Four people are two years in, their initial grants are nearly vested, and you have decided to refresh. You spend a week working out what each grant is worth.
Nobody will evaluate it that way. Each of them will compare the number to their first grant and to what they believe the person next to them received, and both comparisons will be wrong.
Why these three models
The decision is how to allocate a refresh pool and how to explain it. The features that fire are an amount that will be assessed relative to a reference rather than in absolute terms, information that travels between employees regardless of policy, and an underlying asset whose value is concentrated in outcomes almost nobody in the conversation expects.
Three lenses. Reference dependence produces an equilibrium answer about how the number will land. Spread produces a cycle answer about why the comparisons form whatever you do. Concentration produces a random answer about what the grant is actually worth and why most explanations of it are dishonest. The third is the one founders avoid, and avoiding it is what makes the first two harder than they need to be.
1. It will be judged against a reference you did not choose
People do not evaluate an amount against an internal scale of value. They evaluate it as a movement from a comparison point, and the comparison point is supplied by them rather than by you.1
For a refresh there are three references in play and they conflict. The original grant, in share count, which is the one most employees actually use even though share counts are not comparable across rounds. The original grant in percentage terms, which is the sophisticated version and will show a decline because dilution has happened. And what they believe a peer received, which they will estimate whether or not anyone tells them.
The consequence is that a generous refresh presented badly can read as a demotion. A person who received 10,000 shares at joining and 6,000 now will register a cut, even if the 6,000 are worth several times the original grant, because share count is the reference they have and nobody gave them a better one.
Which makes the framing part of the decision rather than a communication afterthought. Give them the reference you want used, explicitly, and give it before the number: here is what your original grant is worth at the last round price, here is what this grant is worth on the same basis, here is your total. The arithmetic is not hidden and doing it for them replaces a comparison they would have made badly.
A worked example fixes the sizes. Someone joined with 10,000 shares when the round price was USD 1, so their grant was framed at USD 10,000. Two years later the round price is USD 4 and you offer 6,000 shares, which is USD 24,000 on the same basis. In share count the grant halved. In value it more than doubled. In percentage of the company it fell in both readings, because two rounds happened in between. All three statements are true, and whichever one you say first is the one they carry home that evening.
Leading with value has its own failure mode: the round price becomes a promise you did not intend to make. Quote USD 24,000 today and a flat or down round next year converts your own frame into a loss, felt as sharply as a pay cut. Say what the price is at the moment you use it. It is what the last investors paid, it is negotiated rather than traded, and it moves in both directions. That sentence costs nothing now and is the only thing that makes next year’s conversation survivable.
The percentage decline is the honest hard part. Everyone’s percentage falls with every round including yours, and saying so plainly is better than allowing someone to discover it and conclude it was concealed.
2. The comparisons form whether or not you allow them
The second lens explains why a confidentiality instruction is not a policy, it is a wish.
Grant information spreads through a team the way anything spreads through people who talk to each other: by transmission between individuals already in contact, at a rate set by how connected they are and how much they want to discuss it.2 Compensation is highly transmissible, because it is useful to the receiver and interesting to the sender, and it spreads in a distorted form, because people report their own generously and other people’s approximately.
Your alternative to a stated rule is not secrecy. It is a rumoured rule that is less fair than the one you actually used.
The practical response is to publish the basis rather than the amounts. Not who got what, which most teams do not want, but how the decision was made: refreshes happen at these intervals, they are sized against these factors, here is the band for each level. That gives every private conversation the same reference, and the private conversations will happen anyway.
Assume the number that spreads is wrong in a predictable direction. People round their own grant up and quote it in whichever unit flatters it, so the version circulating is usually the largest true framing of the largest grant. You cannot correct that by denying it, because denying it confirms who received what. You can defeat it by having published the band first. A rumoured figure outside a published band is checkable. The same figure in a company with no published band is simply believed.
If you cannot publish bands yet, publish the intervals and the factors alone. Refreshes are considered at twenty-four months, sized on scope of role, time in seat and the last performance cycle, and never negotiated individually. That removes most of the guesswork and commits you to nothing about amounts. It is the version a five-person company can write in an afternoon, and it tightens into bands at the next round.
The failure mode of the discretionary alternative is worth naming because it is invisible while it is forming. When refreshes are individually negotiated with no published basis, the people who ask receive more, which over two cycles produces an allocation correlated with willingness to negotiate rather than with contribution. Nobody chose that outcome and everybody can see it afterwards.
A stated basis also protects you in the specific case where you cannot be generous. A refresh that is small because the company cannot afford more is survivable when the rule is visible. The same grant, unexplained, is read as a judgement about the person.
3. Be honest about what it is worth
The third lens is the one that determines whether the other two land, and it asks you to say something uncomfortable out loud.
Startup equity is worth almost nothing in most outcomes and a great deal in a small number of them. The value is concentrated in the tail, which means the expected value is not the typical value and the median outcome for any individual grant is close to zero.3
Most equity conversations are conducted as though the grant has a definite worth. It does not, and employees who have been through this before know it, which means the confident version of the conversation costs you credibility with exactly the people whose judgement you most want.
The honest version is more useful and not much harder. This is what it is worth at the last round price. That price is a negotiated number, not a market. Here is what it becomes in an outcome we would consider good, here is what it becomes if we are acquired for a modest amount, and here is the most likely case, which is that it is worth nothing. The reason to hold it is the first branch.
There is a distinction the whole conversation tends to blur, and it decides whether the grant is worth anything to the person holding it. A grant is usually not shares, it is an option to buy shares at a strike price, and the option is real only if the holder can find the exercise cost in cash on the day it is due. An engineer in Kampala with 6,000 options at a USD 1 strike is being asked for USD 6,000, within ninety days of leaving, on a share nobody will buy from her. For that person the tail is not a small chance of a large gain. It is unreachable, and the grant is a discount on her salary that nobody has priced.
Two fixes are cheap and neither needs a new round. Extend the post-termination exercise window from ninety days to several years, which costs the company nothing today and turns a grant that expires on exit into one that can be held. And state the exercise cost and the likely tax treatment during the refresh conversation, in the currency the person is paid in, on the same page as the number. A founder who has never mentioned the strike price has described an asset the employee cannot buy.
There is a further reason to take the comparison seriously rather than to manage around it. Perceived unfairness is not a private disappointment that resolves itself. People respond to it, at a cost to themselves, in ways that show up as reduced discretionary effort and eventually as an exit.4 An unfair-looking grant is not a mood to be waited out, it is a change in behaviour you have purchased.
Two things follow that founders resist. Equity should not be presented as a substitute for cash to someone who cannot absorb the risk, because for them the tail is unreachable and the grant is a discount on their salary. And a refresh conversation is a good moment to ask whether the person would rather have cash, because for some of your team the answer is yes and pretending otherwise wastes equity that someone else would value properly.
What the three say together
- Give them the reference before the number: original grant at the last round price, this grant on the same basis, total.
- Publish the basis, not the amounts. Intervals, factors, bands. The private conversations will happen regardless.
- State the percentage decline yourself rather than letting it be discovered.
- Be honest about the tail, and ask who would rather have cash.
Where they disagree
The reference argument and the honesty argument pull against each other, and the tension is real.
Reference dependence says frame the grant favourably: choose the comparison that makes it land well, because the number will be evaluated against something and you may as well supply it. The concentration argument says that any framing which makes equity sound valuable is overstating the median case, and that a founder who frames well and is later proved wrong has spent something more durable than a retention conversation.
The resolution is to be precise about the frame and honest about the distribution, which is available and is what good funds do with their own investors. Use the last round price because it is a real number, say explicitly that it is a negotiated price rather than a market one, and give the range of outcomes rather than a point. That is a favourable framing and a true one, and the two are only incompatible if the frame you wanted required a number you cannot defend.
What none of them contain
None of the three handles the person who is leaving anyway. A refresh is often an attempt to retain someone who has already decided, and equity vesting over four years is a weak instrument against a decision made last month. The models describe how the grant is perceived, not whether perception is the binding constraint.
None of them prices what refreshes do to your own dilution over time. Each cycle is a decision that feels like retention and is also an issuance, and a company that refreshes generously for three cycles without a pool plan discovers the total at the worst moment, which is during a fundraise.
And one property the ensemble will not produce: the conversation itself carries information beyond the number. How much you know about what the person actually did, whether you can name a specific thing, whether you seem to have thought about them individually. People report those details years later and rarely report the size of the grant, which suggests the meeting is doing something the arithmetic cannot see.
The one action that survives the ignorance: before any refresh conversation, write the basis you are using in four lines and check it against all four people at once. If the basis produces an allocation you would not defend if it were published, the basis is not what you are actually using, and the real one is worth looking at.
Who has to move
The founder, and before the first refresh rather than during it, because a basis introduced after the first round of individual negotiations is read as a correction to something. The instinct is to handle each conversation separately and generously, which produces four defensible decisions and one indefensible pattern. The cheapest first test is to write the basis and apply it to everyone on paper before speaking to anyone, which takes an hour and surfaces the comparisons in private rather than in a meeting.
Sources and notes
- Sanjit S. Dhami, The Foundations of Behavioral Economic Analysis, Oxford University Press. Reference dependence, under which outcomes are evaluated as movements from a comparison point rather than in absolute terms, and the asymmetry between losses and gains relative to that point, are developed in the part on behavioural decision theory. Used in section 1 for the direction only. No loss-aversion coefficient is quoted, because the commonly cited figure is contested and varies with domain and stake size.
- Albert-László Barabási, Network Science, Cambridge University Press. Spreading processes on networks, and the dependence of propagation on existing contact structure rather than on instruction, are developed in the chapters on spreading phenomena. Used in section 2 for the claim that compensation information circulates regardless of policy, and in a distorted form because each transmission is a self-report.
- Albert-László Barabási, Network Science, Cambridge University Press, on heavy-tailed distributions and the divergence between mean and median. Used in section 3 for the general property that value concentrated in a tail makes the expected value a poor description of the typical case. The stronger scale-free label is deliberately avoided, following Anna D. Broido and Aaron Clauset, Scale-free networks are rare, Nature Communications 10, 2019, article 1017, https://www.nature.com/articles/s41467-019-08746-5, which reports that only a small minority of nearly one thousand network datasets show the strongest evidence for it.
- Jeffrey Carpenter and Andrea Robbett, Game Theory and Behavior, MIT Press. Inequity aversion and reciprocity, and the experimental finding that people reject allocations they regard as unfair even at a cost to themselves, are developed in the chapters on social preferences. Cited for the mechanism underlying section 2: the comparison is not idle curiosity, and a grant perceived as unfair produces a behavioural response rather than only a private disappointment.
A note on a number this article does not give. There is no standard refresh size or interval. The figures circulating come from a market with different dilution norms and a different cost of capital, and they are quoted at founders as findings. Decide the pool first, write the basis, then divide.
Joshua Agonya Pi’Rwot, Founder.