back
loading skill details...
Activate when: user says 'what's a fair price / starting number / ballpark'; counterparty just opened with a number; user is about to negotiate salary, valua...
---
name: anchoring
description: "Activate when: user says 'what's a fair price / starting number / ballpark'; counterparty just opened with a number; user is about to negotiate salary, valuation, or deal terms; user is making a forecast and a number has already been floated; user asks about 'first offer', 'framing effect', or 'being anchored'.
Do NOT activate when: the number in question is a verifiable data point (audited financials, confirmed comparable) and is genuinely informative; or the decision is so trivial that the economic impact of anchoring is negligible. More: deciqai.com/c/anchoring"
---
# Anchoring
## Overview
When people estimate an unknown quantity they start from a reference point — an **anchor** — and adjust. The adjustment is almost always insufficient, so the final answer stays closer to the anchor than it should. This pattern holds even when the anchor is explicitly random and subjects are told to ignore it (Tversky & Kahneman 1974). It survives expertise, explicit warnings, and financial incentives for accuracy.
Strategically: almost every consequential negotiation, valuation, or forecast begins with an anchor. The first number said or written — asking price, salary band, prior-round valuation, revenue projection — anchors everything that follows.
Composes with: `pricing-strategy` · `expected-value-and-kelly` · `probabilistic-thinking` · `signaling-games`
## When to Use
Apply when:
- About to negotiate a price, salary, valuation, contract term, or settlement
- Estimating an uncertain quantity after a number has already been mentioned
- A counterparty has opened and your reasoning is gravitating toward their number
- Resetting expectations after an anchor has already been set
- Sizing an AI startup valuation, raise, or AI-capex/adoption estimate where a headline round or a benchmark like Nvidia's market cap has already been floated as "the market"
**When NOT to use:**
- The number is genuinely informative (verifiable comparable, audited cost basis) — data, not an anchor
- Decision is trivial; you are past the negotiating phase; or anchor was set by you intentionally
## Coaching Novices (Adaptive Front Door)
- **Engine mode:** user has a concrete case → run The Process directly.
- **Coach mode:** user is unfamiliar or has no concrete case → guide step by step.
In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.
1. One-line what-it-is: when you estimate or negotiate a number, the first number you hear pulls your answer toward it — even when you know it's random or wrong.
2. Check fit against When to Use / When NOT to use. If the number is genuine data, say so.
3. Elicit their real situation — a number that's been mentioned, an upcoming negotiation, a figure that's been floated. > **[WAIT — do not advance until user responds]**
4. Walk through The Process one step at a time with their input. > **[WAIT — do not advance until user responds]**
5. Close by naming the one concrete move that fits their situation (open with X / counter with Y / reset to Z). > **[WAIT — do not advance until user responds]**
## The Process
Run the **Anchoring Analysis**. Detect anchors, judge their informational content, decide deliberately.
1. **Identify the anchor.** Name the specific number explicitly — it often arrives as casual mention, "data," or expectation, not as a labeled offer.
2. **Classify informational content.** Genuinely informative → use as evidence. Partial (real signal + proposer's interest) → discount. Pure anchor → treat as zero information; bias still operates, apply structural defense.
3. **Run the anchor-free counterfactual.** What number would you have produced without hearing the anchor? Write it down first.
4. **Consider an opposite-side anchor.** Deliberately generate a number in the opposite direction and articulate the reasons — not to use as a counter, but to break the asymmetry.
5. **If proposer:** open bold but defensible. Bold openers consistently outperform soft ones; absurd anchors lose power and damage the relationship.
6. **If respondent:** name it, reset with your own independently-generated number, or work with it knowingly — discounting your natural agreement point.
7. **Audit hidden anchors in forecasts.** Last quarter, board expectations, prior-year baselines are all anchors. Identify which is doing the work.
8. **Stop-rule:** if the analysis doesn't change the decision, you may have rationalized. Re-run the anchor-free counterfactual.
### Output: the Anchoring Analysis
```
Anchor: [number] | Source: [who/how] | Status: [offer/comparable/casual/expectation]
Informational content: [genuinely informative / partial / pure anchor] — Evidence: […] Counter-evidence: […]
Anchor-free counterfactual: [my number without the anchor] | Gap vs. current "natural" answer: […]
Opposite-anchor test: [what would justify the other extreme] → range implication: […]
Decision: [proposer: bold-but-defensible opening] / [respondent: name/reset/work-with + specific number]
Risk anchor still operating: […] (how I'll know if I've been pulled too far)
```
*→ Method in Action: [Tversky & Kahneman's Wheel of Fortune, 1974](examples/tversky-kahneman-wheel-of-fortune-1974.md) · [Judges Playing Dice with Criminal Sentences, 2006](examples/englich-judges-dice-criminal-sentences-2006.md)*
*→ 2026 lens: [AI Fundraising and Valuation Anchors (2023–2026)](examples/ai-fundraising-valuations-anchoring-2023-2026.md) — headline AI rounds and Nvidia's market cap anchoring the next valuation*
## Pack: Anchoring in Practice
- **Pricing/sales:** list premium tier first; generate willingness-to-pay before seeing the seller's price.
- **Salary:** candidates who name first do better; open bold from market data. Employers: the first number in the offer email anchors negotiation.
- **Real estate:** buyers pull sold (not listed) comparables before viewing the listing.
- **Startup funding:** bold-but-defensible asks correlate with higher closes; investors build their valuation model before reading the pitch deck.
- **Forecasting:** zero-based forecasting and independent pre-aggregation estimates defend against last-year anchoring.
- **Legal:** high opening demands → higher settlements; defendants counter-anchor with evidence-based numbers.
## Applying It Well
- Generate your anchor-free number **before** any external anchor is introduced — structural defense beats procedural.
- Bold-and-flexible beats soft-and-firm: open aggressively, concede during negotiation.
- Recognition is the start of defense, not the conclusion; pair it with counterfactual generation.
- "Consider the opposite" (Mussweiler et al. 2000) reduces but does not eliminate the bias.
*→ Primary sources: [references/sources.md](references/sources.md)*
## Common Rationalizations
**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**
| Fake move | Reality |
|---|---|
| [D] "I'm an expert — I'm not anchored" | Northcraft & Neale 1987 (real-estate agents) and Englich et al. 2006 (judges) show expertise is not protective. |
| [D] "I know the anchor is random, so it won't affect me" | The 1974 wheel told subjects it was random. Effect operated anyway. Knowing is the start of defense, not the conclusion. |
| [D] "I'll consider the anchor and then make my own judgment" | Every anchored subject believed this. Defense is generate-your-number-first, then look at the anchor. |
| [D] "Going first shows my cards — I'll let them open" | Galinsky & Mussweiler 2001: bold first offers produce better outcomes. Letting them open loses money on average. |
| [D] "Our forecast is bottom-up, not anchored" | If it started from last year's number or referenced the board's expectation, it is anchored. |
| [D] Setting an anchor outside the bounds of plausibility | Absurd anchors lose power and damage the negotiation. Bold = maximum within the plausibility range. |
| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |
## Red Flags
- A casual number ("ballpark," "let's just say X") is now the implicit baseline of the conversation
- Your estimate was generated after hearing the anchor, with no anchor-free counterfactual written down
- You are deferring on opening offers to avoid "showing your hand" — this intuition is systematically wrong
- Forecast = "last year ± adjustment" — adjustment is almost always insufficient
- Decision-maker says "the number isn't supposed to influence me, but…" — that "but" is the anchor
## Verification
- [ ] Anchor explicitly identified and named
- [ ] Informational content classified (genuinely informative / partial / pure anchor)
- [ ] Anchor-free counterfactual generated before the anchor settles in
- [ ] Proposer: opening anchor bold but defensible within plausibility range
- [ ] Respondent: chose name / reset / work-with and named the specific number
- [ ] Decision meaningfully different from pre-analysis — if not, may have rationalized rather than de-anchored
- [ ] In forecasting: hidden anchors (last period, board expectation) identified, not just explicit ones
---
*Part of **deciqAI Knowledge Skills** — 227 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. **See it run → https://www.deciqai.com/c/anchoring** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*
*Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/anchoring.json*
don't have the plugin yet? install it then click "run inline in claude" again.
when people estimate an unknown quantity, they start from a reference point (an anchor) and adjust insufficiently. the final answer stays closer to the anchor than it should. use this skill any time a number has been floated in negotiation, valuation, forecasting, or estimation and you need to decide whether to accept it as a baseline or generate an independent one. activate when you're about to name a price, salary range, valuation, forecast, or settlement; when a counterparty has opened; when your reasoning is gravitating toward their number; or when resetting expectations after an anchor has already tilted the conversation. do not activate if the number is a verifiable data point (audited cost, confirmed comparable sale, market-tested price) with genuine informational content, or if the economic impact is negligible.
parameters:
context:
external connections:
The Anchoring Analysis (run these steps in order):
Identify the anchor explicitly. Name the specific number. write it down. note how it arrived (formal offer, casual mention, data, expectation, baseline). anchors often hide inside language like "ballpark," "we're thinking," "last quarter was," or "the market is pricing at." make it visible.
Classify informational content. does this number contain real signal or is it pure positioning? three categories exist: genuinely informative (audited cost, confirmed comparable, market-tested price), partial (real signal mixed with proposer's interest, discount accordingly), pure anchor (proposer's positioning with no factual basis, treat as zero information and apply structural defense).
Generate anchor-free counterfactual. what number would you have produced without hearing the anchor? use market data, cost-plus logic, comparable transactions, or first-principles reasoning. write this down before the anchor settles into your thinking. this is the structural defense.
Run the opposite-anchor test. generate a defensible number in the opposite direction. if anchor is high, what would justify the low end? if low, what would justify high? articulate the reasoning. purpose is to break asymmetry, not to choose the opposite, but to see the plausibility range.
If proposer (you are going first): open bold but defensible. bold first offers consistently outperform soft ones (Galinsky & Mussweiler 2001). defensible means within the plausibility range for your domain. absurd anchors lose power and damage the relationship. choose your opening from the high end of your anchor-free counterfactual or opposite-anchor range, but stay in the realm of justifiable.
If respondent (they opened first): name it, reset, or work with it knowingly. three moves exist. (a) name the anchor: "i hear you at X, and i want to work from a different baseline." (b) reset with your own independently-generated number: "my number, working from first principles, is Y." (c) work with it knowingly: if their number has informational content, discount it for their interest and adjust. do not default to split-the-difference or immediate acceptance.
Audit hidden anchors in forecasts. in forecasting, the anchor is often implicit. last quarter's revenue, the board's expectation, last year's budget, prior-round valuation. identify which one is doing the work in your forecast. if your forecast is "last year ± adjustment," run steps 1-4 on the implicit baseline.
Stop-rule: test whether the analysis changed the decision. if you have run the anchoring analysis and your decision is identical to your pre-analysis intuition, you may have rationalized rather than de-anchored. re-run the anchor-free counterfactual. if it still produces the same output, the analysis was sound. if it diverges, go with the counterfactual.
if the anchor is genuinely informative (audited cost, confirmed comparable, market-tested): treat it as evidence. incorporate it into your estimate. do not run defensive steps 3-4 in full; instead, blend it with other data. the skill does not apply when data is real.
if the anchor is partial (real signal + proposer's interest): discount for the proposer's upside. if a seller says "comps in the neighborhood are X," they are anchoring with real data but choosing the high comp. acknowledge the data, generate your own counterfactual, and reset from your anchor-free number. move: reset.
if the anchor is pure (proposer's positioning, no factual basis): apply full structural defense. generate anchor-free counterfactual, run opposite-anchor test, move: name or reset.
if you are proposer and have generated your anchor-free counterfactual: open at the bold end of the range if you have relationship capital; slightly lower if you don't. if your counterfactual range is [10, 15], open at 15 or 14, not 12.
if you are respondent and they opened first: do not split-the-difference as a default. name, reset, or work-knowingly. if you split, you are surrendering half the value of your anchor-free counterfactual.
if you have no access to comparables or verification data: classify the anchor as "pure anchor" and defend accordingly. do not assume proposer credibility.
if time pressure is acute (immediate decision required): generate anchor-free counterfactual in 15 minutes using first-principles reasoning or internal benchmarks. do not defer; deferral locks in the anchor.
if you are forecasting and the hidden anchor is prior-year results: run zero-based forecast in parallel. if zero-based and prior-year differ by >10%, investigate. do not report prior-year-adjusted as your primary forecast; report both.
if you discover mid-negotiation that the anchor has shifted (they re-anchored lower): treat the new anchor as fresh; restart the process. do not be anchored to being anchored.
the Anchoring Analysis artifact in this format:
ANCHORING ANALYSIS
==================
Anchor: [exact number] | Source: [who stated it and how] | Status: [offer/comparable/casual/expectation/baseline]
Informational content: [genuinely informative / partial / pure anchor]
, Evidence supporting: [facts that make this number real]
, Counter-evidence: [facts that contradict or contextualize it]
Anchor-free counterfactual: [your number, methodology]
, Methodology: [cost-plus / comparable / market rate / first-principles / internal benchmark]
, Confidence: [high / medium / low]
Opposite-anchor test: [what would justify the other extreme]
, Range implication: [low end] to [high end] with reasoning
Decision & Move:
, Your role: [proposer / respondent]
, Move: [opening anchor / name / reset / work-knowingly]
, Your number: [specific figure]
, Stated reasoning: [one sentence, defensible to counterparty]
Risk: anchor still operating: [how will you know if you've been pulled too far?]
Stop-rule check:
, Pre-analysis intuition: [A]
, Post-analysis decision: [B]
, Divergence: [yes/no]
, If yes, adopt: [B]
file location: capture in negotiation notes, forecast document, or decision memo. do not let the analysis exist only in working memory; write it down before you move.
format: plain text, markdown, or email-ready prose. do not require specialized software.
edge case handling: if counterfactual cannot be generated (insufficient data), state "insufficient data for anchor-free counterfactual; classify anchor as pure and apply name/reset defense." if decision is trivial or economic impact <$10k equivalent, note that anchoring analysis may not be worth the time; decide to skip at your discretion.
you know the skill worked if:
you identified an anchor that was not previously explicit. you wrote down a specific number, source, and status that shapes the conversation. before this skill, you might have said "their ask seems high"; now you say "they opened at $X, which is Y% above my anchor-free estimate."
you generated an anchor-free counterfactual before committing to a position. you have a number in writing that you would have produced without external input. this number is your true estimate, not a rationalization of someone else's offer.
your decision diverged from the anchor. if the anchor is $100 and your counterfactual is $65, you either (a) opened at $70 (proposer) or (b) reset the conversation to $65 (respondent), or (c) worked from $65 with clear awareness that you are adjusting upward. you did not default to $85 (the middle ground).
you named the anchor or reset the baseline. in a negotiation, you said something like "i'm seeing that number, and here's my independent calculation" or "let me start from a different baseline." the conversation is no longer anchored to their initial number by default.
in forecasting, you identified a hidden anchor and ran a parallel zero-based estimate. you said "last year was $500k; with no history, i'd forecast $400k" or "with history considered, i'm at $420k." the board now sees both, not just "last year ± 5%."
you did not rationalize post-hoc. you followed the procedure before the anchor settled in, not after. the analysis changed or confirmed your intuition, but it was real decision-making, not justification.
the negotiation or estimate moved meaningfully away from the initial anchor. if proposer, you opened bold and moved the range. if respondent, you reset the baseline or worked knowingly. if forecaster, your estimate is less dependent on prior period. the anchor did not control the outcome.
you can articulate why the anchor matters and when it doesn't. you can tell a counterparty or colleague: "that number is real market data, so we use it; this other one is pure positioning, so we ignore it and reset." you have made the classification explicit and can defend it.
engine mode: user has a concrete case (specific number, negotiation, or forecast). run the anchoring analysis directly.
coach mode: user is unfamiliar or exploring the concept. guide step by step. each [WAIT] is a hard stop; output only that step's question, then wait for response before advancing.
one-line what-it-is. when you estimate or negotiate a number, the first number you hear pulls your answer toward it, even when you know it's random or wrong. [WAIT]
check fit. is the number you're dealing with genuine data (audited, confirmed comparable) or positioning? if genuine, anchoring doesn't apply; you should use the data. [WAIT]
elicit the real situation. what number has been mentioned, and in what context? (upcoming negotiation, salary discussion, valuation, forecast, price quote, board expectation) [WAIT]
walk the process with their input. working from the anchor you named, what number would you have generated before you heard their number? write that down first. [WAIT]
close with the concrete move. based on your analysis, should you open with X, reset to Y, or work knowingly from Z? name one specific action. [WAIT]
pricing & sales:
salary negotiation:
real estate:
startup funding:
forecasting:
legal & settlement:
| rationalization | reality |
|---|---|
| "i'm an expert, i'm not anchored" | northcraft & neale 1987 (real-estate agents) and englich et al. 2006 (judges) show expertise is not protective. anchor effect operates across experience levels. |
| "i know the anchor is random, so it won't affect me" | tversky & kahneman 1974: subjects were told the wheel was random. effect operated anyway. knowing is the start of defense, not the conclusion. |
| "i'll consider the anchor and then make my own judgment" | every anchored subject believed this. structural defense is generate-your-number-first, then look at the anchor; not look-then-generate. |
| "going first shows my cards; i'll let them open" | galinsky & mussweiler 2001: bold first offers produce better outcomes. letting them open loses money on average. |
| "our forecast is bottom-up, not anchored" | if it started from last year's number or referenced board expectation, it is anchored. bottom-up means zero-based, not "detailed build on top of a baseline." |
| "i'll set an anchor outside plausibility to be bold" | absurd anchors lose power and damage the negotiation. bold = maximum defensible within plausibility, not maximum absurd. |