The call
404 Media reported on 15 October, from an internal memo titled “Claude Enterprise Users: Important Updates to Licensing and Usage Costs”, that Accenture is “centrally funding only $29 a month per person towards Claude usage”, effective 1 October (independent). The memo says “all costs are consumption-based”. Usage above $29 needs a Work Breakdown Structure, a project or team code, to pay for it. Accenture’s spokesperson: “At the foundation, we make AI available to all our people.”
What it is, and what it is not
What it is. A budget rule. The centre pays the first $29 of each person’s Claude usage a month; projects pay the rest. Premium plans include $250 of token usage a week. The memo adds that a project code and a licence do “not guarantee indefinite or ongoing access to Claude”, and one employee told 404 Media their access was revoked (all independent, 404 Media).
What it is not. It is not a ban, and it is not a verdict on Claude. It is not an Anthropic price change; the report mentions none. And it is not unusual pricing. Claude Enterprise lists at $20 per seat a month, billed annually, plus usage at API rates (independent, Justin McKelvey, citing Anthropic’s pricing page). Once usage is metered, someone has to own the meter. Accenture has decided the project owns it. Most enterprises have not decided anything yet.
The number nobody is quoting
Coverage leads with $29. The figure that matters is the gap between the two tiers, and what lands on project codes.
| Item | Figure | Label |
|---|---|---|
| Central allowance | $29 a month | independent (404 Media) |
| Premium allowance | $250 a week | independent (404 Media) |
| Premium, monthly | $250 × 52 ÷ 12 = $1,083 | our maths |
| Premium against central | $1,083 ÷ $29 = 37x | our maths |
| Heavy user, reported | up to $1,000 a month | independent (404 Media) |
| Moved to project codes, per heavy user | $1,000 − $29 = $971 a month; $11,652 a year | our maths |
Now what $29 buys. Anthropic lists Sonnet 5.5 at $2 input and $10 output per million tokens and Opus 5.5 at $4 and $20 (vendor, via eesel, citing Anthropic pricing docs and Artificial Analysis). On the Artificial Analysis index at medium effort, a task costs $0.59 on Sonnet 5.5 and $1.34 on Opus 5.5 (independent, same source).
| Model | Index tasks for $29 | Per working day (21 a month) | Output tokens for $29 | Label |
|---|---|---|---|---|
| Claude Sonnet 5.5 | $29 ÷ $0.59 = 49 | about 2 | 2.9 million | our maths |
| Claude Opus 5.5 | $29 ÷ $1.34 = 21 | about 1 | 1.45 million | our maths |
One Opus task a working day. That is a chat allowance, not an agent allowance. Scale it to a 20-person Salesforce delivery team at the reported $500 to $1,000 a head: $10,000 to $20,000 a month, of which the centre funds $580 and $9,420 to $19,420 lands on project codes (our maths). At list prices, a $20 seat plus the $29 allowance is $49 a person a month, so the allowance is already bigger than the seat (our maths).
Missing
What the report does not tell a buyer, as of 16 October:
- Whether a client project’s code passes the Claude cost through to the client.
- How many people sit on the $29 tier and how many on Premium, and who qualifies.
- Which products the $29 covers: chat, Claude Code, API, or all three.
- Accenture’s contracted rate against Anthropic’s list price.
- What happens to work in flight when someone hits the cap mid-month.
- Any measure of output per dollar behind the $29 figure.
Who owns what
The allowance is finance’s job. Whether $29 of Claude finishes the work is engineering’s.
On a Salesforce programme this is a statement-of-work question before it is a tooling one. If your integrator’s people run Claude Code against your org, that spend sits on a project code somewhere: theirs, as overhead in the day rate, or yours, as a pass-through line. Neither is wrong. Not knowing which is. Inside your own team, treat AI spend like any cloud resource: tagged to the project, reviewed monthly, and owned by the person who owns the project budget. If you run delivery in Salesforce, the project record already exists; put AI cost on it the way you put hours.
Commercial decision: a flat central allowance is simple and starves heavy users; full chargeback is fair and slows adoption. Pick the default deliberately, then write the exception route.
Monday morning
- Step 1. Pull the distribution. Ninety days of AI usage per person from your admin consoles: median, top 10%, and top 1%. Owner: IT or AI platform lead.
- Step 2. Set the default from the median, not the mean. Then define the exception route and tie it to a project code with a named budget owner. Owner: finance partner.
- Step 3. Ask every integrator in writing. How is AI usage billed on our statement of work, which models, what cap, and who approves overage? Owner: procurement and the Salesforce programme owner.
- Step 4. Measure cost per finished task. Two real workloads, Sonnet against Opus, so the cap is set on output rather than tokens. Owner: engineering lead.
- Step 5. Review in 30 days. Who hit the cap, what slipped, and what moved to Premium. Owner: CIO.
We move this to WATCH if Accenture or other integrators confirm AI usage is absorbed in day rates and never billed to clients. We stay at PILOT while that is unanswered.
Frequently asked questions
What did Accenture change about Claude?
From 1 October 2026 Accenture centrally funds $29 a month per person of Claude usage. Anything above that needs a project or team code to pay for it. Premium plans include $250 of token usage a week. 404 Media reported it on 15 October from an internal memo.
How much Claude does $29 a month buy?
About 49 Artificial Analysis index tasks on Sonnet 5.5 or 21 on Opus 5.5 at medium effort, which is about one Opus task a working day. In tokens, $29 is 2.9 million Sonnet 5.5 output tokens or 1.45 million on Opus 5.5.
Should we cap AI spend per employee?
Set a default allowance from your median usage, not the mean, and give heavy users a route to a project budget with a named owner. Then ask every integrator how their AI usage is billed to you.
Living page. Last updated 16 Oct 2026. What would change our call: integrators confirming AI usage is never billed through to clients, or Anthropic changing Enterprise pricing.