AI Tool Access & Budget Framework
Leader decision guide for AI tool allocation across business units
Bottom line
Default everyone to ALIA; it's free to your budget. Grant Claude Enterprise only when a person has a specific, frequent, evidenced capability gap that ALIA can't close. Claude Enterprise is a real cost your BU owns, so every seat is a call you're accountable for.
One trap to name up front: more usage is not more value. Someone burning through their token cap, chasing usage for its own sake (what some call "tokenmaxxing"), is not automatically more productive than someone doing the same job on ALIA. Optimise for outcomes per dollar, not tokens consumed.
Four habits that keep this simple
- ALIA is the default for everyone on your team.
- When someone wants Claude Enterprise, it's worth forwarding once it clears all four gates (Section 2).
- A quick monthly read of the usage report, plus the three cost questions, keeps things on track (Section 5).
- Seats that sit unused are worth handing back. Granting and never revoking is how budgets quietly balloon.
1. The Two Tiers
Every employee routes to one of two AI platforms. The default is ALIA. Claude Enterprise (CE) is the exception, reserved for users where ALIA demonstrably cannot deliver.
ALIA
Who pays: the AI Lab, centrally. $0 to your BU.
Internal platform, broad access, full group control. The default for everyone.
- Free to your cost center: no per-seat charge to your P&L
- General productivity and content drafting
- Q&A, summarization, translation
- Internal knowledge retrieval
- Available to all employees
- Internal guardrails, no external dependency
Claude Enterprise
Who pays: charged to your BU's cost center.
Advanced reasoning, very large documents, app integrations
- Tackles complex, multi-part analysis end to end
- App integrations: Office, Granola, Fireflies, Canva, and more
- Works across very large documents at once
- Restricted access, request-based only
- Centrally governed, with a usage cap set per person
The principle
The question is never "who deserves the better tool." It's: does Claude Enterprise unlock something that's currently blocked? If ALIA handles the job, paying for Claude Enterprise is waste. ALIA costs your budget nothing; Claude Enterprise is a real line item, so the bar for the exception is high. Seniority, title, and enthusiasm are not qualifying criteria. A blocked workflow is.
2. The Four Gates
Every request for Claude Enterprise passes through four gates. If it doesn't clear a gate, Claude Enterprise isn't provisioned and the person continues on ALIA. No scoring, no interpretation. Pass or fail.
The answer must be a concrete capability, not a quality difference. "Better writing" is not a gap. "Building pivot tables directly inside Excel" is a gap. "Slightly better reasoning" is not a gap. "Reviewing an 80-page contract in one pass" is a gap.
Examples of real gaps
| Function | The capability they need |
|---|---|
| Finance | Build and update financial models directly inside Excel |
| Strategy | Pull 15+ long documents into one clear thesis in a single pass |
| Product | Work across a long product spec and all its supporting docs at once |
| Investments | Build a financial model and refine it over a long working session |
| Legal | Review highly confidential contracts under a written guarantee that nothing is stored or used to train a model |
| Innovation | Turn research into pitch-ready concepts and visuals through app integrations like Canva |
Examples of NOT gaps
| Claim | Reality |
|---|---|
| "Claude writes better emails" | ALIA writes fine emails. Preference is not a gap. |
| "I prefer the Claude interface" | UX preference. Not a capability blocker. |
| "Claude is faster" | Speed difference. Not a blocked workflow. |
| "I need AI for my presentations" | ALIA generates the content. High-volume deck production can qualify as a temporary gap (see below); the occasional slide does not. |
| "Everyone on my team should have it" | Not a gap. Each person assessed individually. |
| "All my history and saved work is already in Claude" | Not a gap. ALIA is adding import from other AI tools, including memory, so your history and context move over. The switch is temporary, not a lock-in. |
Temporary gaps (e.g. slide / deck generation)
Some gaps are real today but already on ALIA's roadmap. Slide generation is the main one: ALIA can't build decks yet, but the capability is coming, and it will not exist by the time this form goes live, so people will raise it as a gap. Handle it as temporary, not structural:
- High-volume deck builders (daily/weekly, client- or board-facing decks) can pass the gates, but the seat is time-boxed: provisioned with an expiry/review tied to ALIA shipping slide generation, then migrated back. It is not a permanent seat.
- "I need it for the occasional presentation" is not a gap. ALIA drafts the content; assembling slides by hand is a minor workaround, not a blocked workflow.
- Never grant a permanent CE seat on the strength of a capability that is weeks-to-months from being free in ALIA.
The gap existing isn't enough. If someone hits it once a quarter, the workaround cost is trivial. Frequency is the multiplier that turns a capability gap into a budget justification.
| Frequency | Decision | Reasoning |
|---|---|---|
| Daily | Strong candidate. Proceed. | The gap is structural to their role. They work around it every day. |
| Weekly | Moderate. Proceed, start at lower tier. | Meaningful friction but not constant. Base or Standard tier. |
| Monthly | Weak. Challenge it. | Is the workaround really that painful 12 times a year? Probably manageable. |
| Rarely | No. Stay on ALIA. | The tool will sit unused. Waste of budget. |
This is the anti-inflation mechanism. Everyone will say they need the better tool. The gate is simple: show the blocked workflow. Paying for a personal subscription helps the case, but it doesn't replace it, they still walk through what they actually do.
| Evidence type | What it proves | Strength |
|---|---|---|
| Can show 3+ real examples where ALIA couldn't do the task | Tried ALIA, hit the wall, documented it. This is what we're looking for. | Core |
| Can walk through the workaround and what it costs in time | "Without the integration I rebuild this in Excel by hand: 3 hours. With it, 20 minutes." Concrete and quantified. | Core |
| Already pays for a personal Claude / AI subscription | A real signal they rely on it, but not proof on its own. They still show the workflow above. | Supporting |
| "I think it would help me" with no examples | Aspiration, not evidence. | Not evidence |
| Manager vouches for them | Useful context, but the person still demonstrates it themselves. | Not evidence |
Don't guess. Don't let the leader pick the tier that "feels right." Anchor to what's observable about their current usage.
| Tier | Annual Cap | Anchor |
|---|---|---|
| BASE | $250 | Equivalent of a $20/mo plan and they don't max it out. Weekly gap frequency. Plugin-dependent but not daily-intensive. |
| STANDARD | $500 | Equivalent of a $20/mo plan and they regularly max it out. OR: no prior usage data (this is the safe default). Daily-to-weekly frequency. |
| POWER | $750 | Currently on a $100/mo personal plan. Daily frequency. Heavy analytical workflows. |
| MAX | $1,200 | Currently on $100/mo AND hitting limits regularly. The heaviest users. Exception, not norm. |
Default rule
When in doubt, start at Standard ($500). It sits near typical enterprise consumption and is a safe mid-point (confirm the current benchmark with the AI Lab). Raising a cap takes one approval. Revoking a license damages trust. Start low, adjust with data.
3. Where Leaders Get It Wrong
Four instincts that feel reasonable in the moment but lead to the wrong call. The gates exist to catch them.
Reading seniority as need
"She's a VP, she should have the best tool." It's natural to assume seniority means need, but it doesn't. If she uses AI 10 minutes a day for email polish, ALIA is the right fit regardless of title. A junior analyst using Claude for hours a day on financial modeling is the stronger case.
Reading heavy usage as a gap
"He uses AI more than anyone on my team." Heavy use of ALIA doesn't mean Claude Enterprise is needed. If ALIA handles his tasks well and he's simply a power user, he stays on ALIA. High usage with no capability gap is just a happy ALIA user, and remember, more usage isn't more value.
Acting on enthusiasm, not evidence
"She's really keen to explore what Claude can do." Enthusiasm is welcome, but it isn't a blocked workflow. The 60-day revisit is there for exactly this: come back when she can show what ALIA couldn't do for her.
Granting seats, then walking them back
A leader approves 15 seats, the first invoice lands higher than expected, and access gets pulled. Granting a tool and then removing it is hard on people and on trust. The four gates avoid this by only letting through real, evidenced need up front.
4. How Gaps Differ by Function
The four gates are universal. The answers differ by department. Use this as calibration, not as rules.
| Function | Typical gap | Typical frequency | Expected outcome |
|---|---|---|---|
| Finance & Controlling | Excel/PPT plugin for financial modeling and board reporting | Daily for modelers, weekly for controllers | Highest CE penetration. The plugin gap is binary and hard to dispute. |
| Strategy & BD | Multi-source synthesis, long-context reasoning for strategy docs | Daily during strategy cycles, weekly otherwise | High CE penetration. Many already on personal subscriptions (strongest evidence). |
| Investments | Scenario modeling, due diligence synthesis across large datasets | Daily during evaluation periods, weekly otherwise | Passes easily during active deal flow. Tier depends on deal volume. |
| Brand Management | Complex sell-through analysis, brand principal deck creation via plugin | Weekly for directors, monthly for junior BMs | Mixed. Directors likely pass all gates. Junior BMs likely don't pass frequency gate. |
| Own Concepts | Buying/merchandising modeling in Excel; e-commerce analytics synthesis | Daily for buyers, weekly for e-comm analytics | Buyers with heavy Excel workflows pass. Marketing/CRM teams stay on ALIA (content is ALIA's strength). |
| Legal & Compliance | Contract analysis with ZDR compliance guarantees | Weekly for contract-heavy roles | Unique gate: compliance posture (ZDR) is the gap, not reasoning quality. Do NOT provision until ZDR is confirmed in writing. |
| Product & Innovation | Extended context for spec synthesis, technical feasibility reasoning | Spiky: daily during exploration, low during execution | Passes during active build/research phases. May not justify a permanent seat; revisit quarterly. |
| Marketing & Comms | Likely no gap. Content generation is ALIA's core strength. | Low | Most marketing tasks are ALIA-fit. Only VP-level with strategic deck needs or analytics leads might pass Gate 1. |
| Distribution & SC | Demand planning/forecasting modeling | Weekly for planning roles, rare for ops | Classic "revisit later" department. The gap exists at planning level but readiness is low. Start with ALIA. |
| HR & People | Comp & benefits modeling (Excel complexity) | Monthly at best | Almost entirely ALIA-fit. Process work, template work, policy work. One or two comp analysts might pass. |
| Retail Operations | Minimal. Operational work is structured and template-driven. | Very low | ALIA for 90%+. Only regional managers with cross-store reporting might pass. |
| Technology & Data | API access for building, Claude Code for engineering | Daily | Different access model entirely. Engineers need API/Claude Code, not per-seat web licenses. Separate budget, separate gate. |
5. Cost Control
This tool sits on your P&L. The four gates are your cost control mechanism: every seat that passes all four gates is justified. Every seat that doesn't is waste.
Treat AI tooling like headcount, not like software licenses
A Claude Enterprise seat is not a SaaS license you bulk-buy. It's a productivity investment in a specific person. If that person isn't delivering measurably more because of the tool, the seat is wasted OpEx. Evaluate each seat the way you'd evaluate whether someone needs a second monitor or a business class ticket: does the output justify the cost for this specific person?
Team Budget Calculator
Enter your numbers and the totals update live. ALIA is free, so only Claude Enterprise and other tools hit your budget. CE usage shows each tier's annual cap (the ceiling, not a forecast).
Healthy cost ranges
Because ALIA is free to your budget, the number that matters is your Claude Enterprise spend per employee (total CE cost รท full team headcount, including everyone on ALIA). It should fall within these ranges. If you're significantly above, scrutinize whether every CE seat is truly passing all four gates.
| Department Profile | CE Spend / Employee / Year | Why |
|---|---|---|
| Strategy / Finance / Analytics-heavy | $150-350 | High CE penetration justified by a daily app-integration or large-document gap |
| Brand / Own Concepts / Investments | $60-150 | Mixed: some pass all gates, many stay on ALIA |
| Marketing / Innovation / Product | $30-100 | Mostly ALIA-fit; CE seats are the exception |
| Retail Ops / Distribution / HR | $10-50 | Primarily ALIA, minimal CE |
How you'll see usage
You don't need to chase this data. Each month the AI Lab sends every leader a short CE usage report for their team, so the accountability comes with the means to act on it.
What the monthly report shows
- Per CE user: % of their cap used, and the trend vs. prior months
- Flags for anyone under 20% of cap (underused) or hitting their cap early (maxing out)
- Total team CE spend vs. the budget your calculator set
- A reminder that high usage is not the target: the flags are prompts to check value, not to reward consumption
Self-serve dashboards may come later; for now this is a pushed report, no tool to log into.
Monthly cost rhythm
Three questions each month, answered straight from the report above. Build into your existing cost review.
1. Who's not using it?
Any CE user under 20% of their cap for two consecutive months? Either the gap wasn't real (downgrade to ALIA) or they need enablement help. Either way, the seat isn't delivering value today.
2. Who's maxing out?
Any CE user hitting their cap before month-end more than once? Either upgrade their tier (the gap is bigger than estimated) or investigate whether they're using it for tasks ALIA could handle (the gap was misidentified).
3. Is the total tracking to budget?
Your budget calculator gave you an annual number. Are you on track? If you're 20%+ over after 90 days, it means either a gate was too lenient (revisit evidence) or tiers were set too high (adjust down). Reclaim underused seats before adding new ones.
6. Ongoing Governance
Migration triggers
| Signal | Action |
|---|---|
| ALIA user hits the wall with documented examples | Run them through the four gates |
| CE user barely uses it (under 20% cap, 60 days) | Downgrade to ALIA, reclaim seat |
| Personal subscription user doing work stuff | Priority: move to compliant tool immediately (skip to Gate 4) |
| CE user hitting cap regularly | Tier upgrade with usage data as justification |
| New team member in a role where predecessor had CE | Not automatic. Run them through gates independently. |
| ALIA ships slide / deck generation | Migrate all time-boxed "slide gap" users back to ALIA; close those CE seats unless another gate still applies. |
Review cadence
| When | What | Who |
|---|---|---|
| Day 30 | First usage check: is the tool being used? Is the gap real? | AI Lab + Leader |
| Monthly (Q1) | Cost rhythm: underuse, overuse, budget tracking | Leader |
| Quarterly | Cost reconciliation: actual spend vs. budget, seat utilization | Leader + Finance |
| Annually | Full framework review: are the gates still right? Has the tool landscape changed? | AI Lab |
Approvals, escalation, and timing
| Question | Answer |
|---|---|
| How long does a request take? | Target: a decision within 5 working days of manager sign-off. |
| What if a request is declined and the leader disagrees? | Escalate to the AI Lab lead with the business case. Genuine edge cases are decided there, not re-argued at the gate. |
| Who owns the overall budget? | The AI Lab sets the group-level envelope; each leader owns their BU's share within it. |
Leaving, terminations, and renewals
| Event | What happens |
|---|---|
| Someone leaves or is terminated | Access is revoked automatically at offboarding through SSO, the moment employment ends. The seat is reclaimed. No leader action and no cancellation request needed. |
| Someone changes team or role | The seat does not move with them. Their new manager re-justifies it through the four gates, or it's reclaimed. |
| Renewal | Nothing auto-renews. Each seat is re-justified at the quarterly and annual reviews above; a seat that no longer clears the gates is not renewed. |
Feature access layers
Claude Enterprise includes multiple capability surfaces. Not all are provisioned by default.
| Surface | Default access | Gate |
|---|---|---|
| Web + App | Yes, for all CE users | Included in standard provisioning |
| App integrations (Office, Granola, Fireflies, Canva, ...) | Yes, for all CE users | Included; the Office integration is the primary gap driver |
| Claude Code | No | Separate request. AI Lab validates. Engineering/technical roles only. |
| Cowork (computer use) | No | Research preview. Sandbox testing first. Not provisioned broadly. |
Appendix A: The Request Form
The employee fills this in, not their manager, because only the person with the blocked workflow can describe it. Most of it is auto-filled or a single tap, the target is under five minutes. Approvals route automatically after they submit. It's ready for Tech & Data to build in Microsoft Forms; each answer maps to a gate.
| # | Question the employee sees | Answer type | Maps to |
|---|---|---|---|
| 1 | You: name, work email, department, cost center, line manager | Auto-filled from your profile, just confirm | Audit & cost attribution |
| 2 | In one sentence, what can you not do in ALIA today? | Short text | Gate 1: Gap |
| 3 | Which best describes it? App integration (Office, Canva, Granola, Fireflies, ...) · Working across very large documents · Confidential data that can't be stored or used for training · Slide / deck generation · Something else | Single choice. Picking "Slide / deck" shows a short note that this access is temporary, with a tick to acknowledge. | Gate 1: Gap |
| 4 | How often do you hit this? Daily · A few times a week · Weekly · Monthly · Rarely | Single choice. "Monthly" or "Rarely" shows: this is usually an ALIA use case, you can still submit. | Gate 2: Frequency |
| 5 | Show us. Upload up to 3 examples where ALIA fell short, and/or describe your workaround and the time it costs. | File upload + short text. At least one required, intent alone doesn't pass. | Gate 3: Evidence |
| 6 | Do you already pay for a personal Claude / AI plan? No · About $20/mo · $100+/mo | Single choice. Supporting evidence, and it sets your starting tier. | Gates 3 & 4 |
| 7 | What other AI tools do you already use for work? Granola, Fireflies, Otter, Adobe AI, Gamma, Canva, ChatGPT, Copilot, other... and rough monthly spend if you expense any. | Multi-select + optional spend. No wrong answer, this is for visibility and overlap, not judgement. | Tool visibility |
| 8 | Anthropic training: confirm you've completed it and upload your certificate | Tick + file upload. Required: no certificate, no access. | Access requirement |
| 9 | Acknowledge: ALIA is the default and Claude Enterprise is for an evidenced, recurring gap; usage is monitored and a seat under 20% of its cap for 60 days returns to ALIA. | Checkbox | Expectations |
Your starting tier comes from question 6: no paid plan or unsure goes to STANDARD $500 (the safe default), $20/mo to BASE $250 or Standard, $100+/mo to POWER $750 or MAX $1,200. AI Lab confirms or adjusts on review.
What happens after you submit
Questions about access, eligibility, or budget planning? Reach out to the AI Lab team.