Copilot, ChatGPT Enterprise, or Gemini: What an Arizona SMB Should Actually Buy
A buying framework for small businesses choosing between the major AI assistants, based on where your data lives and what your team actually does.
The short answer
For a small business the choice rarely comes down to model quality, because all the major assistants handle ordinary knowledge work competently. It comes down to four factors: where your data already lives, whether your permissions are clean, what work your team actually does, and the contract terms for the specific tier you are buying.
The comparison articles you will find online mostly rank these tools on benchmark scores. That is close to irrelevant for a 40-person company. All of the major assistants are competent at the work a small business needs. The decision comes down to four things that have nothing to do with model quality.
One: where your data already lives
This is the single biggest factor and most buyers underweight it.
If your company runs on Microsoft 365 with SharePoint, Teams, and Outlook, an assistant that reads your tenant natively has an enormous practical advantage. It can answer questions about documents your team already has without anybody uploading anything. That advantage disappears if your files are scattered across personal Google Drives, a legacy file server, and a Dropbox account somebody set up in 2019.
Same logic in reverse for Workspace shops.
The corollary is uncomfortable but important. If your data is a mess, native integration is not a feature, it is a liability. The assistant will confidently surface the wrong version of a document because it has no way to know which of the six copies is current. Cleaning up file structure and permissions before deploying a tenant-integrated assistant is not optional preparation. It is most of the project.
Two: what your permissions look like right now
A tenant-integrated AI assistant respects the permissions you have configured. That sounds reassuring until you check what you actually configured.
In most small business tenants we assess, there is at least one SharePoint site or shared drive that is readable by everyone in the company and contains something that should not be. Payroll spreadsheets in a finance folder inherited from a migration. An HR site with a broken permission inheritance. Before AI, that exposure was theoretical because nobody browses folders they do not need. After AI, an employee can ask a plain-language question and get a straight answer out of it.
Run a permissions audit before you deploy. If you do nothing else on this list, do that.
Three: the actual work your team does
Group the use cases by work type and the answer usually becomes clear.
Document and email work inside your existing productivity suite favors the assistant built into that suite. Drafting, summarizing long threads, pulling a status update out of a Teams channel.
Open-ended reasoning, analysis, research, coding, and anything where you want to iterate on a hard problem tends to favor a standalone assistant with a strong model and a good interface for long conversations. Many technical teams end up with a standalone tool regardless of what the rest of the company uses.
Customer-facing automation, where AI touches a workflow rather than a person, is a different purchase entirely. That is an API and integration project, not a per-seat license, and it should be scoped and budgeted separately.
A lot of companies land on two tools, not one. That is fine. It is cheaper than a bad single choice.
Four: contract terms your attorney would care about
Read the data handling terms of the specific tier you are buying, not the marketing page. The three provisions worth checking:
Whether your inputs and outputs are used to train models. Enterprise tiers generally commit that they are not. Free and consumer tiers frequently reserve the right.
Data residency and retention. How long is your content held, where, and can you shorten it.
Whether the vendor will sign a business associate agreement if you are a covered entity, or support the contractual terms your own client agreements require you to flow down. If you are a healthcare practice or a subcontractor with government work, this question comes first, not fourth.
What we tell clients
For a Microsoft-based company under 100 people doing normal knowledge work, start with the Microsoft assistant for a pilot group, on a clean tenant with permissions audited, and add a standalone tool for your technical and analytical staff who will hit the limits of an in-suite assistant quickly.
For a Google-based company, the same logic with the Google assistant.
For anyone whose data is not in either place in an organized way, spend the first ninety days and the first dollars on data organization. An AI assistant pointed at chaos produces confident chaos.
Get the scorecard
The scorecard is a weighted worksheet covering the four factors above plus cost modeling per user, integration effort, and the contract checklist. Score the two or three tools you are considering, and the sheet gives you a ranked result and a written list of the prerequisites you have not met yet.
That last list is usually the most valuable output.
Frequently asked questions
Is Microsoft Copilot or ChatGPT better for a small business?
It depends on where your data lives. If your company runs on Microsoft 365 with SharePoint and Teams, an assistant that reads your tenant natively has a large practical advantage. For open-ended analysis, research, and coding, technical staff often prefer a standalone assistant regardless of what the rest of the company uses. Many companies end up with both.
What should we check before buying an AI assistant?
Run a permissions audit. A tenant-integrated assistant respects the permissions you configured, and most small business tenants have at least one site readable company-wide that should not be. Before AI that exposure was theoretical, because nobody browses folders they do not need. AI makes it queryable in plain language.
Do we need one AI tool or several?
Two is common and reasonable. An in-suite assistant for document and email work, plus a standalone tool for the technical and analytical staff who will hit the limits of an in-suite assistant quickly. Two tools is cheaper than one bad choice.
What contract terms matter when buying AI software?
Three: whether your inputs are used to train models, data residency and retention, and whether the vendor will sign a business associate agreement or support the flow-down terms your own client contracts require. Read the terms for the tier you are buying, not the marketing page.