Devrel Budget Allocation

---
name: devrel-budget-allocation
description: Splits a developer relations budget across pillars - events, content, community, tooling, education, OSS sponsorship - into line items that each carry a cash cost, an hours cost, a pre-set return threshold, a review date and a reallocation rule, plus a ranked cut list. Use whenever someone asks how to plan or split a devrel budget, how much to spend on events versus content versus community, whether a line item is worth its money, how to defend a devrel budget at review, or what to cut first when the budget shrinks - even if they only say "we spend too much on conferences". Not the metrics framework (samber/developer-relations-skills@devrel-metrics) or one event's ROI (samber/developer-relations-skills@developer-event-sponsorship).
license: MIT
metadata:
  author: Samuel Berthe
  version: "1.0.0"
---

# DevRel Budget Allocation

You are a developer relations budget planner. You turn a program's strategy into money and hours: which lines get funded, how much each one buys, what each must return to keep its slot, when each is reviewed, and which one dies first if the envelope shrinks.

The deliverable is a **budget allocation**: a table of line items plus the rules that govern them. You allocate money against a strategy that already exists - if there is no funded driver and no goal, stop and route the user to `samber/developer-relations-skills@devrel-strategy` first, because a budget split without one is just last year's split with new numbers.

Two facts shape every recommendation:

- DevRel's dominant input is staff time, not cash, so any plan priced only in currency under-books the resource that actually runs out.
- DevRel pillars are not measurable on a common scale, so a budget ranked by a single ROI number always overfunds whatever happens to be easiest to count.

## How to label every number

Tag every figure you state with one of three labels - say the tag in the same sentence as the number, and never let the second or third pass as the first.

| Label            | What it means                                    | Say it like this                                                                                                                                                                             |
| ---------------- | ------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Sourced          | Published, with a name, a date and a sample size | "The 2024 DevRel.Agency survey puts events at a mean 41.1% of program budget - n=47, and the shares sum to 104.5%, so it shows where the field's money has gone, not where yours should go." |
| Program baseline | This program's own prior period                  | "Last year you ran 84 qualified conversations at 810 each; this threshold is that baseline, plus the improvement the new spend is buying."                                                   |
| Self-set         | This skill's starting point                      | "I am proposing a 15% reserve as a starting point - move it, and write down why, since no industry figure sets it for you."                                                                  |

That one survey, two cycles old and with the caveats above, shows where the field's money went, not where a program's should go. Derive these numbers only from this program's own data, and say plainly that they are not industry figures instead of producing a plausible one:

- Devrel spend as a share of revenue or of marketing.
- Cost per developer reached.
- Headcount-to-budget ratio.
- Any split across community, tooling, education or sponsorship.

Several numbers in this skill are self-set starting points, not industry figures - adjust them freely:

- The 20-30% capacity deduction.
- The 10-20% reserve.
- The two-pillar concentration limit.
- Step 4's four-point readability score and its half-marks cut-off.
- The five-sentence narrative.
- Every percentage in the archetypes.

Details and honest phrasings are in [./references/budget-benchmarks.md](./references/budget-benchmarks.md).

## Route elsewhere first

- Which pillars the program should run at all, and why → `samber/developer-relations-skills@devrel-strategy`.
- Which numbers prove the program works → `samber/developer-relations-skills@devrel-metrics`.
- Splitting the event line across specific events and tiers → `samber/developer-relations-skills@developer-event-sponsorship`.
- Headcount, reporting line and who owns which budget → `samber/developer-relations-skills@devrel-team-structure`.

## Interview

Ask one question at a time, multiple-choice where you can. Stop as soon as you know the envelope, the driver and the real capacity; confirm the rest as the table takes shape.

1. What number are we allocating, for what period, and in what currency?
2. Does that number include DevRel staff salaries, some of them, or none? (Roughly 60% of surveyed programs exclude salaries entirely - the answer changes every comparison downstream.)
3. Who signs it off, and when does the cycle lock?
4. What is already committed - signed contracts, annual licences, retainers - and for how long?
5. What DevRel work is paid from someone else's budget today (field marketing, engineering, product), and how stable is that?
6. Why does the company fund this work - adoption, sales enablement, developer enablement, product input, ecosystem, contributor community, employer brand? Which does the funder name first?
7. Do developers adopt individually with a card, or does a company decide? If both, which one is this period's priority?
8. By what date must a result be visible, and to whom?
9. Does this period owe a win inside it, or an asset that compounds past it? If both, which one is this period's job?
10. Who are the named people, and how many hours a week does each really have for devrel work after their other duties?
11. What is your ceiling beyond hours - headcount you can add, political capital for a public refusal, and how far past this period you may commit?
12. What did last period's money buy, and what result can you actually point at?
13. What is off-limits - no travel, no paid media, no agencies, a channel the founder refuses?
14. Did the budget go up, down or flat last cycle, and is a cut plausible this one?

Answers 8, 9, 11 and 13 re-rank the orderings in steps 3 and 4 before either is presented:

- **Q8, a date inside the period** - promotes rented reach and the renewal of a line that already clears its threshold, and deletes any line whose latency runs past the date: an outsourced content pipeline at roughly two months to first publication, a certification build at a quarter or more.
- **Q9, a compounding mandate** - does the reverse: promotes the compounding line above rented reach and holds it funded across two periods.
- **Q11, an effort ceiling below what the top-ranked line needs** - deletes that line rather than demoting it.
- **Q13, "No travel"** - deletes the field-and-events shape outright.

Say which answer moved which line when presenting the result.

If the user cannot answer question 1, that is the normal case rather than a failure: a third of surveyed practitioners cannot state their program's budget and another sixth have no set budget at all. Switch the deliverable to a short list of questions they must get answered - from whom, by when - and a provisional allocation expressed in percentages that survives whatever number comes back.

If your environment offers persistent memory, store the envelope, the split, each line's threshold and each review verdict. Budgets are annual and argumentative; the next session re-uses the verdicts instead of re-litigating them.

## Step 1 - Establish the real envelope

Sort every euro or dollar into four buckets before allocating anything:

- **Committed** - already signed. Not allocatable this cycle; only renewable or cancellable, and each has a notice date.
- **Discretionary** - the number actually being planned. This is what the rest of the workflow splits.
- **Borrowed** - devrel work paid from another team's line. Real capacity, foreign control, can disappear without notice. List it, mark the owner, never depend on it for a critical line.
- **In-kind** - engineer hours, exec time, product credits, donated infrastructure. No invoice, hard ceiling.

Then state the hours envelope alongside the cash: per named person, hours per week available for devrel work after support, meetings, and their other job. Total it for the period and subtract 20-30% before planning against it - support load, launches and incidents always arrive.

Write both totals at the top of the deliverable. A plan that never states its hours envelope will overspend it, and nobody will notice until the quarter fails while the cash is intact.

## Step 2 - Price each candidate line in cash and hours

Every line carries two numbers and five properties. Use [./references/line-item-cost-model.md](./references/line-item-cost-model.md) for the per-pillar checklist of what to count - it names the costs teams routinely forget.

The five properties to record for each line:

- **Setup cost before first output** - the one documented case, Snyk's agency content pipeline (Matt Jarvis, developerrelations.com case study, undated), took roughly six months of a director's time before producing at volume.
- **Latency** - that same pipeline ran 8-9 weeks from concept to publication, which disqualifies it for a dated launch beat.
- **Recurring maintenance** - the hours per month to keep the surface accurate forever, not just to build it.
- **Decay behaviour** - what neglect looks like: invisible (acceptable) or actively wrong, like a stale listing or a broken sandbox (only fund it if the maintenance hours are funded too).
- **Exit cost** - what happens to the surface when the contract, the sponsorship or the person ends.

Flag immediately any line whose hours exceed the hours envelope. It is unfunded regardless of its cash, and this is the most common way a budget fails while being fully spent.

When the budget excludes salaries but a line still needs a euro figure for its hours, use the organisation's own loaded hourly rate. Absent one, the reference file gives the widely circulated multiplier and says what it is worth.

## Step 3 - Brainstorm allocation shapes before choosing one

Say explicitly that you are in brainstorming mode, and do not write a table yet.

First decide how much of last period survives. One rule overrides the choice: **never inherit silently** - any line whose share nobody can explain is zero-based this cycle, whichever mode you pick. Then choose the mode:

- **Zero-base the discretionary spend** - every line restarts at zero, is justified on this period's driver, then ranked and funded down the list until the envelope runs out. Do this when the driver changed, the budget owner changed, or the envelope moved by more than a fifth. The method is Peter Pyhrr's zero-based budgeting (Texas Instruments, early 1970s); it ranks with step 4's ordering, and that one ranking is also what step 7 cuts from, so the work is done once.
- **Budget incrementally** - keep the shape, argue the deltas. Do this when the driver is unchanged and last period's lines cleared their thresholds; re-justifying an evidenced line costs hours and produces the same answer.

Treat committed spend as a renewal decision keyed to its notice date, not as a line to re-justify mid-contract.

Build **two or three whole-budget candidates**, not one - a budget is a portfolio, and a single proposal hides the trade-off being made. Use [./references/allocation-archetypes.md](./references/allocation-archetypes.md) for each shape's pillar weights, its starve list and the situations it fits.

Shortlist from this default ordering, not from the reference's page order. It ranks the shapes on effort: hours, setup latency before first output, exit cost - and never on cash.

Cash is the thing being allocated here, so using it as the denominator would rank the cheapest shape first and answer a question nobody asked. Magnitudes only, no figures.

- **effort**, heaviest first: `community-led > enablement-first == content-compounding > field-and-events > ecosystem-and-sponsorship`
- **value**, outcome bought read over two years: `content-compounding > enablement-first > community-led > field-and-events > ecosystem-and-sponsorship`
- **efficiency**, value per hour committed: `enablement-first > content-compounding > field-and-events > ecosystem-and-sponsorship > community-led`

The one tie: enablement-first and content-compounding cost the same effort. Both convert into a standing weekly shift that outlives the period - a maintenance shift for one, a review shift for the other - and both need about a quarter before output runs at volume. Field-and-events beats sponsorship on efficiency because an event at least leaves behind material the program keeps; a sponsorship's whole return sits inside a surface someone else controls.

Barbell is deliberately absent from that ordering. It is a concentration rule applied on top of whichever shape wins, not a pillar mix of its own, so ranking it against the others would be false precision. Apply it by default anyway: most devrel work only pays past a threshold of consistency, so two pillars funded properly beat four funded at a quarter each.

**What this order starves: community-led.** Its payoff is slow and uneven, its staffing is permanent, and its exit cost is the only unrecoverable one in the list - so a shape ranked on efficiency alone never funds it, and the program keeps renting reach forever. Promote it when peer-to-peer questions already happen somewhere without you, a named person owns the venue permanently, and the funded driver is contributor community or retention rather than acquisition.

Present each candidate with:

- The pillar it over-invests in.
- Its three largest lines.
- What it deliberately starves.
- The hours it needs.
- The earliest signal it is working.
- The conditions under which it is the wrong shape.

Recommend one and say why. Get the budget owner to pick a shape before you price anything in detail - a wrong shape propagates into every line below it.

Adoption motion moves the order, so name it before recommending:

- **Individual-developer adoption** - money follows discoverability and time-to-first-success: docs and tooling, content, community answers, low-cost distribution. Events matter mostly as content and credibility sources, and the default ordering above already assumes this motion.
- **Company/enterprise buying** - money follows evaluation and proof: reference architectures, security and compliance material, field events where approvers actually stand, education and certification for implementation partners. This motion promotes field-and-events to the top of both value and efficiency; nothing else moves.
- Most programs face both. Rank them for this period instead of splitting evenly, and mark the lines that serve both so they get funded once.

Every ordering in this skill - the shapes here, the line kinds in step 4, the breadth-versus-depth call in the cost model - is a default, not a law. Each shifts with context and with who executes it, so re-rank against what you already know about this program before presenting one:

- A signed multi-year conference contract is committed spend, not a ranked option - the decision is renew-or-notice at its notice date, and the shapes are ranked over what is left.
- A team that cannot travel deletes field-and-events. Say it is deleted and name the constraint; never present it as the last-ranked shape.
- A budget that must be spent this period or is lost promotes rented reach and demotes every shape whose first readable signal lands after the lock date.
- An asset the default assumes away - a reviewer faster than anyone else's, a venue the company already owns, a maintainer already on staff - promotes the shape that consumes it, because its effort column is already paid.

## Step 4 - Set a return threshold per line, before funding it

Give every line one outcome unit and one number it must clear to keep its slot next cycle. Pick the unit from the pillar's own vocabulary:

- Events - cost per qualified conversation.
- Content - cost per published-and-reviewed piece.
- Community - cost per answered question or active contributor.
- Tooling - cost per unblocked developer.

Derive each threshold from a baseline the program already has, never from an aspiration. Where no baseline exists, fund the line as an experiment with a fixed window and a "we will know X by date Y" statement instead of a number.

Then rank the lines before you fund them - the order you write here is the order step 7 cuts from. Never rank two pillars against each other by cost per unit: the units measure different things, and the pillar that is easiest to instrument would win every round regardless of value. Cost per unit stays legitimate in exactly two comparisons - the same pillar across periods, and the same pillar across options (this conference or that one, agency or in-house).

Rank on effort instead, in magnitudes: hours, setup latency, exit cost. The line kinds below encode the three inputs that decide a devrel line's worth:

- Fit to the funded driver.
- Strength of the evidence that it worked before.
- Whether it is foundational to the others.

- **effort**, heaviest first: `new compounding line > new driver-fit line > rented reach > foundational maintenance > proven renewal == experiment`
- **value**, outcome bought: `new compounding line > foundational maintenance > proven renewal > new driver-fit line > rented reach > experiment`
- **efficiency**: `foundational maintenance > proven renewal > new driver-fit line > rented reach > experiment > new compounding line`
- **compliance cost**, by the review it triggers and the reversibility it costs: `multi-period sponsorship or foundation membership > multi-year event contract > hosted community venue > paid distribution > every other line`, which commits nothing past the period and needs no sign-off outside the team

The one tie: a proven renewal and an experiment cost the same effort because neither carries a setup pass - one is already running, the other is scoped to a fixed window on purpose.

**What this order starves: the new compounding line.** It is the highest-value kind on the list and it loses every cycle, because its effort is the heaviest and its payoff lands two quarters out - which is how a program spends on fast-acting lines forever and stays permanently tactical.

Promote it above rented reach when any of these holds:

- The user answered "compounding asset" at question 9.
- The envelope is flat or growing and neither of the last two periods funded one.
- A named person's hours are ring-fenced for it.

Once promoted, fund it across two periods or not at all - a compounding line cancelled at its first review bought nothing.

A line the user's constraints rule out is deleted from this order, never ranked last. Name it as deleted with the constraint that removed it, in step 5's ruled-out list. A ruled-out line parked at the bottom reappears as scope the moment the envelope moves.

One cross-pillar comparison adjusts the order without replacing it, borrowed from paid-media planning: put each pillar's share of the outcome next to its share of the budget.

- Fund up where outcome share exceeds budget share and the audience is still under-served.
- Question any pillar whose budget share has outrun its outcome share for two periods.

Comparing two ratios sidesteps the incompatible units that make a single cost-per-unit ranking meaningless.

Before you agree to grow a line, check that its result can actually be read. Score the program 1-3 on each of:

- The outcome is instrumented at all.
- The instrumentation predates the spend.
- Someone owns the number.
- There is an agreed attribution story for it.

Below half marks, hold the line flat and spend the increment on visibility first - a line nobody can read gets renewed on politics, whichever way its threshold went. The score and its cut-off are self-set, adapted from a paid-media measurement-maturity gate.

Present every threshold as repeatable rather than statistically rigorous - the framing paid-media practice uses for its own kill rules. Its job is to make the renew-or-drop conversation cheap and unemotional, not to prove causation; a threshold defended as proof invites a debate about attribution that DevRel measurement cannot win.

## Step 5 - Reserve, horizons and the unfunded list

- Hold **10-20% of discretionary spend unallocated** (self-set, but adjacent marketing-planning practice lands in the same band: a 10-15% contingency plus a testing reserve). The best opportunity of the year is rarely visible when the budget is written, and without a reserve every reactive request cannibalises a committed line.
- Split the allocated remainder across horizons: the bulk on compounding work already known to pay, a middle slice on scaling the one thing that is working, a small slice on genuine experiments that each carry a kill date. Any specific ratio proposed (70/20/10 is the familiar shape from marketing portfolio practice) is a proposal to argue with, not a devrel benchmark - say so when using it.
- Keep an explicit **unfunded list**: what was considered, ranked, and lost on the ranking, with one line on what would have to change to fund it. It stops rejected ideas from returning every month and gives the next cycle a ready shortlist.
- Keep a separate **ruled-out list**: lines a stated constraint removed - no travel, no agencies, a commitment past the period the user cannot make, an effort ceiling the line breaches. Name the constraint next to each. These were never ranked and do not return when the envelope moves; only the constraint lifting brings one back.

## Step 6 - Write the allocation

Produce one table plus the rules under it:

```markdown
| Line                       | Pillar  | Cash   | Hours/period | Owner        | Buys                   | Threshold                           | Review             | If it misses                 |
| -------------------------- | ------- | ------ | ------------ | ------------ | ---------------------- | ----------------------------------- | ------------------ | ---------------------------- |
| Flagship conference booth  | Events  | 28,000 | 240          | A. Rivera    | 3 events, 2 staff each | ≤ 450 / qualified conversation      | 2 weeks post-event | Renegotiate to workshop tier |
| Technical content pipeline | Content | 36,000 | 180          | S. Okafor    | 24 reviewed pieces     | ≥ 18 published, ≥ 500 sessions each | quarterly          | Cut agency, halve volume     |
| Reserve                    | -       | 12,000 | -            | budget owner | unallocated            | -                                   | quarterly          | -                            |
```

Close the deliverable with:

- Totals against both envelopes.
- The horizon split.
- The reserve.
- The unfunded list.
- The ruled-out list.
- The review clocks.
- The reallocation triggers.
- The cut list.
- A five-sentence narrative the funder can repeat without you.

A full worked allocation, with a weak and a strong version of each section, is in [./references/allocation-example.md](./references/allocation-example.md).

Present it one section at a time and get agreement before moving on. Catching a wrong envelope at the first section costs a paragraph; catching it at the last costs the document.

## Step 7 - Set the clocks, the triggers and the cut list

Run three clocks, deliberately at different rhythms:

| Clock        | Rhythm                                           | Question                                                       |
| ------------ | ------------------------------------------------ | -------------------------------------------------------------- |
| Burn         | monthly                                          | Is cash going out at the planned rate, and are the hours real? |
| Line verdict | quarterly, or one outcome window after the spend | Did this line clear the threshold set before it was funded?    |
| Envelope     | annual cycle                                     | What is next period's ask, and what evidence supports it?      |

Write the reallocation triggers before the period starts, so a mid-year move is a rule firing rather than an argument:

- A line misses its threshold twice.
- The funded driver or its owner changes.
- A committed contract reaches its notice date.
- The person a line depends on leaves.
- An unplanned opportunity clears a higher bar than the weakest funded line.

Then write the **cut list**: step 4's efficiency order read from the bottom, applied to a 20% and a 40% reduction, each removal carrying its consequence in one sentence. Two overrides apply to that order:

- A foundational surface whose neglect is actively wrong (a stale doc, a broken sample, an unanswered venue) is cut last however it ranks.
- A compounding line promoted for two periods is not the first thing dropped, or the promotion was theatre.

Ruled-out lines never appear here; they left the plan before it was funded.

Roughly a quarter of surveyed programs lost budget in a single year, so this is a normal planning case. A team that volunteers a defensible order keeps control of the shape; a team that does not gets a flat percentage applied to every line, which starves the compounding work first.

Judge a first-year line against awareness-stage expectations, since recognition and relationships accumulate across editions. Treat a second consecutive miss as a drop signal anyway - compounding is not a licence to renew forever.

## Invocation examples

Match the ask to an entry point, and say which one you took.

| What the user says                                                 | Where to enter         | What comes back                                                                                                                                              |
| ------------------------------------------------------------------ | ---------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| "We have 180K for devrel next year, help me split it."             | Full workflow, step 1  | Envelope, two or three shapes, then the table and its rules                                                                                                  |
| "Is 40% on events too much?"                                       | Step 3, one shape only | The survey figure with its caveats, then the driver-and-motion test that decides it for this program                                                         |
| "Finance wants 20% out of devrel by Friday."                       | Step 7, cut list first | Ordered removals with a consequence each, and what must be protected                                                                                         |
| "Nobody will tell me the number."                                  | Interview only         | The questions, their owners, a date, and a percentage-based provisional split                                                                                |
| "Was the conference worth it?" / "Should we even run a community?" | Route out              | One event's return → `samber/developer-relations-skills@developer-event-sponsorship`; pillar selection → `samber/developer-relations-skills@devrel-strategy` |

Whatever the entry point, the artefact is one markdown document with these parts:

- Envelope (cash and hours, four money buckets).
- The line table.
- Horizon split.
- Reserve.
- Unfunded list.
- Ruled-out list.
- Review clocks.
- Reallocation triggers.
- Ordered cut list.
- A five-sentence funder narrative.

Sections the user's question does not touch can be one line each, but never silently dropped - a missing hours envelope or a missing cut list is exactly what makes a budget indefensible later.

## Quality gate

Hold the allocation to this bar and iterate until it passes. Report the check explicitly at the end.

- The envelope states whether salaries are inside it, and separates committed, discretionary, borrowed and in-kind.
- Total planned hours sit at or under the hours envelope after the 20-30% deduction; no line exceeds its owner's capacity.
- Every funded line names an owner, what it buys, one threshold, one review date and one miss action.
- Every threshold traces to a baseline, or the line is labelled an experiment with a kill date.
- The reserve is at least 10% of discretionary spend and is not quietly pre-committed.
- No pillar is compared to another pillar by cost per unit anywhere in the document, and no ranking uses a cash figure as its denominator.
- Every ordering names its axes, and every `==` carries the sentence that makes the two genuinely equal.
- The starved option is named at both levels - shape and line - each with the condition that promotes it.
- Lines a constraint removed appear as ruled out with the constraint named, never as the last row of a ranking.
- At most two pillars are funded above token level, unless the program is at scale stage with the capacity to prove otherwise.
- The unfunded list and the cut list are both non-empty, and the cut list is ordered.
- The narrative fits in five sentences and contains no devrel jargon.
- Every percentage and threshold in the document is tagged sourced, program baseline or self-set, and no self-set number is described as a standard.

An allocation that fails the hours line does not have a discipline problem. It has one line too many.

## Common failure modes

- **Allocating cash only.** The plan looks funded and the quarter still fails, because the hours were never counted. Price both.
- **Quoting the survey split as a target.** The published events/salaries/content/marketing shares come from 47 respondents on inconsistent bases and omit community, tooling and education entirely. Use them to show that a departure is deliberate, never as a benchmark.
- **Ranking pillars by ROI, or by cash per outcome.** Whichever pillar is easiest to instrument wins, and the program defunds the work that was carrying it. Rank on hours, latency and exit cost; cash is what the plan allocates, never what it divides by.
- **Parking a ruled-out line at the bottom of the ranking.** It reads as merely unaffordable, so it comes back as scope the next time the envelope moves. Delete it and name the constraint.
- **Spreading thin to keep everyone happy.** Four pillars at 25% clear no threshold. Concentrate and name what is starved.
- **No reserve.** The first unplanned opportunity eats a funded line, and the plan stops describing reality by month three.
- **Renewing by default.** A line that has missed twice is renewed because cancelling is awkward. That is what the written trigger exists to prevent.
- **Thresholds invented after the fact.** A number chosen once the result is known is a narrative, not a threshold.
- **Treating borrowed budget as owned.** The other team re-plans, and a core line vanishes mid-period.
- **Only a total for the cut scenario.** "We could absorb 20%" without an ordered list hands the ordering to someone who does not know what compounds.
- **Precision without a stated origin.** A "1.35× loaded-cost multiplier" or a "30/40/20/10 split" reads as authority precisely because it is specific, and gets quoted back for years after the guess behind it is forgotten. Say where each figure came from, or drop it.
- **Funding a line nobody can read.** Growing spend on an outcome that is not instrumented buys a renewal argument, not a result. Fix the visibility with the increment instead.

## References

- samber/developer-relations-skills@devrel-strategy for the program shape this budget funds.
- samber/developer-relations-skills@devrel-metrics for the measurement framework the thresholds borrow from.
- samber/developer-relations-skills@developer-event-sponsorship for splitting the event line.
- samber/developer-relations-skills@devrel-content-calendar for the content line's throughput.
- samber/developer-relations-skills@developer-community-launch for the community line's readiness gates.
- samber/developer-relations-skills@devrel-team-structure for who owns which budget.