Budgets, roadmaps and making the case

1. Step 01

   Build a cost model you can explain

   Finance will model the program whether or not you do. Build the model first, so the drivers are yours. Split the spend into a few lines, and for each one name what drives it and what you can change:

   Cost	What drives it	Levers you control
   In-house people: reviewers, investigators, policy, data and engineering	Volume times handle time, divided by productive hours; hours of coverage, including round-the-clock cover for the most severe harms; the number of languages	Detection and routing that send cases to the right queue, friction that prevents harm before anyone reviews it, and clearer guidance that cuts handle time
   Vendors: review staffing, AI moderation, age assurance, translation	Price per decision or per hour, minimum commitments, surge terms, and wellbeing and quality standards	Contract terms, price adjusted for quality, and consolidating overlapping vendors
   Tools and infrastructure	Classifiers, hash matching, case management and compute	Free and open-source tools, and building only what's specific to you
   Compliance	Risk assessments, transparency reports, audits and legal advice	Logging that turns a report into a query rather than a project (chapter 17)

   The unit that ties the lines together is cost per decision: the fully loaded cost of review divided by the decisions made, by queue and vendor. Never show it alone. Optimizing it on its own lowers quality, so put it next to QA agreement on every slide. In the metrics framework's worked example, one vendor costs $0.38 a decision against another's $0.52, but agrees with expert reviewers 8 points less often, and rework and appeals make the cheaper vendor the more expensive choice.

   Capacity is the biggest driver, and it's forecastable from volume trends (chapter 7). Before you ask for headcount, check the levers that reduce volume or handle time. Free tools such as those from ROOST can cover hash matching and review workflow without a license fee (chapter 9).

   By stage: an early team divides each vendor invoice by the decisions it covered, and tracks people and tools in a spreadsheet. A growing team builds the model with Finance and forecasts on Finance's planning cycle, quarterly for example. At scale, every queue and vendor has a cost per decision, a quality number and a forecast, owned by T&S operations with Finance.

2. Step 02

   Stop being judged by throughput

   If your team took down 40% more harmful content last quarter than the quarter before, is that good news? You don't know yet. It could mean detection improved, harm grew, or automation got more aggressive and legitimate users are paying for it. Volume can't tell you which, and a budget argued on volume invites one response: do the same volume for less.

   Keep these off the executive slide: total items removed, reports received, automation rate on its own, moderator headcount and accounts banned. Each can rise for good or bad reasons, and none says what users experience. Automation rate is an efficiency metric, not a measure of maturity. It counts the decisions people didn't make, not whether they were right.

   Lead with the outcome questions instead:

   • Is exposure to severe harm going down?
   • After enforcement, do bad actors stop, or come back on new accounts?
   • Are we catching high-risk behavior earlier?
   • What are appeals telling us about where our policy or automation is wrong?

   Then frame the budget the same way. "Child safety: 12 reviewers" is a cost line. "Child safety: unsafe contact per 10,000 young users, down from last quarter, and what it takes to keep it falling" is an outcome with a price. Chapter 11 covers the metrics.

   Report operational and outcome numbers side by side, and say so when they disagree. It takes discipline to walk into a business review and say, "Our numbers are up, and I'm not convinced we're safer." That's the conversation that earns credibility, directs investment to the right places and keeps Trust & Safety from being judged like a cost center measured by throughput.

3. Step 03

   Make the revenue case with a matched cohort

   The cut that works is a matched cohort. Take new players whose early sessions included an actioned incident, and compare their 7- and 30-day return against new players with clean sessions, matched on playtime, mode, region and platform. Without the matching, the analysis measures engagement instead of harm. Chapter 11 and the churn after toxic exposure page cover the method. Outside games, match on what drives exposure on your platform, such as audience size on a social app.

   A rough comparison, such as people who reported harassment against everyone else, can tell you where to look. A budget case needs the matched version.

   Then turn the gap into money, using Finance's numbers rather than your own:

   1. Users lost. The retention gap times the number of exposed new users each month.
   2. Revenue at risk. Users lost times the value of a retained user, from the same figures Finance reports to the board.
   3. Skipped spending. The difference in spend between exposed users and their matched group over the same window.

   Here's an illustration of the arithmetic, from the metrics framework's worked example: exposed players retain at 52% after 30 days against 61% for matched players. A 9-point gap across 20,000 exposed players a month is about 1,800 players a month, and Finance can put a value on each one.

   Be honest about what it is. Present the result as a range, and call it an association unless you've run a proper causal analysis. With children's data, involve your privacy team and report only aggregates. An overclaimed number is the fastest way to lose the room the second time.

   Numbers worth tracking:

   • Share of new players whose first five matches include an actioned incident
   • D7 and D30 return for exposed new players against the matched group
   • Voice chat opt-out rate in a player's first week

   Safety exposure belongs as a standard cut on the retention dashboard, owned jointly by T&S and the data team and reviewed alongside every other churn driver. Once it's there, the revenue case is updated every month without anyone having to argue for it.

4. Step 04

   Count the players a top spender drives away

   The same blind spot affects enforcement on high-value users. Banning a top spender hits the revenue report the next day. The players they drove away stay invisible unless someone builds the cohort.

   So build it before the argument starts:

   • When a high-value account is up for enforcement, pull the users it targeted in confirmed incidents over a set window.
   • Compare their return and spend afterwards against matched users who weren't targeted.
   • Put both numbers in the decision record: the account's own revenue, and the return and spend gap among the people it targeted.

   The cohort completes the revenue argument. It doesn't change the rule. Enforcement follows the same policy for every account, because consistency means the same penalty rules for everyone. If someone wants an exception for a spender, the question goes to the executive who owns safety risk, with both numbers in front of them.

5. Step 05

   Build a roadmap leadership can follow

   Leadership says yes to plans it can check. The program maturity model gives you the structure: eight areas (policy, detection and prevention, review operations, quality and appeals, crisis response, metrics and reporting, regulatory readiness and team wellbeing), each rated from 1 (reactive) to 5 (leading).

   The model suggests a target for each stage:

   Stage	Target
   Early	Level 2, and level 3 for crisis response, regulatory readiness and team wellbeing
   Growing	Level 3 in every area
   At scale or regulated	Level 4 in every area

   Crisis response, compliance and wellbeing stay at level 3 even for early teams, because the harm is the same whatever your size.

   Build the roadmap from the gaps, biggest first. When two gaps are the same size, one order that works is crisis response, then regulatory readiness, then detection and prevention, though your own risks may argue for another. Each level has two concrete steps to the next level, so every line on the roadmap is something specific.

   Then check the roadmap against risk. The harm coverage radar compares each harm area's risk with its coverage across policy, detection, enforcement, appeals and measurement. Without that view, budget follows whoever argues loudest rather than where the exposure is.

   Give each line the same six things: the gap, the step, the owner, the cost, the risk it reduces (from the abuse pre-mortem or incident data) and the metric that will show it worked. Legal deadlines go on with their fixed dates (chapter 16). Keep it short, for example one page in three columns: now, next and later. Take a snapshot of the maturity ratings every quarter, so progress is visible without a long explanation.

6. Step 06

   Run the quarterly executive review

   A metric only changes behavior when a named meeting looks at it on a schedule. A quarterly executive review, or whatever cadence matches your planning cycle, brings together executives, Legal and Finance to look at outcomes, risk and investment. Send a pre-read, and spend the meeting on decisions.

   Part	What's in it
   Outcomes	North-star metrics against last quarter and the target band, with ranges
   Where the numbers disagree	Operational metrics next to outcomes, and what you think explains the difference
   Retention and revenue	The exposure cohort: retention gap, revenue at risk and the trend
   Risk	Top risks, accepted risks and what happened to them, and legal duties coming due
   Roadmap	What shipped, what slipped and why, and the maturity snapshot
   Asks and decisions	Each ask with the risk it reduces, its cost and the metric that will show it worked

   Set targets carefully. Measure long enough to see the normal range before you commit to a number, for example 6 to 8 weeks. Set a band, not a point, so the gap between "on track" and "off track" is your early warning. Pair every target with a guardrail, and show the uncertainty on anything from a sample.

   Close the loop on past asks. Each quarter, show what last year's investments promised and what they delivered, including the ones that didn't work. Nothing earns the next yes like a record of reporting honestly on the last one.

7. Step 07

   Plan the cuts before you need them

   Tight years come. A team that has already decided what it would cut, and what it never would, makes better choices than one deciding under a deadline. The first column below is an example to adapt to your own risks.

   Cut first	Never cut
   Response-time targets on low-severity queues: relax them, don't abandon them	Child safety and severe-harm escalation, including round-the-clock cover for the worst harms (chapter 12)
   Overlapping tools and vendors, and minimum commitments you don't use	Duties the law requires, such as child sexual abuse reporting, notices and required risk assessments (chapter 16)
   Full reviews of low-risk launches, replaced with a short checklist	Reviewer wellbeing: exposure limits and specialist support (chapter 14)
   Manual work that automation can take over, only where its overturn rate is at or below human review	Quality sampling and prevalence measurement, because without them you can't see what the other cuts did
   Reports and dashboards nobody uses	Appeals, because they're how you find out a cut went too far

   Every cut accepts a risk, so treat it like any accepted risk. Write down what it saves, what risk it accepts, the metric to watch and a date to revisit. The person who owns the product outcome signs off, and the most serious risks go to the executive who owns safety risk. If the metric moves past an agreed limit, reverse the cut. A cut that nobody measured is a risk nobody accepted.

1. 01

   Defending the budget with volume

   Lead with outcomes and risk, and keep volume in the operations review.

2. 02

   Showing raw retention data

   Build the matched cohort first, or harassment will look like it improves retention.

3. 03

   Claiming causation from a correlation

   Present a range, and call it an association unless you've run a causal analysis.

4. 04

   Optimizing cost per decision alone

   Show it next to QA agreement, and count rework and appeals.

5. 05

   Asking for headcount without a roadmap

   Tie each ask to a gap, a risk and a metric.

6. 06

   Letting a spender's revenue decide enforcement

   Apply the same rules to everyone, with the cohort of the people they drove away in the record.

7. 07

   Cutting what lets you see

   Measurement and quality sampling go last, because they show what every other cut did.

From Steven's writing · 2 posts

1. Oct 1, 2026Essay

   Harassed players look like your best-retained users

   Raw retention data hides the cost of toxicity because harassed players are the most engaged. A matched cohort shows it, and safety exposure belongs on the retention dashboard.

   Also in: 1. What Trust & Safety is for, 11. Measuring what matters

2. Sep 23, 2026Essay

   Activity is easy to measure. Impact is harder.

   A 40% jump in removals could mean better detection, more harm or over-enforcement. Report outcome metrics next to operational ones, and tell leadership when they disagree.

   Also in: 1. What Trust & Safety is for, 11. Measuring what matters

Outside sources

• Take This: Toxic Gamers Are Alienating Your Core Demographicthe 2023 report, with Nielsen data, on how toxicity changes players' spending and play.
• TSPA: Setting up content moderation teamsthe Trust & Safety Professional Association on workforce models, capacity planning and cost.
• Digital Trust & Safety Partnership: Best Practices Frameworkfive industry commitments, including assessing whether your approach to risk is working.
• ROOSTa nonprofit building free, open-source safety tools, including the Osprey rules engine and the Coop review console.

1. Oct 4, 2026
   RevisedWith 18 other chapters

   Practical advice that varies by platform, such as cadences, sample sizes, targets and who owns what, is now set out as options with examples, so each team can choose what fits.

2. Oct 3, 2026
   DraftedWith 17 other chapters

   First full drafts of the other 18 chapters, built on Steven's posts, the handbook's principles and the Workbench's open content, with every legal and factual claim checked against its source. Stories from Steven's own work come next.

3. Oct 3, 2026
   AddedWith 14 other chapters

   Linked the first 16 posts to the chapters they inform, and set out the ten principles behind the handbook.