. Every tool you avoid adding is one less thing to manage.

AI that works inside the systems you already use has three advantages over AI that sits outside them:

  • Your team doesn't change how they work. The help shows up in the inbox, document or record they already have open.
  • Your data stays where it is. It is governed by the contracts, permissions and retention rules you already have, instead of being copied somewhere new.
  • You own the result. When the automation is built in your accounts, it keeps running if you change vendors, change consultants or change your mind.

That last point matters more than it first appears. We think a business should own its automations the way it owns its website or its books. It is the reason behind our No Lock-in Pledge and the reason we build in client-owned accounts.

Step 1: Built-in AI first

Microsoft 365 has Copilot. Google Workspace has Gemini. Zoho has Zia. Most CRMs, helpdesks and accounting packages now ship some AI features of their own. Which ones you have depends on your plan: some come included and some are paid add-ons, so the first job is simply to find out what you are already paying for.

Then ask a practical question for each task you want help with: can the built-in AI do this well enough, today, with a little setup?

Often the answer is yes. Built-in AI is usually good at work that stays inside one application:

  • summarizing a long email thread or meeting
  • drafting a first version of a reply, proposal or policy
  • answering questions from documents the user already has access to
  • pulling a quick summary out of a spreadsheet or report

When that is enough, the right move is to set it up properly and stop there. "Properly" means checking which data it can see, making sure permissions in SharePoint or Drive are right (built-in AI tends to respect existing permissions, which also means it can surface files that were overshared), and showing your team what it's good for. That is a short project, not a platform purchase.

Step 2: Enterprise AI where built-in falls short

Built-in AI tends to struggle in three places:

  • Work that crosses systems. Reading an emailed invoice, checking it against a purchase order in another system and posting it to accounting is not something a single app's assistant is designed to do.
  • Work that should happen without someone asking. Most built-in AI waits for a person to open a panel and type a prompt. A lot of valuable work should start on its own when a document arrives or a record changes.
  • Work that needs your rules. Routing requests the way your business routes them, or checking work against your own standards, needs more structure than a chat box gives you.

This is where automation does the heavy lifting and AI makes it smarter. The automation handles the reliable parts: the trigger, the data moving between systems, the record being updated. The AI handles the judgment steps that used to need a person, such as reading a document, classifying a request or drafting a reply. Our AI page walks through four typical examples.

How it's built matters as much as what it does:

  • Enterprise AI under business terms. We use enterprise AI services whose business terms keep your data out of model training, never consumer chat apps. That is the real fix for shadow AI: give people a safe, sanctioned way to get the help they were looking for.
  • Accounts in your name. The AI service is set up in your account, so usage is billed to you, the data stays under your contract and nothing disappears if you part ways with whoever built it.
  • Not tied to one AI company. Models change quickly. Building so you can switch providers as prices and quality change protects you from betting the workflow on one vendor. We wrote more about that in Beyond the "Seat Tax".
  • A person approves what matters. Anything that goes to a client, moves money or changes important records can require a human sign-off. See Human in the Loop and our AI guardrails guide for how to decide where.

Step 3: New software last

Sometimes nothing you own can do the job. A specialized document-processing need, an industry-specific system or a missing piece of infrastructure may justify a new tool. That's fine. The point of putting it last is that by the time you get here, you know exactly what gap it fills.

Before adding anything new, get clear answers to a few questions:

  • What does it cost in total? Per-seat pricing, usage charges, add-ons and the price after the first year.
  • Where does our data go, and is it used to train anyone's models?
  • Can we get our data and our workflows out if we leave?
  • Does it replace something we already pay for, or does it add to the pile?

Any new software we recommend is listed in the proposal with its cost, so it is a decision you make with the numbers in front of you, not a surprise after the project starts.

A quick audit you can run this week

You don't need outside help to start. Spend an hour on this:

  1. List what you pay for. Every subscription that touches email, documents, customers, finance or support, and the plan level for each.
  2. Note the AI already included. For each one, check which AI features your plan has and whether anyone uses them.
  3. Ask the team what they use. Without blame. Any AI tool people are using on their own tells you where the real demand is, and where your data may already be going.
  4. Pick one workflow that hurts. One repetitive, high-volume task where people spend hours each week.
  5. Run it through the order. Can built-in AI handle it? If not, is it a cross-system or automatic job that an automation with enterprise AI would handle? Only then, is there a gap that needs a new tool?

Most teams find at least one thing they can switch on or tidy up immediately, and one workflow that is worth automating properly.

The bottom line

The question to start with is not "which AI tool should we buy?" It is "what do we already have, and what is it not doing?"

Start with the AI inside the software you already pay for. Build with enterprise AI, in accounts you own, where that falls short. Buy something new only when you have to, and know what it costs. That order keeps your stack small, your data under your control and your automations yours.

If you want a second opinion on one workflow, a free 30-minute call is a good place to start. You'll leave knowing which step of the order it belongs in and roughly what it takes, at a fixed price.