The investor offers a valuation. You spend the next week deciding whether to accept it, counter it, or walk. All three of those responses are the same response.
You are treating their number as the middle of the range. The moment you argue against a number, you have agreed it is the thing being argued about.
Why these three models
The decision is what to do with an offer, and more importantly what to do before one arrives. The features that fire are outcomes evaluated relative to a comparison point rather than absolutely, a market figure whose origin nobody checks, and a back-and-forth whose pattern tells you more than its content.
Three lenses. Reference dependence produces an equilibrium answer about how any offer will be judged. Aggregation produces a random answer about where the comparison number came from and how much it is worth. Behaviour modes produce a cycle answer about whether the negotiation is heading somewhere or merely repeating. The third one is diagnostic rather than tactical and it is the cheapest of the three to run.
1. Nothing is evaluated in absolute terms
People do not assess outcomes against some internal scale of value. They assess them as movements from a reference point, and the same objective position is experienced as a gain or a loss depending on where the reference sits.1
Two consequences matter commercially. The first is that whoever establishes the reference has framed everything that follows. An offer presented as the standard terms for this stage places itself at the reference and pushes your counter into the position of a demand. The same offer presented after you have already circulated your own terms is a discount request.
The second is that losses relative to the reference weigh more heavily than equivalent gains, which means the party defending the reference has an easier job than the party attacking it. This asymmetry is one of the most consistently reproduced findings in the field, though its magnitude varies considerably with context and is contested at the level of the specific coefficient.2
The practical instruction is unusually simple and almost nobody follows it. Say your number first, in writing, before the conversation that would otherwise establish theirs. Not because first offers win, which is a claim about magnitude this article does not make, but because the reference point exists whether or not you set it, and the alternative to setting it is inheriting it.
Going first has one failure mode, and it is the reason most founders avoid it. If your number sits outside what the counterparty can approve, sending it first ends the conversation before you learn anything. The test is cheap. Ask what the largest cheque this fund wrote last year was, or what the buyer’s procurement threshold is above which a second signature is needed. A number inside that ceiling gets argued with. A number above it gets forwarded to someone who never replies.
The same move works below the fundraise. A distributor asking for your price list sounds administrative and is the reference-setting exchange. Send unit price, volume break and payment terms in one document, and the negotiation that follows is about which line of your document applies. Let them say what they usually pay, and every conversation after that is a discount request from a number you never agreed to. If you have no list, one line in writing does the job: this price, at this volume, on these terms, valid thirty days.
If you cannot go first, the recoverable move is to reject the reference explicitly rather than to counter within it. That means naming a different basis for the number altogether: not a higher valuation but a different comparison set, not a lower fee but a different unit of pricing. Countering inside someone else’s frame concedes the frame, and the frame is where the value sits.
2. Where the market number actually comes from
Both sides in these conversations refer to what the market says. It is worth asking what that phrase contains, because it is doing enormous work and it is usually hollow.
A crowd estimate is accurate under specific conditions: the individual judgements must be roughly independent, the errors must go in different directions, and there must be enough of them. Meet those conditions and the aggregate beats almost any individual. Fail them and the aggregate is one opinion repeated, which carries all the confidence of a crowd and none of the information.3
Apply that to your market number. In a small ecosystem, the comparison figure typically comes from a handful of people who talk to each other constantly, who saw the same three recent deals, and whose estimates are therefore not independent in any useful sense. That is not a market price. It is a correlated guess with the appearance of consensus, and its confidence is a property of the repetition rather than of the evidence.
Ask how many independent observations sit behind the number, and the market usually turns out to be four people.
This is usable rather than merely satisfying. Asking which comparable transactions the figure is based on is a legitimate question and it has three possible outcomes. They name specific deals, in which case you now have real data. They name a general sense of the market, in which case the reference has just lost most of its authority. Or they decline, which tells you the number is a negotiating position rather than an estimate. All three outcomes improve your position and the question costs one sentence.
There is a second-order effect worth naming. The four correlated people who price you this round will price you next round, and they will use this round as their comparable. A number you accepted because it looked like the market becomes the market for you specifically, with your signature on it. That argues for correcting a low reference now even at the cost of a slower close, and for writing the reason into the deal note if you accept it anyway, so the next conversation starts from a documented distortion rather than a clean precedent.
If they will not give you comparables, build a thin set yourself. Three founders who closed in the last eighteen months, asked privately about structure rather than headline valuation, tell you more than any published figure. The headline is the number everyone rounds up. The structure is the part nobody bothers to distort. Ask what the liquidation preference was, where the option pool was set and who paid the legal fees. Three real answers beat a general sense of the market and cost three messages.
There is a second use for the same question, and it runs in your favour rather than against theirs. Where the comparison set is genuinely thin, both sides are guessing, and a guess is easier to move than a measurement. Founders in small ecosystems often assume they face a well-priced market and concede accordingly, when the truthful description is that nobody in the room knows what the number should be. That is not licence to invent a figure. It is permission to argue from your own evidence, because there is no external standard in the room that your evidence has to defeat.
3. Read the pattern, not the content
The third lens ignores what is being said entirely and looks at the shape of the exchange over time.
Any back-and-forth exhibits one of a few characteristic patterns. It converges, with each round closing a smaller gap. It oscillates, with positions swinging without narrowing. It escalates, with each round adding demands. Or it goes flat, with rounds continuing and nothing moving. Naming the pattern is more informative than analysing any single round, because the pattern reveals the structure while the content reveals only the current position.4
Each pattern implies a different move. Convergence means keep going and stop optimising the last few percent, since the remaining gap is smaller than the cost of another two weeks. Oscillation usually means the person you are talking to is not the person deciding, and the correct move is to ask who else needs to agree. Escalation means the other side reads delay as weakness, and the response is a deadline rather than a concession. Flat means the deal is already dead and neither party has said so.
A worked instance. Round one, they offer a valuation and ask for a thirty percent option pool. Round two, the valuation moves up slightly and the pool holds. Round three, the valuation moves again, the pool holds, and a condition appears on the board seat. Plotted as a line, valuation converges while terms escalate, which is two negotiations being run as one. The response is not to keep trading valuation. It is to say the valuation is agreed and the terms schedule is now the negotiation, which stops the side that is still moving from paying for the side that is not.
Assume the counterparty is reading the same line. Your rounds are a pattern too, and a founder who concedes a little every round has drawn a converging line straight at their number and shown them where it lands. The defence is not to be erratic, which reads as an approvals problem inside your own company. It is to make one round move nothing at all, with a reason attached, so the line has a flat segment and the extrapolation stops working.
The reason the pattern outranks the content is that the content is chosen by the other side and the pattern is not. Anyone can write a round that sounds like movement. Nobody can make four rounds converge without actually conceding something, which is why the shape is harder to fake than the language. Read the four rounds as a line, because a line cannot be worded persuasively.
Founders almost never run this because each round arrives as an urgent piece of content demanding a reply. Plotting the last four rounds on one line takes two minutes and it regularly shows that a negotiation everyone describes as progressing has not moved since round one.
What the three say together
- Put your number in writing first, or explicitly reject the basis of theirs. Countering inside their frame concedes the frame.
- Ask what independent transactions the market figure is built on. Four correlated opinions is not a market.
- Plot the last four rounds. Converging, oscillating, escalating or flat, and each one calls for a different move.
- Decide your walk-away before the first exchange, because the reference point will move once the conversation starts and your judgement of it will move with it.
Where they disagree
Setting the reference and reading the pattern pull against each other on when to move first.
The reference argument says commit to a number early and in writing, because the frame is decided before the negotiation properly begins. The pattern argument says the first few rounds are your only cheap source of information about who you are dealing with and what they can approve, and a firm early commitment spends that information-gathering opportunity to buy a frame.
The resolution depends on which is scarcer for you. If you have many alternatives, set the reference hard and early; the frame is worth more than the information because you can walk. If this counterparty is one of very few, gather first, because losing the deal by over-anchoring costs more than an unfavourable frame you can still negotiate inside. Founders raising a first round are usually in the second position and behave as though they are in the first, having read advice written for the first.
What none of them contain
None of the three prices the relationship after the deal. An investor you anchor aggressively against is on your board for the next seven years, and a customer whose frame you reject in the negotiation still has to be served afterwards. Every model here optimises the transaction and none of them sees the period after it.
None of them handles the counterparty who has done this two hundred times against your three. Reference dependence describes both of you, but experience changes how much the reference moves someone, and the models treat the two sides symmetrically when the asymmetry is the main feature of the situation.
And none of them can tell you whether your own number is any good. All three are about how a number will be received. Whether the number reflects what your company is actually worth is a separate question and no amount of framing improves a bad answer to it.
The one action that survives the ignorance: before your next negotiation, write your number and its basis in one sentence and send it first. If you cannot write the basis, that is the finding, and the week spent constructing one is better spent than the week you would otherwise spend reacting to theirs.
Who has to move
The founder does, and specifically before the meeting rather than in it, because the reference is set in the first exchange and everything afterwards is negotiation inside it. The instinct is to wait and see what they offer, which feels like gathering information and is actually conceding the frame. The cheapest first test costs one email: state your number and its basis before the call is scheduled. The response tells you more about the counterparty than the call would have, and it arrives before you have spent anything.
Sources and notes
- Sanjit S. Dhami, The Foundations of Behavioral Economic Analysis, Oxford University Press. Reference dependence, the evaluation of outcomes as gains and losses relative to a reference point, and the associated treatment under prospect theory are developed in the part on behavioural decision theory. Used in section 1 for the mechanism.
- Sanjit S. Dhami, The Foundations of Behavioral Economic Analysis, Oxford University Press. The asymmetry between losses and gains relative to the reference point is among the most reproduced findings in the field, while the specific coefficient is contested and varies with domain, stake size and elicitation method. This article deliberately states the direction and no magnitude, because the magnitudes circulating in business writing are typically quoted from a single early estimate as though it were a constant.
- Jeffrey Carpenter and Andrea Robbett, Game Theory and Behavior, MIT Press. The conditions under which aggregated judgements outperform individual ones, and the failure of aggregation when individual estimates are correlated rather than independent, are treated alongside information cascades and social learning. Used in section 2, including the point that correlated estimates produce confidence without adding information.
- Donella H. Meadows, Thinking in Systems: A Primer, Chelsea Green Publishing, and John D. Sterman, Business Dynamics: Systems Thinking and Modeling for a Complex World, McGraw-Hill. The practice of identifying a system’s behaviour over time, whether it converges, oscillates, grows or stagnates, before attempting to explain any single observation, is central to both. Used in section 3, applied here to the sequence of rounds rather than to a physical system.
A note on a number this article does not give. You will find claims that the first offer captures a specific share of the bargaining zone. Those figures come from laboratory settings with a known zone and a fixed number of rounds, and neither condition holds in a fundraise. The direction transfers. The number does not.
Joshua Agonya Pi’Rwot, Founder.