The forecast that disagrees with itself
Every Monday morning since at least 2017, Somos has published six different forecasts for when we'll run out of toll-free numbers — based on six different historical windows: the last year, the last 2 years, 3, 5, 7, and 9 years. They almost never agree. This page shows every one of them, week by week, since we have records.
What Somos thinks today
Six concurrent forecasts published in the latest report (week ending 2026-04-25). Same data, six different answers depending on how far back you look.
| Window | Monthly rate of change | Months to exhaust | Predicted exhaust |
|---|---|---|---|
| 12-month window (1 yr) | 329,685 | 22 | 2028-02-01 |
| 24-month window (2 yr) | 158,945 | 46 | 2030-02-01 |
| 36-month window (3 yr) | 131,977 | 55 | 2030-11-01 |
| 60-month window (5 yr) | 66,342 | 110 | 2035-06-01 |
| 84-month window (7 yr) | 64,689 | 112 | 2035-08-01 |
| 108-month window (9 yr) | 48,261 | 151 | 2038-11-01 |
How predictions have evolved
X axis: when the forecast was made. Y axis: the year it predicted exhaust would occur. The 12-month line (red) is skittish — it lurches around with every week's noise. The 9-year line (deep blue) barely moves. The dashed horizontal line shows today: any forecast above it predicts exhaust still in the future, any forecast below it has already missed.
- 12-month window (1 yr)
- 24-month window (2 yr)
- 36-month window (3 yr)
- 60-month window (5 yr)
- 84-month window (7 yr)
- 108-month window (9 yr)
Forecasts that have already passed
Every prediction whose date is now in the past — sorted by predicted-exhaust date, oldest first. The fact that you're reading this on a working internet connection where toll-free numbers still exist tells you whether each forecast was right.
| Predicted exhaust | Forecast made (week ending) | Window |
|---|---|---|
| 2024-03-01 | 2017-11-04 | 108-month window (9 yr) |
| 2024-05-01 | 2017-11-11 | 108-month window (9 yr) |
| 2024-05-01 | 2017-11-18 | 108-month window (9 yr) |
| 2024-05-01 | 2017-10-28 | 108-month window (9 yr) |
| 2024-07-01 | 2017-11-25 | 108-month window (9 yr) |
| 2024-07-01 | 2017-12-02 | 108-month window (9 yr) |
| 2024-07-01 | 2017-12-16 | 108-month window (9 yr) |
| 2024-07-01 | 2017-12-23 | 108-month window (9 yr) |
| 2024-09-01 | 2017-11-04 | 84-month window (7 yr) |
| 2024-09-01 | 2018-01-06 | 108-month window (9 yr) |
| 2024-09-01 | 2018-01-13 | 108-month window (9 yr) |
| 2024-09-01 | 2018-01-20 | 108-month window (9 yr) |
| 2024-09-01 | 2017-12-30 | 108-month window (9 yr) |
| 2024-12-01 | 2017-10-28 | 84-month window (7 yr) |
| 2024-12-01 | 2017-11-11 | 84-month window (7 yr) |
| 2024-12-01 | 2017-11-18 | 84-month window (7 yr) |
| 2025-01-01 | 2018-02-03 | 108-month window (9 yr) |
| 2025-01-01 | 2018-02-17 | 108-month window (9 yr) |
| 2025-01-01 | 2018-01-27 | 108-month window (9 yr) |
| 2025-01-01 | 2018-02-10 | 108-month window (9 yr) |
| 2025-03-01 | 2017-12-23 | 84-month window (7 yr) |
| 2025-03-01 | 2017-12-02 | 84-month window (7 yr) |
| 2025-03-01 | 2017-12-16 | 84-month window (7 yr) |
| 2025-03-01 | 2017-11-25 | 84-month window (7 yr) |
| 2025-04-01 | 2018-03-17 | 108-month window (9 yr) |
| 2025-04-01 | 2018-03-24 | 108-month window (9 yr) |
| 2025-04-01 | 2018-03-10 | 108-month window (9 yr) |
| 2025-05-01 | 2018-03-31 | 108-month window (9 yr) |
| 2025-05-01 | 2018-04-07 | 108-month window (9 yr) |
| 2025-05-01 | 2018-04-14 | 108-month window (9 yr) |
… plus 43 more.
Why such different answers?
Each forecast is a linear regression. The 12-month line uses only the last year of data. If the last twelve months have been unusually busy (lots of new assignments, few disconnects), it extrapolates that pace forever and gives you a near-term exhaust date. The 108-month line averages across nine years of data, which smooths out the spikes — and so it tends to predict exhaust decades out.
Both forecasts are honest. They just answer different questions. The short window says "if recent trends continue." The long window says "averaged across the historical pattern." Reality usually lands somewhere in between.