Appendix B: What a Privacy Program Actually Costs

Data Privacy Guide: Overview | Part 1 | Part 2 | Part 3 | Part 4 | Part 5 | Part 6 | Part 7 | Part 8 | Part 9 | Part 10 | Appendix B | Appendix C

Why This Appendix Exists

The ten-part series gave a CPO the framework, the regulatory map, the implementation walkthrough, and the technical depth on PETs. It said almost nothing about money. That was a deliberate omission while drafting and a real gap by the time the series went up. Two of this appendix’s own baseline numbers, the largest CCPA settlement and the IBM breach average, were already superseded by the time this piece went live on September 1, 2026. That is not a flaw in the argument. It is the argument: privacy costs move faster than any single appendix can track, which is why the section below records what changed and when.

Cost conversations are where privacy programs stall. The CFO does not push back on whether GDPR matters. The CFO pushes back on the headcount request, the OneTrust renewal, the consultant invoice, and the line item for “PET evaluation” with no benchmark to anchor it. Without numbers, the privacy team loses every internal negotiation and ends up resourced for compliance theater rather than actual protection.

This appendix builds the model from the bottom up. The cost of doing nothing first, because that sets the upper bound on what protection is worth. Then the people, the platforms, the PETs, and the ROI evidence. Every figure in this article comes from a primary survey, a peer-reviewed paper, a regulator announcement, or a procurement database. Vendor marketing claims are flagged when used and treated as ceilings, not midpoints.

The Cost of Doing Nothing

Privacy budgets are easier to defend when the alternative is priced. The data has improved enough in the last two years to do this honestly.

Regulatory Fines as a Baseline

Cumulative GDPR fines have crossed EUR 7.1 billion since the regulation took effect in 2018. 2025 alone added EUR 1.145 billion across approximately 2,161 fines. The CMS GDPR Enforcement Tracker records 2,685 documented fines as of March 2026, up 440 from its 2025 report, and the trend is no longer concentrated on Big Tech. France led on total fine value in 2025 at EUR 486.8 million, Germany on volume with 499 cases, and Spain landed second by value at EUR 45.2 million.

The headline fines remain instructive. Meta’s EUR 1.2 billion penalty for unlawful EU-to-US data transfers, Amazon’s EUR 746 million for advertising-related processing, and TikTok’s EUR 530 million for transfers to China stand as the upper end. They are also the cases that fund every privacy budget request in 2026.

US enforcement has caught up faster than most boards realize. The California Privacy Protection Agency raised penalty caps effective January 1, 2025 to $2,663 per violation and $7,988 per intentional violation, with consumer damages of $107 to $799 per incident. Disney’s $2.75 million CCPA settlement in February 2026 was the largest CCPA settlement at the time, since surpassed (see the update below), but Tractor Supply ($1.35M, October 2025), an unnamed health publisher ($1.55M), and Todd Snyder ($345K) sit in a tier any mid-market company can land in by missing a cookie banner configuration.

Enforcement venue	2025 baseline	What it means for a budget
GDPR (EU collectively)	EUR 1.145B in fines, 2,161 actions	Penalties expanding beyond Big Tech to mid-market
Largest single GDPR fine to date	EUR 1.2B (Meta, 2023, transfers)	Cross-border transfer documentation is not optional
California CCPA / CPRA	$2,663 to $7,988 per violation	A single misconfigured opt-out can cascade across millions of records
Largest CCPA penalty to date	$12.75M (GM, May 2026)	Six- and seven-figure, now eight-figure, US privacy penalties are routine
European DPA breach notifications	443 per day, up 22% YoY	Volume is growing, not slowing

Breach Costs as the Other Baseline

The IBM Cost of a Data Breach Report is the only annual study large enough to anchor breach modeling. The 2025 edition puts the global average at $4.44 million per breach, the first decline in five years. The US average is $10.22 million, up 9% year over year. Healthcare leads at $7.42 million per breach. The average breach lifecycle dropped to 241 days, the lowest in a decade, and organizations using AI extensively in security operations cut that lifecycle by 80 days and saved approximately $1.9 million.

The per-record figures matter for incident response budgeting. Customer PII costs $160 per record globally, $264 in the US, $189 in the EU. 53% of breaches involve stolen or exposed customer PII, making it the most-compromised data type. Breaches disclosed by attackers (extortion-driven) cost $5.08 million versus $4.18 million when detected internally. The 2025 report also flagged a new line item: shadow AI usage adds $670,000 to average breach cost, a number that did not exist in any prior survey.

What this looks like in practice. The right framing for a CFO is not “what could a fine be” but “what does a breach at our scale actually cost.” A mid-market US company processing one million customer records that suffers a reportable breach exposing 5 to 15 percent of holdings, at $264 per record (the US average across IBM’s full breach dataset, not a mega-breach figure), is looking at a conservative $13M to $40M in exposed-record cost. A $2M annual privacy program is cheap insurance against that range.

Privacy Team Staffing Costs

Headcount is the largest line item in any privacy budget and the easiest one to underestimate when starting from a single hire.

Compensation Benchmarks

The IAPP’s 2025-26 Privacy Compensation Survey drew responses from 1,600+ professionals across 60+ countries. Half of all respondents working across privacy and AI governance earn more than $169,700 in total compensation. Nearly 7 in 10 received a bonus (72% in North America). IAPP certifications correlate with higher salaries and multiple certifications drive further increases.

The role-by-role picture below pulls from the salary aggregators that publish raw bands. Treat these as US benchmarks; European DPO compensation runs lower.

Role	Average / typical band	Source
Chief Privacy Officer	$280,000 to $450,000+ base, significant LTIP at public companies	Leonid Group, 2025
Chief Privacy Officer (alt)	$354,423 average	ZipRecruiter, March 2026
Privacy Counsel	$137,473 average, $137,473 median, $124,829 to $144,012 range (middle 50%)	Salary.com, August 2026
Data Protection Officer	$137,022 Glassdoor / $172,933 Salary.com	2025-26
Privacy Engineer	$172,554 average, $139K to $215K range	ZipRecruiter, December 2025
Google Privacy Engineer	$170,399 average, $138K to $212K range	Glassdoor, 2025

Date-stamp aggregator figures like these before using them in a budget defense. The Salary.com Privacy Counsel band above is dated August 2026; the live page moves on its own schedule and can drift materially between when you cite it and when a CFO checks it. An earlier version of this table cited a January 2026 snapshot at $212,457 average, a figure the live page no longer supports as of the August 2026 check: a concrete example of the drift this paragraph warns about.

Team Size Benchmarks

The IAPP-EY Governance survey put the average privacy team at 10 full-time staffers in 2021, a benchmark that predates the contraction described below. Composition varies sharply by region: roughly 70% of European organizations have at least one DPO and average 3 to 4 full-time DPOs each, while only 40% of North American organizations have a DPO and average less than one full-time DPO.

Privacy team sizes are also under pressure. The ISACA State of Privacy 2026 survey found median team size dropped from 8 in 2025 to 5 in 2026, and 43% of respondents report their privacy budget is underfunded. That same survey notes that contract-based hiring has surged: 60%+ of 2025 privacy job offers were contract roles, reflecting both flexibility trends and budget compression.

A Reasonable Year-One Headcount Plan

For a US mid-market B2B SaaS company between 500 and 2,000 employees handling EU and US customer data, a credible starting team looks like this. The salary numbers are loaded (base + bonus + benefits, roughly 1.3x base).

Role	Year 1 fully loaded	Notes
Privacy Counsel or fractional CPO	$275,000	Often a partner-level outside counsel for the first year, then in-housed
Privacy Engineer	$225,000	Builds DSAR automation, data discovery, consent infrastructure
Privacy Program Manager	$180,000	Owns the privacy registry, runs PIAs, coordinates with Legal and Security
GRC analyst (shared with Security)	$90,000 (50% allocated)	Manages vendor reviews, audit evidence, sub-processor registry
Year 1 staffing subtotal	~$770,000	Excludes training, certifications, conference attendance

A Fortune 500 organization handling regulated data routinely runs 15 to 30 privacy professionals across regions, plus embedded privacy champions in product teams (covered in Appendix C). That puts steady-state staffing alone in the $3M to $7M range before any technology spend.

Privacy Technology Costs

Software is where the build vs buy debate gets the most heated and where vendor-published ROI claims most distort the conversation.

Platform Pricing Reality

The cleanest pricing data comes from Vendr, a procurement platform that publishes anonymized transaction medians from actual purchases.

Vendor	Median annual contract	Range	Source
OneTrust	$11,500	$1,620 to $42,534	Vendr, 325 purchases
TrustArc	$22,000	$10,000 to $137,000	Vendr
TrustArc vendor portfolio monitoring	$30,000 to $70,000 (50 to 200 vendors)	up to $150K+ for larger portfolios	Vendr
BigID	Custom enterprise pricing, generally peer to OneTrust	mid-to-low six figures	Enzuzo comparison
Securiti AI	Custom; often 20 to 30% more competitive in competitive deals	Acquired by Veeam for $1.73B (Oct 2025)	Industry reporting

The Vendr medians look modest because they capture the distribution. Enterprise OneTrust deployments are routinely in the mid-to-high six figures annually, and implementation fees add 20 to 40% of annual subscription, typically $200K to $250K for external resources. The Forrester TEI study commissioned by OneTrust models a composite implementation of roughly 9 months and 3 internal FTEs.

Speaking of TEI studies: the OneTrust Forrester analysis claims 227% ROI over three years, $6.9M in benefits against $2.1M in costs, payback in under 7 months. Vendor-commissioned TEI studies are useful for directional analysis and dangerous when treated as planning numbers. The TrustArc TEI similarly claims 126% ROI and a $2.08M total benefit. Both numbers are plausible upper bounds. Neither is what you should put in a budget defense.

DSAR Cost Per Request

Data Subject Access Requests are the operational tax that compounds invisibly until you measure it. Gartner survey data shows the average manual DSAR costs $1,400 to $1,524 and consumes 8 to 12 hours of skilled labor per request at $50 to $250 per hour. DSAR volume jumped 246% over two years per 2024 survey data and 72% since 2021. A company processing 1,000 DSARs annually faces $1.4M to $1.5M in fully-loaded operational cost before any platform is purchased.

Build vs Buy

Vendor analyses estimate that building in-house typically costs 2 to 20x more than buying, and that custom builds take 12 to 24 months to launch versus weeks for vendor platforms. These figures originate from vendor marketing and skew high. The directional claim is consistent with what teams report in practice: orchestrating consent across 11 data systems runs roughly $200K, DSAR processing automation for high-volume workloads $1M+, and data discovery and classification for 24 projects per year $400K+.

For practitioners: The right build-vs-buy decision is rarely “build everything” or “buy everything.” Buy the consent management platform, the privacy registry, and the DSAR workflow because they are commoditized and high-blast-radius if mishandled. Build the integrations into your specific data warehouse, the AI-specific PIA workflow, and the engineering-side automations because no vendor knows your stack. Treat the Skyflow 2-to-20x figure as a sanity check, not a procurement input.

PET Implementation Costs

Parts 7 through 9 of the series explained how PETs work. None of them quoted the compute and personnel cost. This section closes that gap.

Differential Privacy

The peer-reviewed cost picture for DP is clearer than the marketing picture. Federated learning with DP runs roughly 13.1 hours of training time versus 10.4 hours for centralized training without DP, a 26% overhead. Newer techniques like RanN-DP-SGD reduce noise injections by approximately 50% while maintaining performance, with average epoch time of 110.3 seconds versus 109 seconds for standard SGD.

Apple and Google operate DP at hundreds of millions of devices: Apple uses local DP for emoji usage, Safari queries, and HealthKit; Google uses RAPPOR for Chrome telemetry and DP plus federated learning for Gboard. Neither publishes infrastructure cost. For LLMs specifically, ACM Computing Surveys reports that applying DP-SGD to large language models can lead to deficient accuracy and unreasonably high computational and memory overhead, a polite way of saying it does not yet work at frontier scale.

The honest answer for most enterprises in 2026: DP is feasible for telemetry, analytics, and federated training of mid-sized models. It is not yet feasible for fine-tuning frontier LLMs without significant utility loss.

Homomorphic Encryption

FHE remains the highest-overhead PET in operational use. Performance overhead lands in the 1,000x to 1,000,000x slower range versus equivalent plaintext operations, with 10,000x commonly cited as a midpoint. A healthcare benchmark on a single 1KB record reports 2.5 seconds to encrypt and 4.8 seconds per homomorphic computation, with ciphertext expansion of approximately 5x. AI inference under HE runs up to 202.5x slower than standard inference even with 90% model pruning.

GPU acceleration is improving the picture: the MNEMOS framework provides GPU acceleration for TFHE, and Libra enables cross-scheme optimization. Production deployments outside Apple’s PSI implementation remain rare.

Tokenization

Tokenization is the PET with the clearest ROI math. It can reduce PCI compliance scope by up to 90% of time and effort, shifting from SAQ-D to SAQ-A. In-house tokenization infrastructure runs hundreds of thousands of dollars and months to implement; third-party providers launch in minutes. The ROI formula a CFO will accept is straightforward: baseline annual control cost minus tokenization fees minus saved headcount hours minus reduced audit scope minus avoided incident cost.

Confidential Computing

TEEs are the lowest-overhead PET in production use. Typical overhead is approximately 10% for general workloads. NestedSGX reports under 2% for compute-intensive tasks and under 15.68% for I/O-intensive tasks. The cost is mostly hardware: confidential-computing instances on AWS, Azure, and GCP carry a modest premium over standard instance pricing, in the low tens of percent.

Federated Learning and Synthetic Data

The federated learning market is small but growing: $192M in 2025, projected to $562M by 2032 at 16.5% CAGR. The honest figure that should temper enthusiasm: only 5.2% of FL research has reached real-world deployment. Most FL projects in 2026 are still pilots.

Synthetic data is larger and growing faster: $604M in 2025, projected to $7B by 2033 at 31 to 38% CAGR depending on source. Vendor claims of 70% data cost reduction by 2026 are directional and need verification per use case.

PET Cost Summary

PET	Compute overhead	Production maturity	Best fit
Tokenization	Negligible	Mature, commoditized	Payment data, PCI scope reduction
Confidential Computing (TEEs)	~10% (compute-bound), up to 15% (I/O-bound)	Mature, GA on all major clouds	Restricted-tier processing on multi-tenant infrastructure
Differential Privacy	~26% for FL+DP; deficient for frontier LLMs	Mature for analytics, immature for LLMs	Telemetry, federated training, census-scale aggregates
Federated Learning	Variable; coordination overhead dominates	Pilot stage, 5.2% in production	Cross-silo medical, cross-bank fraud
Synthetic Data	Generation cost only	Mature for tabular, immature for high-fidelity behavioral	Test environments, ML training where privacy of source matters
Homomorphic Encryption	1,000x to 1,000,000x	Pre-production for general workloads	PSI, narrow inference workloads, niche regulatory contexts

The ROI Counter-Argument

Cost modeling skews defensive if it stops at expense. The Cisco Privacy Benchmark Studies are the only multi-year longitudinal data with sample sizes large enough to anchor the ROI conversation.

The 2025 study reports that 96% of organizations confirm privacy ROI exceeds cost, with median ROI of 1.6x on average annual spending of $2.7 million. 29% of organizations report returns of 2x or higher. The headline benefits cited: enhanced customer loyalty (79%), improved operational efficiency (78%), increased innovation (78%), reduced security losses (76%). 86% of organizations report positive impact from privacy legislation, up from 80% the prior year.

The 2026 follow-up shows the spending trend that matters most for budget defense: the share of organizations spending $5M or more on privacy rose from 14% in 2024 to 38% in 2026, a 171% increase over two years. 93% plan to allocate more resources over the next two years. The study surveyed 5,200+ professionals across 12 markets, making it the largest available primary research on privacy program economics.

Cisco data point	Year	Source
96% report privacy benefits exceed costs	2025	Cisco 2025 study
Median ROI 1.6x on $2.7M average spend	2025	Cisco 2025 study
29% report ROI of 2x or higher	2025	Cisco 2025 study
99% report at least one tangible benefit	2026	Cisco 2026 study
Share of orgs spending $5M+: 14% to 38%	2024 to 2026	Cisco 2026 study
93% plan increased privacy resources over 2 years	2026	Cisco 2026 study

Two adjacent ROI vectors deserve mention. Trust centers (the Drata / Vanta / Conveyor pattern of automated security questionnaire response) enable 70 to 90% reduction in security review time and 85%+ reduction in manual questionnaire work, with measurable deal velocity impact. Cyber insurance premiums are down 7% globally and 3% in the US in Q4 2025, and combined controls (MFA, third-party risk management, XDR) drive 20 to 50% premium reductions.

Phased Cost Modeling Framework

Cost questions usually arrive in one of two forms: “what does year one cost” and “what does steady state look like.” This section answers both for three organizational tiers.

The Phased Model

Phase 1 (Months 1 to 6): address critical gaps. Initial investment is the highest. Typical activities include privacy registry setup, data inventory, vendor due diligence on existing sub-processors, and DSAR workflow stand-up.

Phase 2 (Months 7 to 12): expand to medium-risk activities. Add PIA process, consent management deployment, and cross-border transfer documentation per Part 6.

Phase 3 (Year 2+): comprehensive coverage and steady-state operations. PETs, AI-specific governance, machine unlearning preparation.

Organizational tier	Year 1 implementation	Year 2+ steady state	Source pattern
Medium (200 to 1,000 employees)	$75,000 to $250,000	$40,000 to $100,000 ongoing	ComplyDog range
Large (1,000 to 10,000 employees)	$250,000 to $1,000,000+	$150,000 to $500,000+ ongoing	ComplyDog range
Global enterprise (10,000+)	$1M to $5M+	$2M+ ongoing	Secure Privacy reports 20% spent over $1M

These ComplyDog and Secure Privacy ranges are vendor-published and skew toward the high end. They are useful as ceilings for budget defense and not as midpoints. Cisco’s $2.7M annual average across all sizes is the better midpoint anchor.

The most important warning in any cost model: organizations underestimate GDPR compliance costs by 40 to 60% when planning. That 40 to 60% buffer is not contingency. It is the actual gap between planning and reality.

Industry Variations

Industry	Year 1 cost range	Annual maintenance	Driver
Healthcare (HIPAA, large systems)	$75,000 to $500,000+	$25,000 to $150,000	High breach cost ($7.42M average) and 89% experienced a breach in past 2 years
Healthcare (small practice)	$8,000 to $25,000	$39 to $99 per month software	Low data volume but personal liability
Financial Services (PCI DSS, large)	$150,000 to $1M+	Same range	QSA engagement $20K to $100K+, penetration testing $5K to $30K+
Fintech (GDPR)	$20,000 to $50,000 (small) up to $10M+ (mega)	Variable	DORA applicability, payment regulator overlap

Training Costs

Training is the line item most cost models forget. Management training runs $1,000 to $3,000 per employee, general staff training $300 to $800, and ongoing online platforms $50 to $200 per employee per year. For a 2,000-person organization that lands at $100K to $400K annually, separate from the platform spend.

What this looks like in practice. When a CFO asks “what does this cost in year three,” the answer is not the year-one number minus implementation fees. It is the year-one number plus DSAR volume growth (typically 30 to 50% YoY in regulated markets), plus PET pilot costs that move from research to operations, plus an additional FTE per major regulation that comes into force. The privacy cost curve does not plateau the way most software cost curves do.

The Numbers I Trust vs the Numbers I Discount

Source quality matters more in cost modeling than in any other section of the privacy series. Vendor-commissioned ROI studies are useful directionally and dangerous as planning inputs. Survey data from large multi-year longitudinal studies (Cisco, IAPP, IBM) is the best available. Procurement databases like Vendr provide medians from real transactions and are underused. Peer-reviewed PET cost data is sparse but trustworthy when it exists.

Source type	Trust level	Use case
Primary survey research (IBM, Cisco, IAPP)	High	Anchor numbers for budget defense
Enforcement databases (CMS Tracker, CPPA announcements)	High	Cost-of-doing-nothing baselines
Procurement databases (Vendr)	Medium-high	Vendor pricing reality checks
Peer-reviewed papers (ACM, NDSS, Nature Sci Reports)	High	PET overhead and benchmark data
Forrester / Gartner analyst research	Medium-high	Direction; treat absolute ROI numbers with skepticism
Vendor-commissioned TEI studies	Low-medium	Plausible upper bounds, never midpoints
Vendor marketing blogs	Low	Useful only for trend direction, never planning
Salary aggregators (Glassdoor, ZipRecruiter)	Medium	Direction; verify against IAPP and Levels.fyi for validation

The discipline that separates good cost models from bad ones is sourcing every number to one of the higher-trust tiers and flagging every figure that comes from a lower-trust source. Worldwide context: Gartner projects $213B in worldwide end-user information security spending in 2025, of which privacy is a growing share. The privacy management software market alone is $2.14B in 2024, projected to $19.77B by 2033 at 27% CAGR.

Where the Series Got the Cost Math Wrong

Three cost-related claims in the original series deserve revision in light of better data.

First, Part 9 implied PET costs were converging toward usability across the board. The 2025 peer-reviewed FHE benchmarks confirm that frontier-scale FHE remains 1,000x to 1,000,000x slower than plaintext for general workloads. PET selection is still a triage exercise where most workloads cannot use the most powerful PETs.

Second, Part 6 treated the OneTrust ROI claim as ambient context. The 227% ROI figure is from a vendor-commissioned Forrester TEI and should not be used in a budget defense without disclosure of its source.

Third, the series did not quantify the cost of doing nothing in dollar terms a CFO would accept. This appendix corrects that gap: the $10.22M US average breach cost, the $264 per-record cost, and the EUR 1.145B in 2025 GDPR fines are the right anchor numbers for the conversation.

What Has Changed Since This Appendix’s Baseline Data (Spring 2026)

May 8, 2026. California’s attorney general settled with General Motors for $12.75 million over OnStar data sold to LexisNexis and Verisk without proper notice, the first CCPA settlement built on data minimization and purpose limitation. At 4.6 times Disney’s $2.75 million, it replaces Disney as the largest CCPA penalty to date; the enforcement table above is stale on that point.

August 2026. IBM’s 2026 Cost of a Data Breach Report, covering breaches from March 2025 to February 2026, reversed the 2025 edition’s decline: the global average rose 12% to $4.99 million and the US average climbed to $11.5 million, up from the $10.22 million this appendix cites throughout. Every dollar figure in this piece keyed to the 2025 IBM report is now a floor, not a ceiling.

Do Next

Priority	Action	Why It Matters
This week	Pull your DSAR volume for the last four quarters. Multiply by $1,400 per request to get current operational tax.	Most organizations have never measured this, and the number is usually large enough to justify a workflow automation investment by itself.
This week	Run the 8-component diagnostic from Part 5 and assign each unbuilt component a Year 1 dollar estimate from the table in this article. Total it.	This produces a defensible bottom-up budget that maps to specific capability gaps rather than a top-down “match the Cisco $2.7M” estimate.
This month	For every vendor proposal you receive, ask the vendor to disclose whether the ROI study they cite was vendor-commissioned. Document which ones are and which are not.	The Forrester TEI pattern is industry-wide. Knowing which ROI claims are vendor-funded changes how you weight them.
This month	Build the cost-of-doing-nothing baseline for your organization: expected breach cost (records exposed times $264 in US, $189 in EU) plus 1-year fine exposure based on regulated revenue.	This is the upper-bound number that justifies the privacy budget. CFOs respond better to “we are insuring against $30M of expected loss” than to “we need a privacy program.”
This quarter	Map every PET in Part 9 to the cost band in this article. Identify which ones are feasible for which Restricted-tier asset given your compute budget.	Tokenization and TEEs are operationally feasible today; FHE and frontier-LLM DP are not. Cost-aware PET selection prevents shelfware.
This quarter	Add the 40 to 60% planning underestimation buffer to your Year 1 budget request. Document it as buffer, not contingency.	This is the single most consistent finding across cost-modeling sources. The buffer is the difference between the budget surviving and the program collapsing in month 9.
This year	Establish a quarterly cost review with Finance that tracks: privacy team headcount, platform spend, DSAR cost per request, and breach-cost-avoided estimates.	Privacy budgets are renewed against trend lines, not absolute numbers. Quarterly reviews produce the trend lines that make the renewal defensible.

Looking Forward

Cost modeling in privacy will get harder before it gets easier. AI Governance is collapsing into privacy budgets at most organizations (per the IAPP’s 2025 governance report, 69% of CPOs now own AI governance). PET infrastructure is moving from research to procurement at the same time machine unlearning emerges as a regulatory requirement without a mature commercial solution.

The Cisco data point that captures the moment best: the share of organizations spending $5M+ on privacy moved from 14% in 2024 to 38% in 2026, a 171% rise over two years. That is not a slow ramp. That is a structural budget reallocation under regulatory and reputational pressure.

Appendix C covers the other reason privacy programs stall: the political dynamics that make even a well-funded program hard to implement. The cost model in this article tells you what to ask for. The next one tells you why getting it approved is only the start of the problem.