Independent Living Resource Center San Francisco v. Lyft, Inc.

• Court: District Court, N.D. California
• Decided: September 1, 2021
• Docket: 3:19-cv-01438
• Status: Unknown

Opinion

1 2 3 4 5 6 UNITED STATES DISTRICT COURT 7 NORTHERN DISTRICT OF CALIFORNIA 8

10 INDEPENDENT LIVING RESOURCE CENTER SAN FRANCISCO, a 11 California non-profit corporation, No. C 19-01438 WHA JUDITH SMITH, an individual, JULIE 12 FULLER, an individual, SASCHA BITTNER, an individual, TARA 13 AYRES, an individual, and FINDINGS OF FACT AND COMMUNITY RESOURCES FOR CONCLUSIONS OF LAW 14 INDEPENDENT LIVING, a California non-profit corporation, 15 Plaintiffs, 16 v. 17 LYFT, INC., 18 Defendant. 19

20 21 INTRODUCTION 22 In this action under the Americans with Disabilities Act (ADA), 42 U.S.C. 12181, et seq., 23 the issue is whether a rideshare platform company discriminates against motorized wheelchair 24 users by failing to offer wheelchair-accessible rides. Specifically, trial addressed the following 25 question: Has defendant discriminated by refusing to adopt plaintiffs’ proposed modification 26 to defendant’s policies, practices, or procedures? The answer is no. For the following reasons, 27 plaintiffs have not met their burden to show that the proposed modification is reasonable, so no 1 Plaintiffs Judith Smith, Julie Fuller, Sascha Bittner, and Tara Ayres are disabled, use 2 motorized wheelchairs, and live in the Bay Area. Plaintiffs Independent Living Resource 3 Center and Community Resources for Independent Living represent Bay Area-resident 4 wheelchair users. Lyft, Inc. operates a transportation service on an app-based platform, a 5 service that matches riders with drivers. 6 In March 2019, plaintiffs filed the instant complaint as a putative class action, alleging 7 violations of the ADA and seeking injunctive and declaratory relief. The complaint explains 8 that Lyft offered “wheelchair access” mode only within San Francisco, and with greater 9 restrictions (e.g. limited hours) than its standard mode. Plaintiffs seek an expansion of 10 wheelchair access mode to Alameda and Contra Costa Counties, as well. Wheelchair means 11 motorized wheelchair. Conventional wheelchairs that are foldable are already accommodated 12 by Lyft in its standard mode. The relief requested would not require Lyft to purchase 13 wheelchair-accessible vehicles (WAVs), but instead to modify its incentive structures to 14 guarantee wheelchair-accessible rides on its platform. 15 Plaintiffs moved for class certification. A March 2020 order herein denied plaintiffs’ 16 motion without prejudice so that plaintiffs could take another stab at defining the class. In 17 September 2020, plaintiffs renewed their motion to certify a class, which was ultimately 18 denied. 19 Simultaneously, both sides cross-moved for summary judgment. The November 2020 20 summary judgment order denied defendant’s motion and granted in part and denied in part 21 plaintiffs’ motion. That order held (Dkt. No. 92 at 9, 12): 22 [W]ithout more specific evidence regarding, for example, how a different combination of these proposed methods would be 23 implemented in the Bay Area, the supply of WAV drivers that could 24 be deployed, or the financial costs of doing so, it is unclear whether the proposed modifications are reasonable . . . . We will have to hold a 25 trial on the main issue of whether the proposed modifications, specifically a rental model or a combination of the models, are 26 reasonable. 27 Three pretrial conferences proved necessary to sift through the parties’ motions in limine 1 At the close of evidence, the parties submitted almost 300 proposed findings. The 2 findings of fact below represent those necessary to address the main conclusions of law. For 3 clarity and ease in presentation, this order will discuss some additional findings with its 4 conclusions of law. All declaratory statements are findings. To the extent, however, that any 5 proposed finding was expressly admitted by the responding party in the recent round of 6 proposals and responses, this order hereby adopts the proposal (to the extent expressly 7 admitted). This order need not cite the record and will do so only where it will likely assist our 8 court of appeals, that is, as the exception and not as the rule. 9 FINDINGS OF FACT 10 1. Lyft is a San Francisco-based company. It launched an on-demand ridesharing 11 marketplace in 2012. The marketplace is generally available, including throughout the three 12 counties at issue (i.e. Alameda and Contra Costa counties, and the City and County of San 13 Francisco). 14 2. In these three counties, Lyft offers the following ride “modes,” among others: (a) 15 Standard (its classic rideshare option), (b) XL (larger vehicles for up to six riders), (c) Lux 16 (high-end or luxury vehicles), (d) Lux Black (high-end or luxury black car), and (e) Lux Black 17 XL (high-end or luxury black SUV for up to six riders). All of these modes can and do 18 accommodate foldable wheelchairs. The issue here concerns motorized wheelchairs. 19 3. Lyft uses financial incentives to influence drivers to join the platform and to alter 20 their driving, including the modes in which they drive. 21 4. At the time of trial, Lyft offered WAV service, a.k.a. “access mode,” in nine U.S. 22 cities. Again, “WAV” refers to motorized wheelchairs. No regulation requires the service in 23 San Francisco or Los Angeles, but in the remaining seven cities, Lyft offers the service 24 because local governments made it a condition of doing business. Lyft does not currently offer 25 WAV rides in Alameda or Contra Costa Counties. 26 5. Lyft chose to begin offering WAV services in San Francisco and Los Angeles 27 because state regulators established a ten-cent surcharge on all rides in California with 1 Angeles and San Francisco as a pilot to try to recover from the surcharge fund. This fund will 2 be described below. 3 6. In developing WAV programs, Lyft has used trial and error in some of its markets 4 to determine the business model (or combination of models) that will, if possible, allow it more 5 effectively to provide WAV service there. This trial and error is sometimes referred to as an 6 “iterative process” of developing WAV programs. 7 7. In all cases, WAV customers pay the same rate for rides as do non-WAV 8 customers. 9 8. Lyft currently uses three primary models, sometimes in combination, in different 10 markets around the country: the “organic independent contractor model,” the “rental model,” 11 and the “partner model.”1 12 9. The “organic independent contractor” model involves recruiting drivers to the 13 Lyft platform who already own their vehicles. This is the usual model with which most 14 customers are familiar. 15 10. Currently, however, Lyft does not allow such drivers to provide WAV rides in the 16 Bay Area. 17 11. Lyft has not attempted to recruit organic independent contractor (IC) WAV 18 drivers in the Bay Area and does not intend to because, among other reasons, it believes the 19 service would prove unreliable (Tr. 456–57). 20 12. Under the “rental model,” which Lyft has tried in New York City and 21 Philadelphia (where local regulators require WAV services), Lyft works with a rental vehicle 22 company, which purchases and maintains WAVs. Would-be drivers, in turn, rent the vehicles 23 in order to drive for Lyft. Lyft subsidizes these rentals. 24 25 26

27 1 At the summary judgment stage and at the final pretrial conference, the district judge asked counsel to present evidence concerning Uber’s WAV practices. At trial, neither side did so. This footnote is included because our court 1 13. Where Lyft uses the rental model, Lyft allows drivers operating WAV rentals to 2 drive in any mode, for any length of time, and whenever they prefer, just the same as standard 3 Lyft drivers. This means rental WAV drivers may pick up standard fares, which they do. 4 14. In the “partner model” or “W-2 model,” Lyft contracts with third parties to 5 provide WAV services for Lyft. In this model, Lyft pays an existing transportation company to 6 provide WAV rides using the Lyft app. The drivers remain employees of the third party (hence 7 the “W-2” label). These drivers pick up WAV customers only. Lyft uses this model in most 8 regions where it provides WAV services. 9 15. This model, however, ranks as the most expensive, with a true cost of $1500 per 10 ride in San Francisco. It is expensive because the WAV driver picks up only WAV customers 11 and otherwise must wait for the next WAV call. In practice, this has meant that partner drivers 12 receive very few calls. They averaged roughly two rides per day in San Francisco (TX 177 at 13 018, Fig. 5; Tr. 596). 14 16. The partner model is also “operationally difficult” to run. For example, Lyft must 15 closely monitor the wait times involved to document Lyft’s eligibility for a state offset, as now 16 described. 17 17. California Public Utilities Code Section 5440.5 requires transportation network 18 companies (TNCs), like Lyft, to charge no less than a five-cent surcharge on each Lyft ride 19 given in California, with the surcharge amount to be set by the California Public Utilities 20 Commission (CPUC). The CPUC has set the surcharge at ten cents per ride. The surcharges 21 flow into the “TNC Access Fund,” which the CPUC earmarks for accessible transportation 22 services. The CPUC administers the fund, offering quarterly opportunities for TNCs to seek 23 reimbursements if they meet certain requirements. The fund is not limited to providing 24 reimbursements to rideshare companies. The fund may be used in other ways to promote 25 accessible transportation. 26 18. A rideshare company like Lyft may apply to recover “offsets” from the fund up to 27 the amount it paid into the fund for that quarter and for that county. The CPUC has set a 1 variety of performance requirements, including the “offset time standard,” which specifies 2 benchmarks for WAV riders’ maximum wait times (a.k.a. ETAs). 3 19. The CPUC also requires that all WAV drivers receive training, and that WAVs 4 pass a safety inspection. WAV drivers must use special straps to secure passengers and their 5 wheelchairs for safety. Ramps or lifts must be used properly to bring the WAV passenger into 6 and out of the WAV vehicle. 7 20. The CPUC promulgates regulations through rulemaking. All proposed regulation 8 proceeds on a “track.” A track may produce an interim or permanent rule. The CPUC may 9 contemplate multiple proposals simultaneously, on multiple tracks, all on the same topic of 10 regulation. The rulemaking on the CPUC performance standard criteria is set to end in 2025. 11 21. From 2019Q3 through 2020Q1, the CPUC merely required Lyft to show a 12 quarter-over-quarter improvement in the median WAV wait times in order for Lyft to draw 13 down on the Access Fund during those quarters. 14 22. As of 2020Q2, however, the CPUC added additional requirements for a total of 15 three criteria necessary to qualify for offsets, i.e. partial reimbursements, from the Access 16 Fund. 17 23. The first criterion is the offset time standard, which the CPUC set as an interim 18 standard in its Track Two rulemaking, in 2020Q2. This standard varies by county. In San 19 Francisco, either 50% of rides each quarter must have wait times within 15 minutes (the “Level 20 1” offset standard) or 75% of rides must have wait times within 30 minutes (the “Level 2” 21 offset standard). In Alameda and Contra Costa counties, 50% of WAV rides each quarter must 22 demonstrate wait times within 25 minutes or 75% of its rides within 50 minutes (TX 14 at 23 086). While a Track Three decision came down in 2021Q2, it left the offset time standard 24 unchanged (TX 76 at 060–61). 25 24. The second criterion is continued quarter-over-quarter improvement in wait times. 26 Specifically (TX 14 at 088–89, emphasis added): 27 To demonstrate improved level of service for offset eligibility, a achieved either a Level 1 or Level 2 Offset Time Standard for a 1 quarter in that implementation year. If a TNC received an offset in 2 the prior quarter, the TNC must achieve an Offset Time Standard that exceeds the percentage achieved in the prior quarter in either, a 3 Level 1 or a Level 2 Offset Time Standard. 4 For example, if a TNC’s median wait time meets the offset time standard in Level 1 in a given 5 quarter, it need not have also improved its median wait time in Level 1 — it could instead have 6 improved its performance in Level 2 (Tr. 475–77; TX 14 at 086–87). 7 25. The third criterion requires meeting a “Trip Completion Standard,” i.e. a TNC 8 must have completed “more wheelchair-accessible vehicle trips compared to the previous 9 quarter in that county” or must have accomplished an increase in the ratio of requested WAV 10 rides to completed WAV rides. This requirement took effect beginning in 2021Q2, with the 11 publication of the CPUC’s Track Three decision. In that decision, the CPUC postponed setting 12 actual required ratios for the Trip Completion Standard pending further study. 13 26. The CPUC also developed an even higher performance standard (the “exemption 14 standard”), which, if achieved, relieves TNCs from paying the ten-cent surcharge in the first 15 place, for four quarters. The exemption standard requirements are more stringent. They 16 require provision of 80% of each quarter’s WAV rides within 16 minutes in San Francisco, 17 80% within 20 minutes in Alameda County, and 80% within 40 minutes in Contra Costa 18 County (TX 14 at 047–49). To date, Lyft has not met this standard in San Francisco. 19 27. The CPUC is currently considering updating wait-time and trip-completion 20 requirements in its ongoing Track Four rulemaking. There is no expectation of a ruling soon. 21 28. In Lyft’s proposal for Track Four, it explained the problems of the quarter-over- 22 quarter time improvement. It also proposed to the CPUC that TNCs would be required to 23 provide 54.5% of rides in San Francisco within 15 minutes in 2021Q3 and improve on that 24 standard until they would be required to provide 80% of rides within 15 minutes in 2025Q4. 25 Additionally, TNCs would have to provide 82% of WAV rides within 30 minutes in 2021Q3 26 and improve on that standard until they would be required to provide 99% of rides within 30 27 minutes in 2025Q4. Finally, in Lyft’s proposal, entities would have to show a ride-completion 1 rate of 71.5% and ride acceptance rate of 81.5% for 2021Q3, and eventually show a ride 2 completion rate of 80% and ride acceptance rate of 90% in 2025Q4 (TX 97 at 6). 3 29. The CPUC may not (or may) accept Lyft’s proposals, and future offset 4 requirements may become more severe, not less. 5 30. Section 5440.5(e) of the California Public Utilities Code provides that the Access 6 Fund program, including the surcharge, will sunset in January 2026. 7 31. In July 2019, Lyft piloted a WAV partner model in San Francisco. Local law did 8 not require it to do so. It contracted with First Transit, Inc., a private provider of WAV rides. 9 32. Initially, Lyft’s contract with First Transit provided for five vehicles to offer rides 10 between seven in the morning and midnight, daily, in San Francisco. First Transit continues to 11 operate the pilot 17 hours per day. The pilot used the five vehicles between July 2019 and 12 midway through the pandemic. During the pandemic, however, Lyft observed that both traffic 13 and demand for rides decreased in San Francisco. Simultaneously, Lyft found it could better 14 predict the locations of ride requests, place partner vehicles near those areas, and thereby 15 decrease the time it took a driver to reach the would-be rider. Lyft renegotiated its agreement 16 with First Transit “in the middle of the pandemic” and reduced the W-2 vehicles driving in San 17 Francisco from five to two (Tr. 215). 18 33. To repeat, the CPUC did not mandate specific wait times for TNCs to draw down 19 on the Access Fund in 2019Q3 or in 2020Q1. Instead, it required quarter-over-quarter 20 improvement. Starting in 2020Q2, however, the CPUC began to require wait times of 50% of 21 rides within 15 minutes or 75% of rides within 30 minutes (in addition to other requirements 22 like quarter-over-quarter improvement). 23 34. The First Transit program operates at a cost of $58.62 per hour per vehicle (Tr. 24 408). 25 35. The First Transit program has seen a relatively flat demand for rides in San 26 Francisco; it has averaged approximately 60 rides per month since its inception in July 2019 27 and hit a peak of 125 per month in February 2020 (Tr. 153, 821). 1 36. Ina 12-month period (2019Q4 through 2020Q3), Lyft paid First Transit $1.5 2 million. During the same contract period, Lyft provided 955 WAV rides in San Francisco, for 3 an average per-ride cost of $1500 to Lyft (before any offset) (TX 177 at 018, Fig. 5; Tr. 78). 4 37. Lyft requested CPUC offset funds for its San Francisco pilot during four quarters 5 (2019Q3 through 2020Q2). During that period, Lyft qualified for and received offset funds in 6 the amount of $1.47 million. Therefore, during nearly the same period, Lyft received from the 7 a CPUC almost enough to cover the four quarters of the WAV program (see below, TX 98): 8 9 as ee

a [las aise aa agrome| a1 2020, | 02. 2020" | Tata Francisco ? □□□ a SS ccc 13 14 Anticipating that it would not meet offset standards in 2020Q4, Lyft did not apply for 3 15 reimbursement for that quarter. At the time of trial, Lyft’s request for CPUC offset for 2021Q1 16 remained pending. 3 17 38. In contrast to San Francisco, the East Bay counties have contributed, via the 18 surcharge, far less to the Access Fund. In 2019Q4, Alameda County contributed $240,000, and 19 Contra Costa County paid just $60,000 to the Access Fund. These were the only pre-pandemic 20 contribution figures that emerged at trial for the two East Bay counties (Tr. 564). All of these 21 payments were collected by Lyft via the surcharge. 22 39, Isabella Gerundio, Program Manager for Wheelchair-Accessible Vehicles at Lyft, 23 testified about Lyft’s WAV program. She described some of the challenges associated with the 24 partner model: first, ensuring that W-2 WAV drivers are awake and alert and quick to respond 25 to ride requests; second, ensuring availability of rides during shift changes, since shift changes 26 take WAV drivers offline for a period of time and increase wait times overall; and third, since 27 First Transit does not own surplus WAV vehicles, and requires two to three months to procure 28 and ready anew WAV, the model requires Lyft to guess the number of WAVs that demand

1 will support. If Lyft underestimates, the model will fall short of demand until First Transit can 2 onboard more vehicles (Tr. 438–41). 3 40. While Lyft began its partnership with First Transit in July 2019, it began planning 4 for a more cost-effective alternative that would enable it to expand to serve San Francisco, 5 Alameda, and Contra Costa Counties. In 2019, Lyft’s “Science” team prepared an internal 6 memo detailing a pilot proposal to offer wheelchair access mode in the three counties using the 7 rental model. Although the parties refer sometimes to the proposal as the “rental pilot” or the 8 “rental program,” the actual idea would have involved Lyft partnering with First Transit to 9 provide five W-2 vehicles and to add sixty-five rental vehicles through a partnership with 10 Hertz. 11 41. At trial, this was known as the “65+5” proposal. 12 42. This proposal mirrored, in part, Lyft’s WAV services in New York City. The 13 65+5 proposal, however, would have utilized just two of the three Access models to provide 14 WAV services, while New York City uses all three. An additional difference is that New York 15 City’s regulators have set stricter wait-time standards than any other regulator. Lyft has 16 generally achieved those standards: in June 2019, Lyft successfully met its requirements to 17 provide 60% of WAV rides in under 15 minutes and 90% in under 30 minutes (Tr. 144). As of 18 June 2021, however, Lyft was unsure if it could meet New York City’s updated requirements 19 to provide 80% of its WAV rides within ten minutes and 90% within fifteen minutes, however 20 (Tr. 116, 124–25, 375). From 2019 to 2020, Lyft spent approximately $200 per ride to provide 21 WAV services in New York City, though costs have decreased over time (Tr. 150, 82). 22 43. As recently as May 2020, Lyft had hoped the rental WAV program would work 23 around the country. Lyft hoped the model would prove less expensive than the partner model, 24 while still meeting demand and CPUC offset standard. 25 44. The 2019 memo proposed serving all three counties. Lyft would have contracted 26 with Hertz to provide ten WAVs at first (eventually ramping up to 65) for rental directly to 27 Lyft’s ICs. The contract for 65 vehicles would have cost Lyft one million dollars per year. 1 shared the financial risks of the pilot. The ICs would then have driven for Lyft, whenever, in 2 whatever mode, and for as many hours as they chose (see TX 150). 3 45. In the 65+5 proposal, Lyft estimated that 65 rental WAV vehicles would have 4 been able to handle 80% of the demand for WAV rides in the three counties. It provided for a 5 continued contract with First Transit for five WAVs, which, the 2019 memo hypothesized, 6 would have adequately covered the remaining 20% of demand. 7 46. In planning the 65+5 proposal, Lyft had assumed that rental ICs would have 8 accepted 90% of the WAV rides that came in and that they each would have driven 35 hours 9 per week. 10 47. At that time, Lyft estimated the cost of the WAV program would have been two 11 million dollars over two years. Additionally, the five WAV vehicles through First Transit, at a 12 cost of $360,000 per vehicle per year, would have totaled $1.8 million each year. Thus, the 13 total 65+5 proposal would have cost $2.8 million per year, according to Lyft’s estimate in 2019 14 (Tr. 105, 564–65, 788). 15 48. Lyft, however, put its plans for the 65+5 proposal on hold after Hertz announced 16 bankruptcy. Lyft has not explored a rental program for the Bay Area with any other company. 17 49. Since the pandemic began, Gerundio testified, her confidence about the viability 18 of the rental model has plunged (Tr. 420): 19 We’ve learned a lot over the year, almost painfully, with the rental program in New York City and ICs in general across all of our other 20 markets. And what we’ve found is that, even though we would 21 really optimistically want the rental program to work, it just doesn’t because they are so busy with other standard rides that they just 22 won’t be able to be available for the WAV rides. And we still cannot control how many . . . driver hours they would drive and 23 when they would drive and where. 24 50. Lyft has had to provide special incentives to New York City ICs to rent WAVs 25 (Tr. 87–93, 201). 26 51. Lyft has had to offer additional incentives to induce rental ICs to drive in WAV 27 mode. Even so, rental drivers in New York City spent a tiny percentage of their operating time 1 in WAV mode: between January 2021 and May 2021, they spent just one to three percent of 2 total operating time in WAV mode (Tr. 197; TX 209): 3 52. This low rate is due to low demand for WAV rides. In response to that low 4 demand, rental drivers switched over to the standard Lyft mode in order to pick up more riders. Rental Utilization (Access vs. Non-Access) | fi Explore 6 =< Feseaiilicacn pear sect: cing elise , 35%

8 E 05/26/2021 25% + access_utilization 21398%

10 □ 11 tel aan cieeraari oyogizon □□□□□□□□□□□□□□□□□□□□□□□□□□□□□□□□□□□□□□□ oajosf0n o4pig2021 08/03/2021 os/17/2021 12

13 53. If designing a WAV program for the Bay Area today, Gerundio (and Lyft) would 14 not use the rental model. This is true despite Lyft’s success in tweaking the app to double the 3 15 demand for WAV rides in New York City. (Lyft accomplished this increase by “toggling on” 16 the WAV mode option for all New York City Lyft riders.) 3 17 54. Plaintiffs civil engineering expert, Alice Grossman, Ph.D., did not refute the 18 weaknesses of the rental model. Nor did she opine on the specifics of a model that could work 19 for the three counties. Dr. Grossman testified that neither the data needed to estimate the 20 number of W-2 or rental WAVs, nor the cost of WAV service in the three Bay Area counties 21 will exist until Lyft attempts to implement the service. She also opined that the 65+5 proposal 22 was a good “starting point” to see what the correct numbers might be (Tr. 640, 667-68). Lyft’s 23 success in meeting “target performance outcomes” in “other urban regions,” she testified, made 24 her optimistic that “maybe we can try to reach” performance targets of 80% of rides within 30 25 minutes and 100% within 65 minutes (Tr. 652, 654). 26 55. It is true, as Dr. Grossman pointed out, that Lyft generally has been able to meet 27 its wait-time standards for WAV rides. In fact, a 2019 internal Lyft report indicated that 28 during “a summer month” that year, 35% of Lyft’s WAV rides nationwide arrived in under 15

1 minutes and 76% arrived in under 30 minutes (Tr. 111–12). Nationwide regulatory standards 2 appear, however, generally less stringent (with the sole exception of New York City) than the 3 CPUC’s exemption standard (TX 20 at 008). New York, for its part, has a very dense 4 population and provided by far the most WAV rides in 2019, thus skewing the nationwide 5 figure. Furthermore, Dr. Grossman did not perform regression analysis or another detailed 6 comparison of New York City (or any other region) to the Bay Area (TX 79 at 013; TX 220 at 7 3; Tr. 633–34). 8 56. All in all, Dr. Grossman did not adequately compare the Bay Area to other cities, 9 such that those results could be relied upon. Nor did she propose a model that appears 10 adequate to meet the CPUC offset standard in the Bay Area. 11 57. This order finds that the 2019 proposal for 65 rental WAVs and five W-2 WAVs 12 will not, going forward, meet CPUC wait-time standards because rental WAVs are an 13 unreliable means of increasing the supply of WAV rides for would-be customers. This order 14 also finds that the W-2 model represents the only currently-available option able to provide 15 service reliable enough to meet CPUC offset requirements (Tr. 457–58, 108, 567, 568, 570). 16 58. Specifically, a 15-WAV partner program represents the program most capable of 17 meeting CPUC offset benchmarks, at a cost of $5.4 million per year. Gerundio testified that 18 Lyft’s current, conservative, estimate based on its recent experience is that it would take “five 19 vehicles for each county, and not five vehicles for one county” to meet CPUC wait time 20 requirements in the three counties. Thus, it would take at least 15 partner vehicles to cover all 21 three Bay Area Counties just to meet CPUC offset requirements (Tr. 559–68). 22 59. If plaintiffs insist on a rental program along with the W-2 vehicles, though, the 23 cost of the program would amount to $5.4 million plus the one million for the 65 vehicles, for a 24 total of $6.4 million per year (ibid.). 25 60. The price estimates above do not account for potential cross-dispatching of W-2 26 drivers, meaning allowing the W-2s to pick up standard Lyft passengers. Shortly before trial, 27 Lyft began writing algorithms to permit this, while prioritizing WAV riders and minimizing 1 opportunities to accept rides and for corresponding cashflow. As of trial, Lyft had 2 experimented with these algorithms, but had not yet had success. (Cross-dispatching instead 3 increased wait times and decreased WAV drivers’ availability for wheelchair users.) It is 4 impossible, based on our trial record, to predict how much, if any, these measures would save 5 (Tr. 308, 603, 605, 606). 6 61. With regard to cross-dispatch testing, comparisons to other jurisdictions, the 7 CPUC quarter-over-quarter improvement requirement, the viability of the WAV rental 8 program, and the number of W-2 vehicles needed to meet the CPUC offset standard, this order 9 adopts and accepts plaintiffs’ proposed findings Nos. 77, 84, and 85. This order adopts and 10 accepts defense proposed findings Nos. 35, 58, 78, 89, and 91. 11 ANALYSIS AND CONCLUSIONS OF LAW 12 1. Title III of the ADA pertains to places of public accommodation and provides in 13 part, “No individual shall be discriminated against on the basis of disability. . . .” 42 U.S.C. § 14 12182(a). In turn, Section 12184 of Title III applies to “a private entity that is primarily 15 engaged in the business of transporting people and whose operations affect commerce.” That 16 section further defines discrimination, in relevant part, via a cross-reference to another section, 17 which specifically defines discrimination to include a failure to “make reasonable 18 modifications consistent with those required under section 12182(b)(2)(A)(ii) of this title.” 19 2. In turn, 42 U.S.C. § 12182(b)(2)(A)(ii) requires modifications 20 when such modifications are necessary to afford such goods, services, facilities, privileges, advantages, or accommodations to 21 individuals with disabilities, unless the entity can demonstrate that 22 making such modifications would fundamentally alter the nature of such goods, services, facilities, privileges, advantages, or 23 accommodations. 24 Our summary judgment order found that Section 12184(b)(2)(A)(ii) applies to Lyft as a private 25 entity whose primary business involves transporting people, that plaintiffs are disabled within 26 the meaning of the ADA, that they requested a modification, that a proposed modification 27 appeared necessary, and that no fundamental alteration defense is available. At summary 1 3. Lyft now asks that this order revisit the fundamental alteration defense (Dkt. No. 2 92). Based on the record at trial, this order declines to do so. WAV services would not 3 fundamentally alter Lyft’s business because WAVs are not a “new transportation service.” 4 Lyft already provides WAV services where compelled to do so by local governments and 5 voluntarily in Los Angeles and San Francisco (on a pilot basis) (Tr. 662). 6 4. At summary judgment, plaintiffs pursued an additional theory of discrimination 7 under a different subpart of Section 12184, namely subsection (b)(3). This subsection requires 8 businesses of public accommodation to provide “a level of service equivalent . . . to the level of 9 service provided to the general public,” and provides that failure to do so is discrimination. 10 Our summary judgment order found for plaintiffs on the existence of a discriminatory policy, 11 practice, or procedure under that subsection. At trial, however, plaintiffs abandoned the 12 Section 12184(b)(3) claim and proceeded solely with a reasonable modification claim under 13 the standard mentioned above (Tr. at 26). Analytically, then, this order begins with the 14 premise that no discrimination exists unless the evidence shows that Lyft has failed to 15 implement a requested — and reasonable — modification (Dkt. No. 92 at 12). 16 * * * 17 5. Lyft concentrates its fire on plaintiffs’ failure to propose a concrete modification. 18 Indeed, instead of offering a concrete modification, plaintiffs propose that Lyft undertake an 19 “iterative process” of trial and error to try to find a reasonable modification. 20 6. Our federal courts have routinely rejected claims under Section 21 12182(b)(2)(A)(ii) when plaintiffs have failed to propose concrete modifications. Our court of 22 appeals has described the task of the district court as evaluating the reasonableness of the 23 “requested modification.” Fortyune v. Am. Multi-Cinema, Inc., 364 F.3d 1075, 1082 (9th Cir. 24 2004). In that case, a cinephile wheelchair user proposed that the theater reserve a seat 25 adjacent to the wheelchair area for his companion. In Elliott v. Harris, 205 F. App'x 255, 258 26 (5th Cir. 2006), the plaintiff requested an accommodation in a horseback-riding facility. That 27 plaintiff appealed on the ground that the district court had not instructed the jury to consider 1 Appeals for the Fifth Circuit disagreed that the district court should have instructed on any 2 other possible modification since the plaintiff had not “suggested any” other. Ibid. 3 7. Our court of appeals has approved of proposals that are concrete, not 4 experimental, under Title III. In Lentini v. Cal. Ctr. for the Arts, Escondido, 370 F.3d 837, 846 5 (9th Cir. 2004), a concert hall was required to grant a patron’s request to attend with a service 6 dog. In Baughman v. Walt Disney, 685 F.3d 1131 (9th Cir. 2012), a parkgoer whose disability 7 allowed her to stand, but made it very difficult to walk or to stand from a seated position, won 8 her request to use a Segway, rather than a wheelchair, to improve her experience of the park. 9 Finally, the Supreme Court’s decision in PGA Tour v. Martin, 532 U.S. 661 (2001), also 10 involved a clear-cut proposal: a golfer submitted a written request to use a golfcart, rather than 11 walk, while competing in a tournament, and prevailed. 12 8. District courts have followed suit. In Smith v. Walgreens Boots All., Inc., No. 20- 13 CV-05451-CRB, 2021 WL 391308, at *9 (N.D. Cal. Feb. 3, 2021) (Judge Charles R. Breyer), 14 the court determined that the plaintiff had not proposed a modification to a drugstore refusing 15 to fill opioid prescriptions and rejected the claim on that basis. Similarly, in Mannick v. Kaiser 16 Found. Health Plan, Inc., No. C 03-5905 PJH, 2006 WL 2168877, at *12 (N.D. Cal. July 31, 17 2006) (Judge Phyllis J. Hamilton), the “[p]laintiff did not meet his burden because he did not 18 show that he requested a transfer as a reasonable modification.” 19 9. District courts evaluating reasonable modification proposals aimed at 20 transportation providers under Sections 12182(b)(2)(A)(ii) have also uniformly considered 21 concrete proposals. For example, in Doud v. Yellow Cab Co. of Reno, No. 3:13-CV-00664- 22 MMD, 2014 WL 4302552, at *2 (D. Nev. Aug. 28, 2014) (Judge Miranda M. Du), a would-be 23 taxi customer used a scooter that disassembled into stowable parts. Drivers refused to serve 24 her and her husband, who did not use a scooter. See id. at *3. The Douds successfully 25 “ask[ed] the Court to require Yellow to incorporate[] into [its] training that drivers must not 26 prohibit Mrs. Doud from riding in its ordinary taxis.” Id. at *2 (cleaned up). 27 10. In Dahlberg v. Avis Rent A Car Sys., Inc., 92 F. Supp. 2d 1091, 1108 (D. Colo. 1 driving vehicles with hand-held controls. Despite relying on Section 12182(b)(2)(A)(ii), the 2 wheelchair-user plaintiff did not suggest a particular modification to the policy, which Avis 3 had already changed in light of a settlement agreement between Avis and the Department of 4 Justice. The court concluded, the plaintiff 5 attempts to deflect his burden of suggesting a reasonable modification onto Avis by contending that Avis is asking this court 6 to conclude as a matter of law that Avis cannot process reservations 7 for hand-controlled vehicles through [its reservation] system in the same manner as other reservations. 8 9 Ibid. Moreover, in each of the successful proposed modifications, not only was the proposal 10 concrete and specific, but it was straightforward and simple. For example, reserving the 11 adjacent movie-theater seat for the companion of the disabled patron is immediately easy to 12 grasp. 13 11. No federal court has ever held that an iterative, experimental, or trial-and-error 14 proposal constituted a reasonable modification to a policy, practice, or procedure under Title 15 III. 16 12. Accordingly, this order holds that a Section 12182(b)(2)(A)(ii) plaintiff is 17 required to propose a concrete modification rather than merely propose that the district court 18 order a defendant to undertake an iterative trial-and-error process to try to find a proposed 19 modification. A concrete proposal need not outline every detail of the modification it 20 proposes. But the devil’s in the details. A proposal must have enough meat on its bones to 21 allow a fact finder to rate it as “reasonable” (or not), as the statute requires. 22 13. Turning to reasonableness, our court of appeals has said: 23 [T]he determination as to whether a particular modification is “reasonable” involves a fact-specific, case-by-case inquiry that 24 considers, among other factors, the effectiveness of the modification in light of the nature of the disability in question, and the cost to the 25 organization that would implement it. 26 Fortyune v. Am. Multi-Cinema, Inc., 364 F.3d 1075, 1083 (9th Cir. 2004), quoting Staron v. 27 1 14. Plaintiffs admit that they have the burden to prove reasonableness but say that the 2 burden is best defined as the “run of cases” standard, described by the Fifth Circuit in Johnson 3 v. Gambrinus Co./Spoetzl Brewery, 116 F.3d 1052, 1059–60 (5th Cir. 1997), a decision 4 considering a patron’s request to bring a service animal despite the no-animal policy. That 5 decision referred to ADA “regulation and commentary,” which “reflect an administrative 6 determination that modifying a no animals policy to allow a service animal full access with its 7 owner in a place of public accommodation is generally reasonable, or . . . reasonable in the run 8 of cases,” and that only occasionally would allowing guide dogs fundamentally alter a service 9 or pose a risk to safety. Id. at 1060. 2 10 15. This order presumes that plaintiffs are correct as to the “run of cases” standard. 11 This order need not make a final determination on the legal question, however. Even under 12 this standard, plaintiffs have not shown a modification that would be reasonable in the run of 13 cases and in any event, Lyft has shown the proposed modification would be unreasonable. 14 Significantly, the phrase “in the run of case” underscores all the more that the statutory 15 provision in question requires a concrete proposal, for otherwise the proposal could not be 16 tested against the run of cases. Again, we have no concrete proposal from plaintiffs to test 17 against the run of cases. This point is dispositive. 18 16. In post-trial briefing, plaintiffs argued (Dkt. No. 202 at ¶ 102) (emphasis added): 19 Plaintiffs propose as a remedy that Lyft be required to meet wait 20 time benchmarks (benchmarks no longer than those set by the CPUC to qualify for exemption from payment into the Access Fund 21 . . .) that are well within Lyft’s capacity and are in line with what the CPUC has determined is feasible, and reliability standards [sic]. 22 If Lyft meets these standards, it will be exempt from paying into the Access Fund, leaving Lyft with significant funds to cover a large 23 majority, if not all, of the cost of its Bay Area WAV program . . . . 24 Plaintiffs’ proposed remedy would provide Lyft a full year to engage in the industry-standard iterative process . . . that Lyft itself 25 uses for both its Standard and Access services . . . before having to 26

27 2 The Ninth Circuit in Lopez v. Catalina Channel Express, Inc., 974 F.3d 1030, 1036 (9th Cir. 2020), meet reasonable benchmarks for WAV service, but Plaintiffs do not 1 ask the Court to order that Lyft engage in an iterative process. 2 17. This is not a concrete proposal or modification. It is a performance standard. It 3 begs the question of how it could be done. Plaintiffs answer, in effect, how it could be done 4 would be Lyft’s problem, offering an iterative process to help Lyft make “any changes [that] 5 are needed to the combination of models, number of vehicles, vehicle placement, or other 6 inputs” (Dkt. No. 201 ¶ 149). This is word play. Plaintiffs propose that Lyft meet a 7 performance standard but fail to show how it could be done, leaving it to Lyft and trial and 8 error to figure it out. 9 * * * 10 11 18. The closest our plaintiffs come to a concrete proposed modification is to argue 12 that Lyft itself generated an internal proposal in 2019 to roll out WAV service in three 13 counties. This order will evaluate that internal proposal. As stated in the findings, the proposal 14 contemplated 65 WAV rental vehicles and five partner W-2 WAVs (the “65+5” proposal) at a 15 then-estimated cost of $2.8 million (before any funds from the CPUC Access Fund). That 16 proposal, however, was not a final plan and was never adopted. 17 19. Turning to the specifics, Lyft showed at trial that the 2019 proposal would cost far 18 more than $2.8 million and that offsets from the CPUC could not be sustained over time. To 19 this, plaintiffs reply that Lyft should be ordered to undertake the program anyway and that 20 through an iterative process of trial and error, the parties and the Court would eventually arrive 21 at the most cost-effective combination of rental units versus W-2s. As this order will now 22 explain, however, Lyft showed at trial that the probable cost of an effective program meeting 23 the performance standard for all three counties would rate as unreasonable. 24 * * * 25 20. The 65+5 proposal never became a final plan. It remained an internal pitch. 26 Before it could be further vetted, the pandemic hit. The contemplated rental partner, the Hertz 27 Corporation, went bankrupt. To reiterate, in developing its 65+5 proposal, Lyft estimated that 1 the three counties would require 65 rental WAVs (rented from Hertz) and five W-2 WAVs 2 (from First Transit) to meet the CPUC Access Fund “offset time standard.” 3 21. Meeting the offset time standard would be important to Lyft because doing so 4 would entitle it to recoup some funds from the CPUC. When transportation network 5 companies (TNCs) like Lyft (and Uber, Via, etc.) meet the offset time standard, they may draw 6 down on reimbursements from the CPUC up to their total quarterly contribution for that county 7 to the Access Fund (which TNCs pay into at a rate of 10-cents per ride). The CPUC offset 8 time standard varies by county. In San Francisco, either 50% of rides must have wait times 9 within 15 minutes or 75% of rides must have wait times within 30 minutes, determined each 10 quarter. In Alameda and Contra Costa Counties, 50% of WAV rides must have wait times 11 within 25 minutes or 75% of rides within 50 minutes, also determined by quarter. Importantly, 12 to qualify for reimbursement, TNCs must improve wait times quarter over quarter. 13 22. Plaintiffs’ proposal, however, asks Lyft to meet an even-more stringent time 14 standard also set by the CPUC. The “exemption standard,” once met, would entitle a 15 qualifying TNC to avoid paying 10-cents per ride into the access fund, for four quarters. To be 16 clear, TNCs need not meet the exemption standard in order to draw down on the offset fund 17 each quarter. Rather, the CPUC established the exemption standard to financially incentivize 18 wait times even shorter than what the offset standard requires (see Dkt. No. 202 ¶ 102; Dkt. 19 No. 201 ¶ 152). 20 23. The exemption criteria are more difficult to achieve because they require a TNC: 21 (1) to provide shorter wait times (80% of WAV rides within 16 minutes in San Francisco, 80% 22 within 20 minutes in Alameda County, and 80% within 40 minutes in Contra Costa County); 23 (2) to maintain these shorter wait times over four consecutive quarters; and (3) to show 24 quarter-over-quarter improvement to ETA benchmarks (a requirement shared with the offset 25 standard). The payoff for meeting the exemption criteria is that for the four quarters following 26 its success with the above criteria, the TNC would be exempt from paying the 10-cents per-ride 27 fee into the Access Fund. This benefits TNCs because meeting the exemption criteria would 1 enable them to save the funds, rest assured of those cost-savings, and avoid the administrative 2 burden of proving eligibility. 3 24. Based on Lyft’s presentation of more recent experience with the rental program, 4 this order finds that $2.8 million (in the internal memo) was merely an initial estimate that was 5 inaccurately low. Costs would far exceed $2.8 million. Lyft has further shown that the rental 6 model is unworkable and that the three-county WAV program would require 15 (not five) 7 partner vehicles. 8 25. The partner program is very expensive. Partner WAV rides cost $1500 per ride 9 in San Francisco (TX 177 at 018, Fig. 5). Partner WAV drivers are dedicated to WAV riders, 10 but are very underutilized, as the actual San Francisco pilot has shown. As a result, partner 11 drivers spent substantial time idle, waiting for riders, literally just sitting for hours in parked 12 vans. 13 26. The rental model is unworkable. Unlike the partner model, which Lyft has tried 14 in San Francisco, the rental model has not been tried at all in California. Data from Lyft’s 15 experience in New York City, however, show that rental WAV drivers spend most of their 16 operating time in non-WAV mode picking up ordinary passengers. They spent only two to 17 three percent of their driving time in access mode in June 2021, and one to three percent 18 between January 2021 and May 2021 (TX 209; Tr. 197). This imbalance occurred because 19 WAV rides were infrequent and rental drivers switched over to the standard Lyft mode in order 20 to make ends meet. As the W-2 pilot has proven in actual practice, a pure WAV rental model 21 would also lead to hours of idle drivers sitting in parked vans waiting for a WAV call if drivers 22 were dedicated solely to WAV service. It would be unrealistic to expect drivers to sign up for 23 a dedicated WAV rental program. 24 27. As a result, Lyft adjusted its estimate of the number of W-2 vehicles it would 25 need to meet the CPUC standards. Lyft analysts explained at trial that the three counties would 26 require 15 W-2 vehicles (not five) to satisfy demand and meet the CPUC Access Fund offset 27 time standard. (But there would be no rental drivers.) 1 28. Plaintiffs question this. But their expert, Dr. Grossman, failed to offer a counter 2 estimate. Instead, Dr. Grossman testified that she did not know the number of vehicles needed 3 and that the number of W-2 vehicles required would have to be determined by trial and error 4 (Tr. 667–69). Because of the low WAV-mode utilization shown in New York City (an even 5 denser urban area than the Bay Area), this order finds that five W-2 WAVs will not suffice and 6 that 15 W-2 WAVs would be required. 7 29. Trial testimony from Isabella Gerundio, Program Manager for Wheelchair- 8 Accessible Vehicles at Lyft, showed that the only program capable of reliably meeting CPUC 9 offset benchmarks would be a 15-WAV partner program, at a cost of $5.4 million (see Tr. 567– 10 68, 570). If Lyft implemented a WAV program in the Bay Area today, Gerundio (and Lyft) 11 would not include the rental model at all. This order accepts the testimony that 15 WAVs 12 would, in fact, suffice to meet CPUC offset requirements without any rental vehicles. 13 30. If plaintiffs insist on a rental program to boot, the program would still require 15 14 WAVs, as well, at a total cost of $6.4 million (not $2.8 million). Some but not all of this could 15 be recouped from the Access Fund if (but only if) Lyft were to meet the offset standards, which 16 subject will be covered below. 17 31. The $5.4 million, however, does not account for plaintiffs’ proposed wait times, 18 i.e. the more stringent exemption standard. The exemption standard requires drivers to reach 19 WAV passengers more quickly. Meeting it would demand still more W-2 WAV vehicles, 20 since Lyft has established that only by adding more W-2s can Lyft reliably reduce wait times 21 (Dkt. No. 201 ¶ 152; see, e.g., Tr. 458). Therefore, plaintiffs’ proposal would certainly cost 22 more than $5.4 million. 23 32. Plaintiffs respond that Lyft has failed to consider cost-saving methods, and that it 24 has not disproven the $2.8 million cost estimate. 25 33. First, plaintiffs point out that WAV costs in New York City have decreased over 26 time (Tr. 82). True, this may yet occur in the Bay Area, but to uncertain effect. Remember 27 that we have real cost-per-ride data from the W-2 pilot in San Francisco and that figure has 1 34. Second, plaintiffs argue that Lyft could modify its user app so that customers 2 would not have to “toggle on” the WAV mode option. This would cause the WAV option to 3 immediately appear for any Lyft user, instead of requiring a user to change the app’s settings 4 manually. With the WAV option immediately visible in the application, Bay Area demand 5 would likely increase, as happened in New York City. But, again, how much in cost savings 6 would occur remains unproven and speculative (Tr. 146–49). 7 35. Third, plaintiffs argue that Lyft could tweak the algorithm that matches riders and 8 drivers to solve the problem that rental WAV drivers, to make money, often switch to standard 9 mode. Otherwise, drivers cannot make ends meet. Plaintiffs failed, however, to establish that 10 potential changes to the algorithm could meaningfully alter WAV mode utilization by rental 11 drivers. Testimony showed that tests of “cross-dispatching,” to date, have increased wait times 12 and lowered WAV availability (Tr. 303–08, 603–07). 13 36. Fourth, plaintiffs argue that Lyft could raise prices to help pay for WAV. To 14 calculate the yield of a five-cent per-ride price increase, plaintiffs estimate the quarterly 15 number of rides Lyft will provide in the three counties. They do so by adding Lyft’s Access 16 Fund contribution from San Francisco for 2020Q1 and the contributions from Alameda and 17 Contra Costa Counties for 2019Q4 and annualizing, for a total of $3.37 million. They work 18 backwards for the number of rides (dividing by 10 cents) and then multiply by five cents for 19 the yield. At a five-cent per-ride fee, plaintiffs posit, Lyft could raise $1.685 million in 20 revenues to pay for WAV each year (Dkt. No. 201 ¶¶ 16, 17, 121). 21 37. Plaintiffs’ number, $1.685 million, is sleight of hand. For one thing, it mixes and 22 matches different quarters. Instead, let’s take a single quarter, 2019Q4. For that quarter, the 23 Access Fund contribution from San Francisco was $477,177, $240,000 from Alameda County, 24 and $60,000 from Contra Costa County, for a total of $777,177. Annualizing this figure comes 25 out to $3.11 million at the ten-cent rate. At five cents per ride, the annual yield for all three 26 counties would be half as much, or $1.554 million, not $1.685 million. More importantly, it 27 still would not come close to covering the $5.4 million per year that the WAV program would 1 38. Fifth and sixth, plaintiffs claim that Lyft could still increase driver incentives to 2 bump up supply and that Lyft has not “consulted external parties” to plan its program (Dkt. No. 3 201 ¶ 86). Plaintiffs do not establish how these measures would save costs. 4 39. At base, plaintiffs argue that the untapped cost savers could zero out expensive 5 problems such as drivers’ unwillingness to operate in WAV mode and reluctance to rent 6 WAVs. This order finds that plaintiffs have not shown that the cost will be as low as $2.8 7 million and that Lyft has shown that a WAV program in San Francisco, Alameda, and Contra 8 Costa Counties would exceed $5.4 million per year in order to meet the requested exemption 9 time standard. 10 40. This order next turns to the feasibility of cost-savings from the CPUC. 11 * * * 12 41. Lyft could save $3.11 million over four quarters if it were to meet either the CPUC 13 offset criteria or the exemption standard (TX 98). This order concludes that Lyft cannot expect 14 to sustain the CPUC Access Fund requirements in the long term. Critically, the CPUC 15 requires quarter-over-quarter improvement in overall ETAs for TNCs to qualify for the Access 16 Fund. Such ever-better improvement will be unsustainable. 17 42. True, the CPUC could change its mind about that requirement. It may continue 18 its rulemaking until 2025, just before the TNC Access Fund sunsets in 2026, and the CPUC 19 could still eliminate the quarter-over-quarter improvement requirement (TX 14 at 019). Lyft 20 has explained the difficulties to the CPUC, but as of trial and this order, the tough requirement 21 remains in place. This order cannot assume any change will follow. This order has waited for 22 any new word from the CPUC on its thinking, but the time has come to issue a decision in this 23 case. 24 43. The CPUC has baked in a bit of flexibility by allowing Lyft to improve, for a 25 given quarter, in either the Level 1 tranche or the Level 2 tranche (see supra, Findings of Fact 26 ¶¶ 23–24). The evidence, however, indicates that it is difficult to predict the factors that affect 27 wait times. Problems could arise with supply, demand, traffic, and/or vehicle costs (to name a 1 problems could emerge if Lyft were to improve in its wait times. Lyft could overshoot a 2 level’s improvement goal for a quarter and find itself unable to improve on that level in the 3 quarter that followed. Similarly, exogenous factors could shorten ETAs, for example by 4 decreasing traffic, which would force an improvement in wait times. This could set an 5 impossible-to-surpass performance standard for the next quarter. Indeed, improved traffic 6 could dramatically improve the percentages of rides meeting the wait time benchmarks in both 7 levels, leaving little room for improvement in either. Perpetual kaizen is impossible. For this 8 reason alone, this order finds that Lyft will not receive, on a quarter-over-quarter basis, any 9 CPUC offset. (Since this order finds the quarter-over-quarter improvement unworkable, it 10 need not address the other CPUC offset standards, including the “Trip Completion Standard” 11 or the wait-time criteria.) 12 44. Of course, plaintiffs request the exemption time standard as part of their proposed 13 modification (Dkt. No. 201 ¶ 152). Apart from the quarter-over-quarter improvement 14 requirement, Lyft would be unlikely to meet the tougher exemption time standard. It has never 15 met this standard in San Francisco. Plaintiffs answer that Lyft can do so, largely because Lyft 16 has managed to meet regulatory requirements in other cities. In those places, however, local 17 regulators require Lyft to provide WAVs and to meet benchmarks as conditions of operating. 18 Lyft’s accomplishments there are not clearly comparable to the Bay Area. 19 45. For instance, New York City regulators require WAV service as a condition of 20 operation. Out of all regions, New York City has set Lyft the strictest ETA requirements. In 21 June 2019, in New York City, with its very dense population, Lyft accomplished 60% of WAV 22 rides in under 15 minutes, and 90% under 30, using a hybrid of rental, organic driver, and 23 partner models.3 24 46. Other cities’ standards are less strict than the exemption standard, meaning that 25 Lyft’s ability to meet those standards cannot inform the likelihood that Lyft will meet the 26 27 3 During trial, Lyft was working to meet revised wait time requirements in New York City, which are: 80% of WAV rides within 10 minutes and 90% within 15. Lyft employees testified that it was not projected to approach that 1 CPUC exemption standard. While plaintiffs point to a 2019 internal report indicating that 35% 2 of WAV rides nationwide saw wait times under 15 minutes and 76% under 30 minutes, New 3 York City data, boasting the largest number of WAV rides, drove this result. 4 47. Lyft’s ability to meet New York City time standards and its nation-wide wait 5 times are encouraging, but plaintiffs have failed to establish that the Bay Area is comparable to 6 New York City. Without that showing, New York City data cannot reliably model Bay Area 7 performance. Plaintiffs’ expert Dr. Grossman admits, “New York City policies, transportation 8 systems, and socio-demographics vary from other U.S. urban regions;” she also acknowledges 9 that her report did not conduct a “robust statistical analysis with control variables” (TX 79 at 10 015; see also TX 220 at 3). Dr. Grossman suggests that “there are parallels to other cities,” but 11 she does not explain how New York City’s data adequately compare on the most relevant 12 metrics, including population density and its impact on demand for rides and wait times (TX 13 79 at 015–16). 14 48. The fundamental problem is that New York City has an extremely dense 15 population, even compared to San Francisco. There are far more WAV rides per block and the 16 greater density makes it feasible to ensure shorter wait times. Since the New York City figures 17 largely drive the nationwide data, neither represents a reliable comparison to the Bay Area. 18 The evidence does not suggest that Lyft will accomplish quarter-over-quarter improvement. 19 Therefore, this order cannot subtract the CPUC offsets from gross costs. 20 * * * 21 49. In Fortyune, 364 F.3d at 1083 (internal citations omitted), our court of appeals 22 held, “[A]n accommodation is not reasonable if it imposes undue financial and administrative 23 burdens.” 24 50. This order has already determined the gross cost of the three-county WAV 25 program. It earlier discussed the feasibility of meeting CPUC offset or exemption criteria. 26 And, it has summarized the average and peak monthly demands for WAV rides in San 27 Francisco, which came to 60 and 125, respectively (Tr. 414, 153). This order finds the net per- 1 51. First, to repeat, no other significant contributions appear. This order has already 2 concluded it cannot discount gross costs with CPUC offsets. Nor do WAV passengers pay a 3 special rate. Therefore, offsets and contributions take nothing off the true price tag per ride. 4 52. Second, at a net cost of $5.4 million per year across the three counties and 60 5 rides per month (the First Transit average in San Francisco), the per-ride cost of plaintiffs’ 6 proposal would amount to $2500 per ride. (In so calculating, this order assumes for 7 simplicity’s sake that each of our three counties would see the same total demand for WAV 8 rides per month and so receive equal shares of the WAV budget.) Alternatively, if the number 9 of rides per month were to hit 125 (First Transit’s record high), the per-ride cost would dip to 10 $1200. Even if plaintiffs correctly estimated that the three-county WAV program would cost 11 $2.8 million per year, the per-ride cost would equal $1296 (at 60 rides per month) or $622 (at 12 125 rides per month) (Tr. 153, 821). Finally, even if Lyft managed to increase demand by 13 100% (by changing the toggle settings to make the WAV option appear), as plaintiffs propose, 14 the range of prices above would still rate as excessive and unreasonable.4 15 53. Plaintiffs respond that all of these numbers represent chump change for a 16 company of Lyft’s size. Plaintiffs also urge that this order define “undue . . . burden” 17 consistent with Section 504 of the Rehabilitation Act and the interpretive regulations of Title 18 III, whose definitions both account for the size of a defendant’s bank account. See 45 C.F.R. 19 §§ 84.12(c)(1), (c)(3) and 28 C.F.R. § 36.104. Lyft counters that a company’s size does not 20 matter, and that nothing in Section 12184’s reasonable modification provision suggests a 21 contrary definition. 22 54. Even a vast bottom line does not transform exorbitant modifications into 23 reasonable ones, so this order need not decide the appropriate definition of “undue . . . burden.” 24 Fortyune, 364 F.3d at 1083. Plaintiffs’ proposed price tag far exceeds those of prior approved 25

26 4 True, the possible $622 cost per ride could drop by half to $311 if demand increased 100%. This order finds, however, that possibly remains too remote. It would require a precarious confluence of events: maintaining a 27 pre-pandemic peak monthly demand of 125 WAV rides, a very unlikely $2.8 million per-year cost, and an increased demand of 100%. It would be unreasonable to impose a modification based on such an unlikely 1 modifications under Title III. The Fortyune decision required reserved movie seats and that of 2 PGA Tour v. Martin, 532 U.S. 661 (2001), found that the tournament should have permitted 3 one of its competitors to use a golf cart. Our summary judgment order held that a $925 per- 4 ride cost to Lyft would be unreasonable. This order now concludes that per-ride net costs of 5 $622, $1200, $1296, or $2500 would be too. This is dispositive by itself. 6 55. The real cost per WAV ride in San Francisco is $1500. We must ask whether it 7 would be reasonable to order Lyft to supply such an inherently costly alternative. In the other 8 two counties the real cost per ride would be even greater because those counties have less 9 density than San Francisco and, by definition, the travel time between rides will be greater. 10 Again, would it be reasonable to order Lyft to supply such an inherently costly alternative? 11 56. Plaintiffs remind us that some of this cost could be borne by the CPUC offsets. 12 Lyft, however, cannot count on the CPUC offsets because meeting the quarter-over-quarter 13 performance standard cannot be sustained. But even if Lyft received the Access Fund offsets 14 in all three counties, the offsets would fall far short. If we annualize Access Fund 15 contributions from all three counties from 2019Q4 (again, the only pre-pandemic quarter for 16 which trial revealed Access Fund contributions paid by Alameda and Contra Costa Counties), 17 the offsets would amount to just $3.11 million for the year. At a base cost of $5.4 million per 18 year, Lyft would still pay $2.29 million per year for the WAV program, even with the CPUC 19 offset funds (Tr. 564; TX 98). 20 57. But, even if somehow the CPUC fund would cover all of Lyft’s expenses, 21 someone would still bear the burden of the astronomical per-ride cost. In that contingency, the 22 CPUC funds, which would perhaps otherwise go to support other accessible transportation 23 services, would be diverted. So, there would be a real detriment, a real opportunity cost. 24 Again, we must ask, is it reasonable to order Lyft to supply a service that will cost society, if 25 not Lyft, $1500 or more per ride? This order holds that it would be unreasonable. 26 58. In addition, implementing plaintiffs’ proposed modification would impose a 27 significant “administrative burden.” At trial, Gerundio described the enormous effort required 1 refereeing compliance with CPUC benchmarks. She also detailed the formidable challenge of 2 getting drivers to rent WAV vehicles and to drive in WAV mode. Administering the WAV 3 program would require substantial time and effort. 4 59. Meanwhile, plaintiffs’ proposal could negatively impact existing paratransit 5 services by siphoning demand. Existing paratransit services provide WAV rides at low cost to 6 consumers. Private-public partnerships in the three counties support a complete existing 7 infrastructure for WAV, with employees and customers who rely on it. Though the parties did 8 not address the latter point at trial, it seems clear that introducing a WAV competitor could 9 negatively affect the demand for existing public WAV services. Paratransit budgets and 10 corresponding quality likely hinge on a baseline of demand. Decreasing demand, in turn, could 11 injure that system, which our taxpayers fund. Eroding the existing WAV transportation 12 infrastructure underscores the substantial burden that this proposed modification would require. 13 60. One may ask whether it is unreasonable to ask Lyft to do nationwide, WAV-wise, 14 what it already does in New York City. If the ADA is to be read to require everywhere what a 15 single city or group of cities require as a price of doing business, then, yes, we should ask Lyft 16 to do it nationwide. But does the ADA go this far? No court has ever so held. Instead, the law 17 requires reasonableness and specifically reasonableness in the specifics of the case. This order 18 has tried to answer that question based on the specifics of our case. 19 61. Since this order has found that the proposal is unreasonable based upon its cost 20 and administrative burden alone, it does not reach the element of effectiveness. 21 62. Relief is DENIED. Lyft’s Rule 52(c) motion is DENIED AS MOOT. 22 23 24 25 26 27 1 CONCLUSION 2 For the foregoing reasons, plaintiffs have not carried their burden to prove that Lyft 3 discriminated against disabled individuals in violation of the ADA. They have therefore 4 established no entitlement to relief on their claim. Accordingly, judgment will be entered in 5 favor of Lyft. 6 7 IT IS SO ORDERED. 8 9 Dated: September 1, 2021. i " Pee 11 ~ WILLIAM ALSUP 12 UNITED STATES DISTRICT JUDGE

© 15 16

it

4 18 19 20 21 22 23 24 25 26 27 28