Chargeback & showback models

# Chargeback and showback models

When Spotify's data platform team ran the numbers in the late 2010s, they found a single behavior driving a disproportionate share of their cloud bill: analysts re-running full-table scans across petabytes because querying was free at the point of use. Nobody was malicious. Nobody was even careless. The problem was structural, the people generating the cost never saw the cost. The platform absorbed it, the CFO noticed it, and the CDO owned it.

This is the failure mode that chargeback and showback exist to fix. You already know the platform economics from the architecture module. What this lesson is about is the *incentive engineering*: how attributing data cost, and eventually data value, to the teams that consume it changes what they do, and how you roll that out without igniting a political war that gets your program killed in month three.

The three models and what each one actually buys you

Cost attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → lives on a spectrum, and confusing the levels is the most common way CDOs lose credibility early.

Showback exposes consumption without moving money. You tell the marketing team, "Your dashboards and pipelines cost $340K last quarter." No invoice, no budget transfer. The purpose is visibility and behavior change through social pressure and awareness.

Chargeback actually transfers cost to the consuming team's P&L or budget. The $340K leaves marketing's budget and offsets the platform's. This creates real financial accountability but also real conflict, because now you're touching other executives' numbers.

Value attribution is the frontier most organizations never reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.View full definition →: mapping not just what a team *spent* on data but what that data *produced*. This is where the CDO conversation shifts from "data is a cost center" to "here's the return on the data investment", the theme running through this whole module.

Here's the judgment call that matters: these are stages, not options. Attempting chargeback before showback has matured is the single most reliable way to trigger the turf war. You need a period where teams can see their consumption and adjust *before* the numbers hit their budget. Skipping showback is like sending someone a utility bill for a house that never had a meter, they'll dispute every line, and they'll be right to.

| Model | Money moves? | Primary effect | Political risk |

|-------|-------------|----------------|----------------|

| Showback | No | Awareness, self-correction | Low |

| Chargeback | Yes | Hard accountability | High |

| Value attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → | Yes (net) | Reframes data as investment | Medium (needs trust) |

The strategic sequencing insight: run showback for at least two full budget cycles before proposing chargeback. During that window you're not just building data, you're building the legitimacy of the numbers themselves.

Building an attribution modelattribution modelA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → that survives scrutiny

The technical foundation is straightforward; the *defensibility* is where CDOs fail. When you tell a peer executive their team owes $340K, the first thing they'll do is attack your methodology. If the model is a black box, you lose.

Tag everything at the source

AttributionAttributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → is impossible without a consistent tagging discipline enforced at the point of resource creation. This is a governance extension, not a new capability. Every query, pipelinepipelineAll active sales opportunities across the stages of the sales process, together with their combined potential value and probability of closing.View full definition →, storage object, and compute job needs to carry a cost-center or team identifier. In modern warehouses this is enforceable:

-- Snowflake: enforce team attribution via query tags
ALTER SESSION SET QUERY_TAG = '{"team":"marketing-analytics","project":"attribution-model","env":"prod"}';

-- Then aggregate cost by tag
SELECT
  PARSE_JSON(query_tag):team::string AS team,
  SUM(credits_used_cloud_services) AS credits,
  SUM(credits_used_cloud_services) * 3.00 AS est_cost_usd  -- your credit rate
FROM snowflake.account_usage.query_history
WHERE start_time >= DATEADD(month, -1, CURRENT_DATE())
GROUP BY 1
ORDER BY est_cost_usd DESC;

The point isn't the SQLSQLSales Qualified Lead: a prospect the sales team has validated as ready for direct outreach and a proposal, having passed clear qualification criteria.View full definition →, it's that you cannot attribute what you cannot tag, and tagging must be mandatory infrastructure policy, not a request. Untagged resources should be treated as a governance violation and, in mature setups, routed to a "shared" or "unattributed" bucket that you deliberately keep visible and uncomfortable, so teams have an incentive to claim their consumption.

Decide your cost allocation philosophy

Three cost types need different treatment, and how you handle each signals your fairness:

• Direct consumption (a team's own queries, their pipelines, their storage): attribute fully and directly. This is the uncontroversial part.
• Shared platform cost (the warehouse baseline, the data catalogdata catalogA centralized inventory of an organization's data assets, enriched with metadata, that helps people find, understand, and trust the data they need.View full definition →, the governance tooling, your team's salaries): this is where fights happen. You can allocate proportionally to consumption, split evenly, or absorb it centrally. Absorb the pure-platform overhead centrally, if you charge teams for your own headcount, you've made data infrastructure feel like a tax rather than a service.
• Data acquisition cost (that $2M third-party datathird-party dataData purchased from external aggregators, collected from audiences you don't own. It is bought or licensed rather than gathered through your own direct relationships.View full definition → feed): allocate to the teams that actually use it, which forces a healthy conversation about whether the feed is worth renewing.

The principle: charge for marginal consumption, absorb fixed platform cost. This mirrors how mature FinOps organizations treat cloud, and it's defensible because it maps cost to controllable behavior. Nobody can reduce your platform's baseline; everybody can reduce their own wasteful queries.

Make the model legible before it's binding

Publish the methodology as a one-page document that a non-technical executive can read: what's measured, what rate is applied, what's excluded, and how to dispute a charge. The dispute process is not optional. A chargeback system without an appeals mechanism is a declaration of war. When someone contests a number and you investigate transparently, you build the trust that makes the *next* hundred charges go uncontested.

The behavior change is the whole point

Cost recovery is not why you do this. The finance department could recover cost with a crude even-split. You do chargeback because it changes what people do, and the changes are predictable enough to design for.

When Spotify eventually gave teams visibility into their own compute cost, the behavior shifts were immediate and rational: teams added partition filters they'd been too lazy to write, they killed dashboards nobody opened, they questioned whether that hourly refresh really needed to be hourly. None of this required a mandate. It required a mirror.

But watch for the perverse incentives, because they're real and they're serious:

• Under-investment in valuable analysis. If querying data feels expensive, a product team may stop exploring, and you've now taxed the exact experimentation your data strategy is supposed to encourage. This is why showback often changes behavior more healthily than chargeback: awareness without financial pain.
• Shadow data estates. Charge too aggressively for the central platform and teams will spin up their own untracked pipelines to escape it, destroying the governance you spent years building. Your chargeback rate is implicitly competing with the "cost" of defection.
• Gaming the attribution. Teams will label expensive workloads as "shared" or "R&D" if those buckets are cheaper. Your tagging categories are an incentive surface, design them knowing people will optimize against them.

The design response is to pair every cost signal with a value signal and an efficiency path. Don't just tell marketing they spent $340K. Tell them: here's your spend, here's the estimated revenue influenced by these assets, and here are three specific optimizations that would cut 20% without losing anything. You're not the tax collector; you're the efficiency partner who happens to keep score.

Rolling it out without a turf war

This is the section that separates CDOs who scale their programs from those who get quietly defunded. The technical model is 30% of the work. The political rollout is 70%.

Get the CFO as co-sponsor before you talk to anyone else

Chargeback moves money between budgets, which means it is a *finance* mechanism, not a data mechanism. If it looks like the CDO grabbing power over other executives' budgets, you lose. If it looks like a joint CDO-CFO initiative to improve capital allocation, you win. The CFO's authority makes the budget transfers legitimate in a way your authority never can. Never propose chargeback as a solo CDO initiative. Walk in with the CFO beside you, or don't walk in.

Sequence your rollout to build allies, not enemies

Start with a friendly, high-maturity team as a pilot, ideally one that already suspects it's efficient and wants the data to prove it. Let them help refine the model. Their public endorsement ("this actually helped us cut waste") is worth more than any deck you build. Then expand to teams that are heavy consumers, where the ROIROIReturn on Investment: the ratio of net profit to the cost of an investment. A 300% ROI means each dollar invested returns $3.View full definition → conversation is most productive. Save the politically sensitive, low-maturity teams for last, when the system is proven and normalized.

Frame it as budget, not blame

Language decides outcomes. "You're overspending" triggers defense. "You now own and control this budget line, here's how to optimize it" triggers ownership. The reframe that works: you're not taking money from teams; you're giving them control over a cost they previously couldn't see or influence. A budget you control is power. A cost you can't see is a threat. Sell the former.

Set the intervention threshold deliberately

Don't chargeback everything from day one. A common mature pattern: showback for all, chargeback only above a materiality threshold (say, teams consuming more than $250K annually), where the amounts justify the administrative and political overhead. Small consumers stay on showback indefinitely, chasing $12K from a five-person team costs more in goodwill than it recovers in dollars.

Instrument the trust, not just the cost

Track your dispute rate and dispute-resolution time as first-class metrics. A rising dispute rate is an early warning that your model is losing legitimacy. A dispute rate near zero after the first two cycles is the signal you've earned the right to advance from showback to chargeback, or from chargeback to full value attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition →. The health of the program is measured by trust, not by dollars recovered.

The endgame, which connects back to this module's core thesis: once teams accept cost attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → and the numbers are trusted, you extend the same machinery to *value*. The same tags that told you marketing spent $340K can be joined to the revenue those assets influenced. Now the board conversation changes from "the data team is expensive" to "every dollar of data investment returns X", and that is the argument that funds your next three years.

Key Takeaways

1. Sequence showback before chargeback, always. Run showback for at least two budget cycles so teams can see and self-correct before money moves. Skipping this step is the number-one cause of failed programs and turf wars.

2. Charge for marginal consumption, absorb fixed platform cost centrally. Attributing your own headcount and baseline infrastructure makes data feel like a tax; attributing controllable consumption drives real efficiency behavior.

3. Co-sponsor chargeback with the CFO, never solo. Budget transfers need finance's authority to be legitimate. As a solo CDO initiative it reads as a power grab and dies.

4. Design against perverse incentives explicitly. Aggressive charging creates shadow data estates, under-investment, and attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition → gaming. Pair every cost signal with a value signal and a concrete efficiency path.

5. Measure trust, not just recovery. Track dispute rate and resolution time; a low, stable dispute rate is your green light to advance from showback to chargeback to full value attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition →, the move that reframes data from cost center to investment.