Every investor believes their estimates are roughly accurate. Almost none have checked, and the ones who do reliably find a consistent bias of a knowable size.
That is good news, because a consistent bias is correctable with arithmetic. What cannot be corrected is a bias nobody has measured.
Why Estimates Drift One Way
Not random error, which would cancel out over enough deals.
You want the deal to work, and that pressure operates on every soft input simultaneously. The resale figure comes in a little high, the repairs a little light, the timeline a little short.
None of those feel like distortion at the time. Each is defensible on its own. The problem is that they all lean the same way, so a deal with four estimates has four errors pushing in the same direction.
Which is why the correction has to be measured rather than intended. An investor who resolves to be more conservative will be, for two deals.
What to Record
Two numbers per input per deal, and nothing else.
After-repair value. What you estimated, and what it actually sold for.
Repairs. What you estimated, and what the project actually cost, including the items outside the renovation scope.
Timeline. The planned duration and the actual one, from acquisition to sale rather than only the construction period.
Holding costs. Modeled against actual, per holding costs investors forget.
Rent, on holds. Estimated against achieved, and the vacancy assumption against the reality.
Profit. The modeled figure against what actually landed.
Six pairs. A spreadsheet with one row per deal, filled in twenty minutes at the end of a project.
Reading the Pattern
The individual deal tells you nothing. Five deals tell you almost everything.
Look at the direction first. If four of five repair estimates came in low, that is a bias rather than a run of surprises.
Then look at the size. If the average shortfall is a consistent percentage, that is your correction factor, and applying it makes you immediately more accurate without any improvement in judgment.
Then look at the spread. A consistent twenty percent under is easier to correct than an estimate that is sometimes right and occasionally wildly wrong, and a wide spread suggests the problem is the method rather than the calibration.
The most useful output is a short list: I run this percentage light on repairs, this percentage optimistic on resale, and this many weeks short on timeline.
Where the Bias Usually Sits
Common patterns, worth checking yours against.
Repairs light, and light in specific places. Frequently concentrated in the items that were not visible: electrical behind walls, plumbing under floors, and whatever was discovered during demolition.
Timeline short. Nearly universal, and the gap is usually larger on the parts nobody schedules: permits, materials arriving, and the interval between finishing and selling.
Resale optimistic. Usually from comparable selection rather than from market movement, detailed in how to calculate ARV.
Holding costs absent. Not underestimated so much as omitted, particularly the transaction costs at each end.
Rent optimistic on holds. And the vacancy allowance too generous.
Applying the Correction
The mechanical part, and it is simpler than the measuring.
If your repair estimates run a certain percentage light, add that percentage to every estimate as a separate line rather than inflating individual items. Keeping it explicit preserves the line-item accuracy and makes the correction visible, explored in estimating repairs.
If your timeline runs short by a consistent margin, extend every projection by it, and size your financing to the corrected number rather than the optimistic one.
If your resale figures run high, either apply a discount or fix the comparable selection method, since this one usually has a specific cause rather than being general optimism.
Then re-measure after another five deals, because the correction changes the behavior it was measuring.
The Deals You Did Not Do
The harder half, and almost nobody tracks it.
Grading only the deals you did measures your execution and not your judgment. The properties you passed on are equally informative and invisible.
The method is a quarterly look at what happened to the properties you declined. Public records show whether they sold, to whom and for what.
Two patterns worth finding. Several selling close to what you offered means your threshold is set tighter than the market requires, and workable deals are going past you. Several selling well above your figure means your inputs are out of step with what buyers there will pay.
And if most did not sell at all, your judgment was right, which is worth knowing because the doubt about passed deals is corrosive without evidence, discussed in when to walk away.
Grading the Whole Analysis, Not Only the Inputs
One level up, and where the useful strategic finding sits.
Compare modeled profit against actual profit across all deals, and look at the distribution rather than the average.
An investor whose deals cluster near the model has a reliable process. One whose outcomes vary widely has a process that is not doing much, even where the average happens to look acceptable.
Then split by deal type. Frequently the pattern is that one category is consistently accurate and another is not, which tells you where to be more conservative and occasionally which deals to stop doing.
Grading Your Team's Numbers Too
Once anyone else produces estimates, the same calibration applies to them and it is easier to do than for yourself.
An acquisitions person's after-repair figures, a caller's assessment of motivation, a contractor's timeline. Each is an estimate with a track record available.
Contractor timelines in particular are worth tracking by contractor rather than in aggregate. Most run consistently over by a knowable margin, and knowing which one runs closest to schedule is worth more than a small difference in price, described in getting contractor bids.
The way to make this useful rather than adversarial is to share it. A contractor shown that their last four projects each ran a few weeks over usually adjusts, and one who does not has told you something.
Same with an acquisitions person. Showing someone their own calibration is the fastest coaching available, and a conversation about numbers rather than about performance, per training someone to talk to sellers.
The Objection to Doing This
It is uncomfortable, which is the actual reason it does not happen.
Writing down that you estimated a repair figure and spent considerably more is an admission, and doing it six times produces a document that is unflattering.
Two things help. It is private, and its purpose is arithmetic rather than judgment. And the investors who improve fastest are uniformly the ones who look, because the alternative is repeating the same error indefinitely while believing each instance was bad luck.
The framing that works: you are not grading yourself, you are calibrating an instrument. An instrument that reads consistently low is not a bad instrument once you know by how much.
The Deals That Went Badly Are the Most Informative
These deserve a separate look rather than being averaged into the pattern.
For any deal that lost money or made far less than modeled, write a short account: what the estimate was, what actually happened, at what point it became clear, and what the earliest available warning was.
That last question is the valuable one. Most bad deals had a signal, usually at the analysis stage, that was visible and discounted. A comparable that did not quite fit. A repair item that was guessed rather than priced. A timeline that assumed nothing would go wrong.
Reading five of those accounts together produces a list of warning signs specific to how you make decisions, which is more useful than any generic checklist.
And it is worth writing while the deal is fresh rather than at year end, because the reconstruction after six months is considerably kinder to the decision than the facts were, covered in keeping records.
Starting From Nothing
If you have no record of past deals, you cannot reconstruct one reliably, and memory is biased in the same direction as the original estimates.
Start with the next deal. Write the six estimates down before you commit, in a place you will find later, and record the actuals when it closes.
Five deals is enough for a usable pattern. That may be a year, which feels slow, and that is the only route to estimates that are accurate rather than confident.
In the meantime, apply a deliberate conservatism to the soft inputs, since being roughly right about your own bias is better than assuming it does not exist, set out in analyzing a real estate deal.