Zestimate versus after-repair value

A Zestimate is Zillow's estimate of what the home would sell for today, as it is. An after-repair value is the value after the repairs in the report. They answer different questions even when both numbers sit on the same house. A current-value estimate is not an after-repair value. The Zestimate is not the ARV.

The first number describes the house the records describe today. The second number describes the house after the work in the rehab budget. Swapping them changes the offer, the profit, and the refinance check, because the later formulas on this site start from the after-repair value.

What a Zestimate is allowed to answer

Zillow publishes the definition on its Zestimate page. Checked on 2026-11-11. That day the page said the accuracy block was last refreshed 2026-11-08. Two sentences were still there. "The Zestimate® is Zillow's estimate of a home's market value." "It is not an appraisal and can't be used in place of an appraisal." A later answer on the same page says: "It is a computer-generated estimate of the value of a home today, given the available data."

Today is the part that matters for a flip. The estimate follows the home as the data describes it now. The after-repair value is the number the report uses for the house after the repairs. A kitchen that is still on the punch list is not what a current-value estimate is pricing. The ARV prices the sold comps instead, and the rehab budget is the money you plan to spend to reach that repaired house.

The same page says: "Unreported additions, updates and remodels aren't reflected in the Zestimate." Facts you have already reported, such as a finished bed or bath, can move today's estimate. That movement is still an estimate of the home as it sits once those facts are on the record. It is not a price for a rehab that exists only in your scope of work.

Where the after-repair value comes from

The report's ARV is a weighted median of adjusted sold comps. Each comp keeps its adjusted sale price, or the raw sold price when no bedroom or bathroom adjustment applies. The calculator sorts those prices and walks a running weight until the running total reaches half of the total weight. That price, rounded to the nearest dollar, is the ARV. A high sale moves the result only when its weight is large enough to cross that halfway point.

The weights are the recency, distance, and similarity multipliers. The table lives on the guide for how to pull comps. This page does not rebuild that table. The 0 to 1 match score is on the comp similarity guide. Open it when a sale looks close on the map and still should not lead the median.

When a Zillow lookup succeeds, the pipeline stores the Zestimate on the property. The written analysis can mention that stored number. The stored field is not the ARV. A sentence in the narrative that repeats the Zestimate is still the Zestimate. The number the offer uses is the weighted median.

The product also uses Redfin as a comp source. It can store a Redfin estimate on the same property. A Redfin estimate is an automated estimate of current value, not an after-repair value. This page does not quote a Redfin error rate.

What Zillow publishes about accuracy

On 2026-11-11, the Zestimate page said: "The nationwide median error rate for the Zestimate is 1.80% for homes that are on the market and 7.21% for off-market homes." The next sentence said: "In simple terms, this means that for homes currently listed for sale, the Zestimate is within 1.80% of the final sale price half of the time, and within 7.21% half of the time for off-market homes."

Those medians are versus the sale price of the home as it sits, half the time. They are not an error versus an after-repair value, and they are not our confidence score. An on-market figure of 1.80% means half of the listed homes in that national comparison sold within 1.80% of the Zestimate shown before the home went pending. It does not mean a house you still have to repair will sell within 1.80% of either the Zestimate or the ARV.

The same page has a table of metro areas. This guide does not copy that table. A national median is not a promise for the block you are underwriting. If you want the metro rows, open the Zestimate page that was checked. If the two medians change, the sentences on that page are the source. This guide is tied to the check on 2026-11-11, when both figures were 1.80% on the market and 7.21% off the market, and the accuracy block showed a last refresh of 2026-11-08.

Confidence is not a Zestimate error rate

The report shows a confidence score from 0 to 100, and the screen prints a percent sign beside it. The sign means the score is scaled to 100. It does not mean the score is a percent error. It does not mean the ARV is that many percent away from a future sale, and it does not mean the score is Zillow's 1.80% or 7.21%.

The score is the sum of four pieces. The result is clamped between 0 and 100, and then it is rounded to the nearest whole number.

Comp count is the number of comps times 6, capped at 30. One comp is 6 points. Five comps are 30 points. A sixth comp does not add count points. This piece measures how many sold comps entered the median.

Average similarity times 25 is the second piece. The match score runs from 0 to 1, so a full set of perfect matches scores 25. A missing similarity counts as 0.5 in that average. A similarity of zero counts as 0.5 as well. This piece measures how close the comps are on the match score. It does not measure how close their prices are.

The share of comps sold within six months, times 25, is the third piece. A month in that test is 30 days, the same month length the comp weights use. Five comps all inside the window score 25. Three of five inside the window score 15. This piece measures recency of the set. It does not measure the Zestimate.

Price consistency is the fourth piece. Take the standard deviation of the adjusted sale prices and divide by their mean. That ratio is the coefficient of variation. The points are the maximum of 0 and 20 minus that coefficient times 100. A tight cluster scores near 20. A wide spread can score 0 even when the other three pieces are full. If the mean is 0, the coefficient is treated as 1 and this piece is 0. The standard deviation divides the squared gaps by the count of comps, not by the count minus one.

With no comps, the calculator does not run those four pieces. It returns an ARV of 0 and a confidence of 0.

Here is an invented set, so the arithmetic can be checked by hand. Five adjusted sale prices: $180,000, $190,000, $200,000, $210,000, and $220,000. Each similarity is 0.80. Each sale is inside the six-month window. These prices are not a closed deal.

Comp count is the minimum of 5 times 6 and 30, which is 30. Average similarity times 25 is 0.80 times 25, which is 20. The recent share times 25 is 25. The mean is $200,000. The squared gaps from the mean sum to 1,000,000,000. Divided by 5, the variance is 200,000,000. The standard deviation is the square root of 200,000,000, about $14,142. The coefficient of variation is about 0.0707. Times 100 that is about 7.07. Price consistency is about 20 minus 7.07, which is about 12.93. The sum is about 87.93. Clamped to the 0 to 100 range, it is still about 87.93. Rounded, the score is 88.

An 88 on that invented set means the five prices are fairly close, recent, and numerous. It does not mean the ARV is within 12 percent of a sale, as if 100 minus 88 were an error band. It does not mean a Zestimate with a different figure is off by 12 points. A percent error against a sale price of the home as it sits is the median Zillow published, quoted above.

What to do with the two numbers

Use sold comps for the offer. Type the weighted median in as the after-repair value, then take the repair cost off that value with the other reserves. The offer page is the maximum allowable offer calculator. The ARV calculator is where a full report builds that weighted median from the comps it found.

Treat a Zestimate that disagrees with the comps as a reason to read the comps, not as a reason to replace the ARV. A higher Zestimate is not extra profit. A lower Zestimate is not a reason to drop a tight set of recent sold comps. Both numbers can be on the screen and still answer different questions. The same rule applies to a Redfin estimate stored on the property. It can say what an automated model thinks of today's value. The offer still starts from the sold comps.

Questions

Open the sample analysis after you sign up. The sample is an example report with invented figures. The sample analysis is created after signup.