AI spend by role
A company can count the tokens each team uses. The count does not say whether the work in a role suits AI, or whether anything came of it.
Is your AI spend in the right roles?
JobRoute joins three things for every role: the tokens the role uses (a token is the unit AI vendors bill by, roughly a piece of a word), how much of the role's work AI can do, and what the role produced. You see where the spend fits the work and where it does not.
The free sample scores 10 of your roles task by task, from job titles alone. Connecting token use and output starts with a call.
One role, three feeds
Paralegals and Legal Assistants
- Token use
- What the team spends on AI, by model, read from the usage reports of your AI tools.
- Work AI can do
- Task exposure 66 out of 100. 9 of 12 tasks score 60 or more for AI exposure, 2 observed in AI use.
- Model tier the work needs: Standard
- Output
- The measure the team already tracks, such as documents drafted or matters closed.
Three feeds, joined for every role
A spend figure is hard to judge until you know what the role's work is and what the role produced. JobRoute puts the three side by side.
- Token use: what the role spends
JobRoute reads the usage reports of the AI tools your company already pays for, with read access only: which model, how many tokens, in which team.
- The role's tasks: what AI can do there
Every task in the role is scored for how much of it AI can do, and for the size of model it needs. It is the same public score shown on every job page, explained in how it works.
- Output: what the role produced
The measure the team already tracks, read from the system where the work happens and set beside token use for the same team and period.
The result is one line for each role: the spend, whether the work suits AI, and whether output moved.
Every role has its own output measure
Engineering teams count pull requests. Legal, finance, support and sales each have a system that already records what the team produced. Your company picks the measure for each role, and JobRoute reads it.
| Role | Output measure | Read from |
|---|---|---|
| Software developer | Pull requests merged, review time | The code hosting system |
| Paralegal | Documents drafted, matters closed | The document and matter systems |
| Finance analyst | Days to close, invoices processed | The finance system |
| Support agent | Tickets resolved, time to resolve | The help desk |
| Sales representative | Meetings booked, pipeline created | The CRM |
Examples of measures a team can start with. When an output measure moves in the same period as token use, that is an association. It does not prove the tokens caused the change, so JobRoute shows the two side by side and leaves the judgment to you.
Where each role lands
Two questions place a role on the map: how much the role uses AI, and how much of its work AI can do. The output measure then shows whether the use is paying off.
-
High use, few tasks fit
Question the spend
Heavy token use in a role where little of the work suits AI.
-
High use, many tasks fit
Right place
Check that output moved, and that the model is no bigger than the tasks need.
-
Low use, few tasks fit
Leave alone
There is little here for AI to do, so there is no rollout to push.
-
Low use, many tasks fit
Missed
Work AI can do, done without it. Train and roll out here first.
Token use is high in the top row and low in the bottom row. The share of the role's work AI can do is low on the left and high on the right.
What a leader reads
One line per role. Each row opens to the teams behind it, the models they use and the tasks where a smaller model would do the same work.
The second row in this example is the one a usage count leaves out. People use AI on work that suits it and the output measure has stayed flat, which points to how the work is organized around the tool.
| Role | Token use | Work AI can do | Output | The read |
|---|---|---|---|---|
| Software developer | High | High | Up | Keep funding |
| Financial analyst | High | High | Flat | Used, not yet paying off |
| Paralegal | Low | High | Flat | Train and roll out first |
| Production supervisor | High | Low | Flat | Question the spend |
| Maintenance technician | Low | Low | Flat | Leave alone |
Example layout with made-up rows. Not customer data. Token use, output and the read are invented to show the layout. The work AI can do column follows the public task score on each role's job page.
Measured by role and team
The question is where AI fits the work, and that is a question about roles. Measuring groups is also what an HR leader, and a works council where there is one, can agree to.
- Groups first
- Every report is by role and team. The roster behind it has no field for names or emails, and employee IDs are hashed before anything is stored. How we handle your data.
- No token score for people
- A token count is not a performance measure. A ranking by token count rewards use, whatever the use was for, so JobRoute does not produce one.
- A question for the review
- A manager can bring the role's figures to a performance review: what the work calls for, what the team used and what it produced. When one person's usage is far from that and output has not moved, the manager asks why and the person answers. A person makes the decision.
- The rules that apply
- Several jurisdictions regulate automated decisions about work and pay, and systems that monitor how workers perform. Our AI and employment compliance calendar lists them with dates and sources. This is general information, not legal advice.
Background reading.
The public task data behind the score, and the rules on AI at work.
Questions about AI spend by role
What is a token, and why look at token use by role?
A token is the unit AI vendors bill by, roughly a piece of a word. Token use shows how much a team draws on AI and what that costs. Grouped by role and set beside the work each role does, it shows whether the spend sits where AI can do a large share of the work.
Where does the token use data come from?
From the usage reports of the AI tools your company already pays for. JobRoute reads them with read access only and groups the usage by role and team: which model, how many tokens, in which team.
How does JobRoute know how much of a role's work AI can do?
Each role is broken into its tasks, and each task is scored from 0 to 100 for AI exposure from public data. The same score records which size of model each task needs. Read how it works or the methodology, and check any role on its job page.
Does JobRoute track individual employees?
Reports are by role and team. A token count is not a performance measure, and JobRoute does not produce a token score or ranking for people. The roster has no field for names or emails. Read how we handle your data.
Can the figures be used in a performance review?
Yes, as questions for the conversation. JobRoute gives the manager the figures for the role: how much of the work AI can do, what the team used and what it produced. The company already holds each person's own usage in its AI tools. When that usage is far from what the role's work calls for and output has not moved, the manager asks why and the person answers. JobRoute does not score or rank people, and a person makes every decision. Rules on automated decisions about work and pay may apply, so check with counsel. This is general information, not legal advice.
How is this different from an AI usage or spend dashboard?
A usage or spend dashboard shows who is active and what it costs. JobRoute starts from what each role's work is, task by task, so it can say whether the spend is in roles where the work suits AI, and it sets the role's own output measure beside it.
See where your roles land.
Send us up to 10 job titles, with no employee data, and get the task score and model tier for each within two business days. Or book a call to talk through your AI tools and the output measures you already track.
A founder replies within one business day.