Applications in Sales & Business

Why sales is the ecosystem most exposed to Goodhart

Every function in a company is measured, but sales combines three traits that make it the #1 risk zone for Goodhart:

1. Fine measurability: every call, every email, every deal is tracked to the second in the CRM.
2. Indexed bonuses: 30 to 60% of compensation depends on KPIs.
3. Information asymmetry: the SDR knows what was said on the call; the manager only sees the counter.

Result: as soon as a KPI becomes a bonus, optimization starts within 30 days. That is why an experienced rep reads your new commission plan the way a developer reads the spec of a security flaw — looking for the shortest path.

Case #1: the "number of calls" KPI

Typical context: sales leadership who believes "more calls = more pipe". Sets 80 calls/day per SDR as a KPI.

What happens in 4 weeks:

How to diagnose:

• Call duration distribution — abnormal spike on 0–10 seconds
• Call/booked-meeting correlation collapsing (from 0.7 to 0.1)
• Unique-calls/total ratio dropping (same numbers redialed)

Antidote:

• Main KPI = qualified meetings booked; secondary KPI = calls (capped at 100/day to discourage pure gaming)
• Quality audit: 5 calls listened to per SDR/week, blind-scored by a coach

Case #2: the "MRR" KPI

Typical context: B2B SaaS that sets signed MRR as the AEs' main KPI.

What happens in 6 months:

graph LR
    A[KPI: signed MRR] --> B[AEs push annual contracts]
    B --> C[Aggressive discounts T-1 day from month-end]
    C --> D[Customers not really qualified]
    D --> E[Massive churn at renewal]
    E --> F[NRR collapses 6-12 months later]
    style F fill:#ffcdd2

The temporal trap: the MRR KPI is immediate, churn is delayed. The AE who games the system gets promoted before the result lands.

Antidote:

• Co-main KPI: MRR at 6 months post-signature (excludes fast churners)
• Clawback: 50% of commission if churn within 90 days
• Qualitative tracking: 3 deals/AE/quarter reviewed by RevOps with a quality score

Case #3: the "NPS" KPI

Typical context: CS leadership that turns NPS into a monthly bonus.

What happens:

• CSMs select customers surveyed (hostile customers "not available")
• They brief customers: "if all is well, give 9 or 10"
• They trigger the survey right after a success, never after an incident
• NPS goes from 32 to 71 in 3 quarters — without any product improvement

Diagnosis:

• Abnormally high response rate (>40%)
• Bimodal distribution flattening
• Decorrelation NPS / actual churn

Antidote:

• Automated random sampling on 100% of base (no cherry-picking possible)
• Measure delayed NPS (90 days after contact, not immediately)
• Cross-reference with an external qualitative score (quarterly call review)

Case #4: "MQLs"

Typical context: marketing paid on the number of MQLs delivered to sales.

What happens:

• The MQL definition silently widens
• Scoring thresholds drop
• Most leads are actually junk
• Sales waste 40% of their time on unqualified MQLs
• The marketing/sales relationship deteriorates silently

Antidote:

• Marketing KPI = SQL accepted + not rejected within 7 days, not MQL
• MQL definition frozen by a tripartite committee (marketing, sales, RevOps)
• Quarterly audit: 50 MQLs blind-reviewed, true-positive rate tracked

Case #5: NRR (Net Revenue Retention)

Typical context: SaaS steering on NRR as North Star KPI.

What happens — extremal variant:

• In normal range, NRR ≈ customer health
• At extremes (large accounts), correlation breaks
• CSMs concentrate 80% of their time on the top 10 accounts (NRR explodes) while the mid-market churns massively
• 18 months later, average NRR remains high but concentration risk has become deadly

This is Extremal Goodhart: the metric works at the center of the distribution, not at the extremes — exactly where optimization pushes.

Antidote:

• Decompose NRR by segment (logo, mid-market, enterprise)
• Cap enterprise NRR in the global calculation (cap at 130% per account)
• Track the distribution of NRRs, not just the mean

The inverted KPI: the "anti-Goodhart" metric by construction

A powerful technique (popularized by Andy Grove at Intel, High Output Management, 1983): for each offensive KPI, create a defensive KPI.

The defensive KPI is structurally immune to gaming the offensive KPI — gaming one degrades the other.

The "no bonuses on metrics < 6 months" rule

Empirical conclusion of multiple studies (Larkin, Strategic Management Journal, 2014): any metric with a feedback loop shorter than 6 months is massively gamed. Metrics that resist have long feedback loops (12-month retention, LTV, NRR, reorder rate).

For a VP Sales, this means a hard rule:

80% of bonus on long-loop metrics, 20% on quick pipeline metrics.

This inverts most teams' practice (90% short-term).

Diagnostic: is your organization infected by Goodhart?

Diagnostic test in 7 questions. Count your "yes":

1. Do you have a single North Star KPI communicated each month?
2. Does more than 50% of bonus depend on fewer than 3 metrics?
3. Have the metrics been stable for the past 18 months?
4. Do seniors explain in meetings how they hit their numbers?
5. Has an "input/output" ratio of a team changed abruptly after a KPI change?
6. Are metrics self-reported (CRM filled by the person being evaluated)?
7. Is there at least one monthly qualitative audit on the reported numbers?

5+ "yes" score: your organization is probably already infected at level 2-3.

Synthesis for a sales leader

Three actions this week:

1. List your 3 main KPIs and identify the most likely Goodhart variant for each
2. Create a defensive KPI for each (the anti-gaming mirror)
3. Schedule a monthly qualitative audit: 5 calls/CSM, 5 deals/AE, 50 MQLs blind-reviewed

The next chapter addresses the AI version of the problem — reward hacking in modern LLMs.