Results and benchmarks · updated 30 October 2026
What our processes take, measured.Medians with the number of runs behind them. Estimates are marked, and what is not measured yet says so.
By process
Every process, by hand and with Titleman
The label says what kind of figure each Titleman time is. Time by hand is always an estimate for a typical deal, not a measurement of your team.
| Process | For | By hand (estimate) | With Titleman | Runs behind it | Latest 30 days |
|---|---|---|---|---|---|
| Screening an incoming OM | Investors, lenders, brokers | about 2 hours | about 30 to 40 minutes of engine time, about 12 minutes of your check Estimate | no clock yet | |
| Underwriting a deal | Investors | about 22 hours | about 58 minutes (the middle half of runs took 35 to 74 minutes) Measured | 44 runs | about 74 minutes, median of 114 runs |
| Credit file and DSCR stress | Lenders | about 22 hours | about 1 hour Estimate | to measure | |
| Valuing a property (value opinion) | Brokers, owners | about 12 hours | about 45 minutes (the middle half of runs took 34 to 62 minutes) Measured | 66 runs | being re-cut |
| Valuing a list of addresses | Brokers, owners | about 12 hours per address | about 28 minutes for a list (the middle half of runs took 21 to 38 minutes) Measured | 138 runs | about 19 minutes, median of 39 runs |
| What may be built on a site | Developers | about 1 day | about 27 minutes Measured | 76 runs | about 43 minutes, median of 49 runs |
| Comparable sales set | Everyone | 1 to 2 hours per set | about 18 minutes Measured | 76 runs | about 25 minutes, median of 27 runs |
| Rent analysis | Owners, investors | no estimate | about 16 minutes Measured | 90 runs | about 23 minutes, median of 19 runs |
| Data room check before signing | Investors, lenders | 4.5 hours of reading, without cross-checking | about 78 minutes with full cross-check Example | 1 test case (6 red flags in 46 items) | |
| OM, IC memo and investment deck | Brokers, developers | about 30 designer hours for an OM | about 1 hour from the same run Estimate | to measure | |
| Finding reasons to call on a territory | Brokers, lenders, investors | about 1 day a week | a weekly list; every row has its reason, date and owner Estimate | not a time claim | |
| Who owns it and who decides | Brokers, investors | to measure | to measure To measure | to measure | |
| Portfolio monitoring | Owners, lenders | no estimate | a property passport in about 4 minutes To measure | few runs |
A human working with an AI assistant is a third column. It is not here until our test on five incoming OMs, three ways, has run.
Accuracy
Value Opinion: 13% median error on 30 sales
Median error 13.0% against the recorded sale price; the price fell inside our stated range in 14 of 30 sales. Blind backtest, February 2025 to August 2026, measured on 28 October 2026. Thirty sales is a small sample, so read the property types as indications.
Closest
| Multifamily | 4.4% median error | 3 sales |
| Industrial | 2.3% | 2 sales |
| Residential | 9.9% | 3 sales |
| Office | 10.4% | 4 sales |
Weakest
| Land | 56% median error | 2 sales |
| Medical office | 40% | 3 sales |
| Restaurant | 27% | 3 sales |
These are sales that carry a going business or an assemblage of parcels. We show them because a number without its weak spots is not a number to trust.
- Sales valued
- 30 properties with a recorded sale, February 2025 to August 2026
- Median absolute error
- 13.0% (average 19.2%)
- Sale price inside our stated range
- 14 of 30 (47%)
- Within 10% of the sale price
- 13 of 30 (43%)
- Within 20% of the sale price
- 20 of 30 (67%)
How we know
Method
- Titleman time is the median time from the request to the delivered files, over completed production runs created in the 30 days to 18 October 2026. One clock for every product. The middle half is the range between the 25th and the 75th percentile.
- Time by hand is an estimate for a typical deal: the figures Titleman publishes in its deal-hours tool, or a rule of thumb. It is not a measurement of your team and it is marked as an estimate wherever it appears.
- Time with Titleman is not the time of the whole process: the output is a draft and your professional reviews and signs it. We assume that check is 10% of the hand hours and say so; we will replace the assumption with a measurement.
- Each row also shows the same clock read again for the 30 days to 29 October 2026, so that a slower or faster month is visible. The value opinion is not shown there until its cut is made comparable.
- We compare Titleman with the process, not with other companies.
- A row marked «to measure» has no number because we have not measured it. A row marked «example» is one real run, not a median.
- We are testing a human working with an AI assistant on five incoming OMs, three ways, with the answer key written in advance. That column is not here until the test is done.
- Backtest. For each property the engine produced a standard Value Opinion as of the day before its recorded sale. The sale price and date were not given in the request.
- Backtest. The request told the engine to leave out everything dated after the valuation date, including the property’s own sale. That is an instruction, not a filter on the data: in 25 of the 30 answers the engine saw the later sale on record and listed it as not used.
- Backtest. The point value is the figure the Value Opinion states; where it states only a range, we took its midpoint (one sale). Error is the stated value minus the recorded price, divided by the price.
- Backtest. All 30 valued sales are counted. Properties where the engine gave no value (wrong parcel in the register, a stopped run) were set aside and are not scored; none was dropped for a poor result.
- Backtest. Thirty sales is a small sample and every property type has two to four of them: read the type rows as indications. We will re-measure on a larger set and replace these figures.
Updated 30 October 2026. Titleman is not a broker, appraiser or law firm; every output is a draft for a licensed professional to review and sign.