Buy a subscription when the task is common, one person does it, and they can paste in what the tool needs. Pay for implementation when the work crosses several systems, needs your own data and rules, has to respect permissions, or needs a person to approve the result at a fixed step. Most businesses end up with both: rented AI capability, plus workflow design they own.
The workflow at a glance
- The buyer question: We already pay for AI seats and the work hasn’t changed. Do we need another tool, or something else?
- The current process: Individuals use chat assistants for drafting. Results vary by person, outputs are copied by hand into email, the CRM or documents, and nobody can say what the seats are worth.
- What can improve: One defined workflow where AI runs at a known step, with the right context fed in automatically, the output written to the right system, and a named person approving it.
- Tools and access needed: A list of current subscriptions and the AI features in each plan (check vendor documentation, not marketing pages), admin access to the systems the workflow touches, and a written AI-use rule.
- Where people still approve decisions: Anything sent to a client, anything involving price, commitments, personnel or legal terms, and the decision to switch any automated step on.
- How to measure whether it works: Time per item, edits needed per AI draft, errors caught at review, and whether seat usage tracks the chosen workflow or stays spread across ad-hoc use.
Examples marked “Hypothetical example” are illustrations, not client results.
Why more seats often don’t change the work
Seats give individuals AI. They don’t change how work moves between people and systems. In McKinsey’s 2025 survey, redesigning workflows had the biggest effect on whether organizations reported bottom-line (EBIT) impact from generative AI, out of 25 attributes tested. Only 21% of respondents said their organizations had fundamentally redesigned at least some workflows (McKinsey, 2025).
Microsoft’s 2024 Work Trend Index found that 75% of knowledge workers were using AI at work and 78% of those users were bringing their own tools (Microsoft, 2024). Usage is already high. Many teams lack a shared process, shared context and a clear point where a person checks the output.
What a subscription does well
A general AI subscription fits when:
- the task is common (drafting, summarizing, rewording, brainstorming),
- one person starts and finishes it,
- the information fits in what they can paste or upload,
- a mistake is easy to spot and cheap to fix.
Vendors handle the model, updates and hosting. You still own how it’s used. As one build-vs-buy guide puts it, buying “does not remove implementation. The business still owns configuration, permissions, training, process change, vendor oversight and the consequences of output” (Most AI Labs).
Hypothetical example (not a client case): A consultant uses an AI assistant to tidy meeting notes before writing them up. It’s one person, low risk, and easy to check. Another subscription or a custom build would add nothing.
When you need implementation work
Implementation means designing and connecting a workflow. Nobody writes a model from scratch. It is worth paying for when one or more of these is true:
- Several systems. An inquiry arrives by form and needs a CRM record, a reply draft and a task for the owner. A chat window can’t do that by itself.
- Your context, every time. Your service descriptions, approved proposal wording or client history should arrive automatically. Today they get pasted in by whoever remembers.
- Permissions. Not everyone should see every client file, and the workflow should respect that.
- A fixed approval step. The design decides where review happens, so it doesn’t depend on habit.
- Measurement. A defined workflow has a baseline and an acceptance test. Ad-hoc use doesn’t.
The same guide frames the decision as “which layers should the business own, and which should it rent,” and notes that a vendor may manage the model while the business owns the workflow (Most AI Labs).
Hypothetical example (not a client case): A design studio pays for AI seats, but each proposal still takes an evening. A new tool won’t fix that. A workflow will: discovery notes saved to the CRM trigger a draft built from the studio’s approved template and past wording, the draft lands in the shared drive, and the founder edits and approves scope and price before anything is sent.
A five-question test before you buy anything
Ask these about the specific job you want help with:
| Question | If yes, lean toward |
|---|---|
| Does one person do it start to finish? | Subscription |
| Does it touch two or more systems? | Implementation |
| Does it need your own documents or records each time? | Implementation |
| Would a mistake reach a client or cost money? | Implementation with an approval step |
| Do you already pay for a tool with this feature? | Configure what you have first |
The last row matters most. Before buying anything, list what you pay for now and what each plan includes. Many CRMs and office suites include automation features at certain tiers. HubSpot, for example, documents workflows with enrollment triggers and actions (HubSpot Knowledge Base), but availability depends on your plan. Confirm it in the vendor documentation.
Count the full cost
Seat prices are easy to compare and misleading on their own. For a subscription, add administration, training and the time staff spend moving outputs between tools. For implementation, add discovery, build, testing, the ongoing cost of any automation platform or model usage, and maintenance when a connected tool changes. Third-party guides recommend comparing total cost over several years, not monthly price alone (Dr Dave Heath). We don’t publish standard prices here because the right answer depends on scope. See what business automation costs and how to scope it.
Data and risk rules apply either way
Rent or build, write down what data can go into which tool. The FTC’s guide for businesses recommends knowing what personal information you hold, keeping only what you need, and protecting it (FTC). NIST’s generative AI profile lists confabulation, meaning confidently stated false content, among the risks to manage (NIST AI 600-1). So put the approval steps in the design, where they don’t depend on someone remembering.
Where people stay in charge
A good implementation makes approvals more reliable:
- Drafting. AI can prepare a proposal, a reply or a report summary. A person approves scope, price and anything sent outside the business.
- Switching on automation. Run new steps in “draft only” mode on real cases first. Then a named owner decides what, if anything, runs without review.
- Exceptions. Anything outside the rules goes to a person. The system doesn’t guess.
How to measure whether it works
Take a baseline on the chosen workflow before changing anything: time per item, items per week, edits needed, and errors found. After launch, measure the same things on the same kind of work. Track adoption too. If people go around the workflow, fix the design. More seats won’t help.
Checklist: subscription or implementation?
- I’ve named one specific job, not “use AI more.”
- I’ve listed current subscriptions and the AI features in each plan.
- I know which systems the job touches.
- I know which outputs need human approval.
- I have a baseline for time and errors.
- I’ve written a data rule for what goes into AI tools.
Related guides
- How to choose the first business workflow to automate
- How to implement AI across your team’s existing tools
- What business automation costs and how to scope a project
If you’re paying for AI seats and the work hasn’t changed, bring us the one job you’d most like to fix. Grownup Business implements AI-assisted workflows in the tools you already use, with approval points agreed up front and a written scope before any payment. Request a demo to discuss your workflow.
If Grownup also manages your Instagram, the same approach can cover inquiries that start there. That’s an optional, separate service (Instagram management); you don’t need it to start a business workflow.