TL;DR. AI reaches your company as a small set of products, sold by a handful of serious vendors, on pricing terms that change every quarter, deployed to specific roles. Five decisions sit on your desk. You already know how to make all of them. This frame makes those decisions easier.
You are getting AI handed to you in the wrong shape.
A vendor wants forty-five minutes. A consultant wants a six-month engagement. A direct report forwarded a podcast. The board wants a strategy. Your CFO wants a number. None of these conversations get easier if you read another industry briefing. They get easier when you stop accepting the frame the briefing arrives in.
This guide sets out the frame that every other guide on this site assumes. It is short on purpose. Once your mental model is right, the rest is execution.
AI leaves you with a handful of operating decisions
The picture you’re getting comes from three places. Vendor sales motions. Strategy decks from the major consultancies. The tech-press coverage that lands in your LinkedIn feed. All three describe AI the same way. A wave. A transformation. A roadmap. Something happening to your industry that you must respond to.
That framing isn’t neutral. It’s the framing that justifies a six-figure consulting engagement, a thirty-slide deck, and a quarterly steering committee. It makes you a passenger and the deck author the pilot. It also doesn’t help you decide whether to buy ChatGPT Business or Claude Team for your operations group on Monday.
The useful frame is operational. AI, as it touches your company, is a small set of software products. Made by a few serious vendors. Sold under pricing terms that changed last quarter and will change again next. Deployed to specific roles that get specific leverage. The interesting questions are: which product, which vendor, which tier, which roles, what to measure, what to restrict.
“What does AI mean for our industry” sits at the wrong altitude. You don’t ask what email means for your industry. You ask who needs Outlook and who needs Gmail and how much you’re paying per seat.
Five operating decisions determine your AI program
Strip the deck off and there are five decisions that sit on your desk. Every other AI question your org will ask this year is downstream of one of them.
Vendor. Which of the four serious vendors (OpenAI, Anthropic, Google, Microsoft) will your org pay? Probably one or two. The decision is which vendor’s commercial terms, security posture, and integration story fit your stack, rather than which model tops benchmark X. Vendors have business models. Read the business model before the release notes.
Tier. Which plan, at which seat fee and consumption commit, with which data terms. The Team plan and the Enterprise plan from the same vendor are different products with different capabilities and different admin controls. Team remains a flat seat plan. Enterprise adds access plus consumption. The distinction that will outlast current prices is whether the tier lets you see the meter and the automations. Pro and Max have no admin surface. Team gives you an org-wide on/off switch, but no audit log or Compliance API. Enterprise gives you all three. The default for a mid-market org is still almost always one tier cheaper for cost, but never below the tier where you can see consumption and scheduled tasks. Ignoring this is how budgets bloat.
Roles. Which jobs, on which teams, get a seat. A correct deployment looks lopsided: heavy in roles where the leverage is real and absent in roles where it is not yet. The leverage is uneven. Your seat distribution should be too.
Measure. What you watch to know it is working. Seat activation, training hours, and vendor-supplied “productivity uplift” charts do not answer that question. Watch observable changes in what your org ships, who ships it, and how fast. Add consumption by team for the metered work. The meter is a leverage map as well as a bill: it shows where useful work is landing. If your dashboard would look the same whether AI worked or not, it is not telling you anything useful.
Restrict. What stays off limits: customer PII in the consumer tier, proprietary code in a free account, or auto-generated communications without a human name on them. This is also where you decide which work may run unattended, because the tools now schedule themselves: an agent that compiles a report every Monday or a routine that touches a production system on a cadence. The rule for unattended work is one line. Only automate work two people can mechanically agree is right. Anything that needs a senior person to read it carefully before anyone trusts it does not go on a clock. This prevents scheduled automation that IT never approved, quietly running part of the business without an owner, monitoring, or visibility after its creator moves teams. The list is short and concrete. It fits on one page. A forty-page governance document mainly defends the organization; it does not govern the work.
Vendor, tier, roles, measure, restrict. Every guide on this site goes deep on one of these. If you cannot defend each with a specific answer and a specific number, you have a line item rather than a program.
The frame changes the questions you ask
A defensible budget. When the CFO asks why AI spend went up forty percent year over year, answer with “we moved twelve roles from the Team tier to the Enterprise tier because the data residency terms required it, and added eighteen seats in the operations group where the per-seat return is observable.” That gives them a decision they can evaluate. “Transformation” leaves the decision undefined.
A filter for vendor pitches. Every vendor meeting becomes a thirty-second exercise. Which of the five decisions is this pitch trying to change? Usually it’s tier. They want you on the more expensive plan. Sometimes it’s vendor. They want to displace your incumbent. Occasionally it’s roles. They want to expand the deployment. Once you know which decision is on the table, you know what to ask and what to ignore.
A way to evaluate any new model release in under a minute. The new model from Vendor X either changes one of your five decisions or it doesn’t. Most don’t. The ones that do are obvious: a price drop that changes your tier math, a capability that opens a new role, a data term that resolves a restriction. The rest is reading material for the engineering team. When you do want the current state of the board rather than the frame, the Q3 2026 briefing tracks what actually moved this quarter; this guide is the part that survives it.
A more useful vocabulary. “Our Claude Team contract,” “our policy on customer data in third-party tools,” and “the eight roles where we’ve seen real leverage” name the work. The language is smaller because the decisions are concrete.
The frame replaces strategy theater with decisions
You give up the comfort of the deck. There is no longer a forty-slide narrative about your AI journey. There is a contract, a policy page, a seat list, and a quarterly review of whether the seat list still makes sense.
You give up the strategic consulting engagement. Implementation help on a specific decision (which tier, which roles) can be useful and cheap. The strategic engagement that produces the deck is a tax on not having a frame.
You give up the org chart with an AI Center of Excellence on it. Centers of Excellence are what organizations build when they’re trying to look like they have a strategy without committing to one. The five decisions live with the people who already make those kinds of decisions. Procurement owns vendor and tier. Line managers own roles. The CFO owns measure. Legal and security own restrict. AI doesn’t need a new department. It needs the existing ones to do their jobs with one more product in the catalog.
The five decisions fit on one screen
| Decision | Default for a mid-market org | Where to go deeper |
|---|---|---|
| Vendor | Foundation on Anthropic; a secondary chat preference (OpenAI/Google) where users prefer it | Choosing Tools |
| Tier | One step below what the vendor recommends for cost, but never below the tier where you can see consumption and scheduled tasks | Choosing Tools, Evaluating Spend |
| Roles | Concentrated in operations, engineering, marketing, finance. Not everywhere. | Recognizing Leverage, Driving Adoption |
| Measure | Observable shipping changes by role; not seat activation | Measuring Returns |
| Restrict | One-page policy: customer data, proprietary IP, external comms | Building an AI Policy, Managing Risk |
The table is the entire leadership view of AI on a single screen. Every meeting, every pitch, and every budget question lands somewhere on it.
One unanswered decision tells you what to read next
The decision on that table you can’t defend with a specific answer and number is the guide to read next.
When the next vendor email arrives, its proposal belongs to one of the five decisions; its justification follows from the same frame. AI strategy resolves into a contract, a policy, a seat list, and a measurement.