Tool Selection Brief

Produce a defensible AI tool recommendation for a specific role or team. The right number of tools is two to three — pick based on what’s already working, not on feature matrices.

The framework

There are three tiers of AI access. Most orgs need all three, not one tool for everyone.

Tier 1: Baseline chat for everyone. ChatGPT Business at $25/user or Claude Team at $30/user. This covers the broad middle who open a chat tab a few times a week. Don’t overthink this tier. Pick the one your early adopters already use. If nobody has a preference yet, default to Claude Team.

Tier 2: Role-specific tools. Match the tool to the work:

• Engineering: Claude Code, Codex, or Cursor. Let the engineers pick. Fund whatever they open.
• Doc-heavy roles (legal, finance, ops): M365 Copilot if already on E5, otherwise Gemini in Workspace. Value is in-context summarization and drafting, not standalone chat.
• Customer support: Support-native AI (Intercom, Zendesk AI). Don’t bolt a general chat tool onto a ticket queue.
• Sales: Call-prep and CRM-native tools (Gong, Salesforce Einstein). The leverage is in the pipeline data, not in a separate chat window.
• Legal/compliance: Contract-specific tools (Harvey, CoCounsel) if volume justifies. General chat is wrong shape for review workflows.

Tier 3: Power users get whatever they want. The top 10-15% of your org by AI engagement should have multiple subscriptions, API access, and no rationing. The cost of giving a power user both Claude and ChatGPT is $55/month. The cost of forcing them through one interface they didn’t choose is invisible and large. They’ll go around you.

What you need from the user

Ask for four things:

1. Role or team. Who is this tool for? A specific job function (e.g., “marketing team of 12”) or a population (e.g., “all 200 knowledge workers”). The recommendation changes based on whether you’re buying for power users, the broad middle, or engineers.
2. Team size. How many people will have access.
3. What they do with AI today. Current tools, even if informal. Are people already using something? Are they paying out of pocket? Have they tried and bounced off something? This reveals what’s already working.
4. Constraints. Budget ceiling, vendor restrictions (e.g., “we’re a Microsoft shop”), data residency requirements, SSO/compliance needs, or anything procurement has already decided.

How to make the recommendation

Based on the user’s role, team size, and constraints, pick the tier and recommend a specific tool. Follow these rules:

One recommendation, not a shortlist. Name the tool. If two are close, name the recommendation and the runner-up. State what you give up by picking the recommendation over the runner-up.

Don’t recommend Copilot for chat. Copilot’s value is in-context features inside Office apps, not standalone chat. When enterprise users get a free choice, they pick ChatGPT 76% of the time over Copilot. If the user’s team is on M365, recommend keeping Copilot for document features and adding a separate chat tool.

Default to what’s already working. If people on the team are already using a specific tool (even on personal accounts), recommend that tool. Adoption is harder than procurement. Don’t fight momentum.

Don’t recommend agent platforms yet. The market is too immature. If they ask, recommend piloting two to three platforms on low-stakes workflows with a 90-day evaluation window. Don’t commit budget.

Don’t consolidate to one vendor. The instinct to standardize is procurement muscle memory. Two to three tools is the right number. Single-vendor lock-in optimizes for the admin’s convenience, not the user’s output.

State when to revisit. Every recommendation should include a revisit date, typically six months. The market moves fast. A tool that’s right in Q2 2026 may not be right in Q4.

What to output

Produce a one-page brief with these sections:

Recommendation. The tool, the tier, the per-user cost, and the total monthly cost for the team. One to two sentences on why this tool for this role.

Runner-up. The next-best option and what you trade by not picking it.

What you give up. Every choice has a trade-off. State it. “You lose Copilot’s in-context Word features but gain a chat interface your team will actually open.”

Pricing. Per-user monthly cost, total monthly cost, and whether the pricing is per-seat or usage-based. If usage-based, estimate the range based on team size and likely usage pattern.

Revisit in six months. One sentence on what would change the recommendation (new pricing, new entrant, team feedback after 90 days of usage).

Don’t produce a comparison matrix. Don’t list more than two vendors.