The Receiving End

Capture Corner is the premium BD intelligence companion to Mission Meets Tech. Public-record sourced. Independent analysis. Not a recommendation, not vendor advocacy, not capture material. Built for federal health BD, capture, and proposal leaders who need analytical depth, not headlines.

This issue is the companion to "The Floor at Winston-Salem," published this week. The public piece follows a veteran's claims folder from a regional-office floor failing under the weight of paper in 2012 to a 76-day average in 2026, and reads the turnaround as an order of operations: digitize the intake, automate the evidence, route the work, then decide. This one works the buy underneath each layer: what the FY2027 request puts against VBA automation and what it does not, the proof-of-concept contract that reaches its end date this Friday, the intake task order with a March 2027 end and the expired parent vehicle it still rides on, the ADS development question the public record cannot answer, and the criteria a team will have to clear because VA's Inspector General and privacy office have already written them down.

---

1. The picture at a glance

Field	Detail
The result being bought against	75.57-day fiscal-year-to-date average to complete through September 19, 2026; 73,868 rating claims older than 125 days; 3,009,832 completions FYTD [CC1]
Where ADS stands	202 diagnostic codes, more than 540,000 compensation claims touched in 2025, 50 percent of compensation claims by September 2025; FY2027 plan moves it above 70 percent of the rating inventory and adds 60 codes needing more complex logic [CC2]
The proposal for claims AI	$130 million in the FY2027 President's Budget for VBA automation and AI investments in claims processing, described as reducing dependence on surge staffing; a request, not an appropriation [CC3]
The enterprise automation line	$47.759 million requested for FY2027 against $43.070 million enacted in FY2026, up $4.689 million or 10.9 percent; department-wide automation, decision support, and AI infrastructure and governance, not VBA-only [CC4]
Intake requirement	Veterans Claims Intake Program stated as a $425.5 million FY2027 budget-planning requirement; C&P Claims Modernization separately requested at $118.784 million; different labels, not additive without tracing [CC2]
The pilot with a date	36C10X25P0051, SteerBridge Strategies LLC, ACE AI proof-of-concept pilot, $4,998,352.63 obligated and base plus options, awarded August 4, 2025, current and potential end October 2, 2026 [CC5][CC6]
The intake task order	36C10D23N0026 under parent 36C10E19D0016, Deloitte Consulting LLP, $48,282,895.46 obligated, $51,340,053.61 base plus options, current end March 27, 2027, potential end March 27, 2028 [CC7]
The parent that is not open	VICCS parent 36C10E19D0016 shows a $900 million base plus options amount and an end date of March 8, 2025; the task order outlives it; ordering authority for new work is not established by the public record [CC8]
ADS development ownership	Unresolved. The Agile Six purchase order for ABD applications, 36C10B24P0010, ended April 15, 2025 at a reported $4,230,661; a newer Agile Six order under the SPRUCE IDIQ is labeled "Cross Benefits Crew," not ADS [CC9][CC10][CC11]
Oversight already on the books	2023 OIG: 16 of 60 automated hypertension claims inaccurate, directed side-by-side comparison; April 2026 OIG: DIC decisions automated without human involvement, 2 percent legal error, $2.7 million improper; GAO July 2026 testimony: 15 of 43 disability compensation recommendations open [CC12][CC13][CC14]
Overtime as the competing lever	$67.0 million allocated to overtime in the budget, credited with adding capacity without onboarding delay; claims-examiner workforce down 2,950, about 14 percent, September 2024 to June 2026 per third-party OPM analysis [CC2][CC15]

---

2. Read the budget as a sequence of layers, because that is how VA built it

The public piece argues that VA's result came from four layers built in order: intake digitization, automated evidence development, outside-records exchange, and work routing, with a human decision preserved at the end. The FY2027 request can be read the same way, and reading it that way tells a capture team which layer the next dollar lands on.

The intake layer is the largest number and the least discussed. VCIP appears in Volume 3 as a $425.5 million budget-planning requirement for 2027, and the program's stated output is more than 55.5 million images a month with 80 percent or more of C&P mail automated. [CC2] That figure is a planning requirement, which is a different thing from an appropriation, a single award, or a solicitation, and it should not be summed with the $118.784 million C&P Claims Modernization request that sits four pages away in the same volume until someone traces the funding relationship between them. [CC2] What the number does establish is that intake remains the most expensive layer in the stack, and the task order currently executing it is discussed in section 4.

The evidence layer is where the $130 million goes. The President's Budget, in the VA chapter, provides $130 million for VBA automation and AI investments in claims processing and describes the purpose as reducing dependence on surge staffing and extra labor costs. [CC3] Two readings follow. First, this is the administration's own statement that automation is meant to replace overtime as the capacity lever, which matters because overtime is what VA's witnesses credited first in April. [CC16] Second, the line is a request; nothing in it has been appropriated, obligated, or competed as of this analysis, and with the government operating under a continuing resolution through December 11, new starts are constrained until Congress acts. [CC17] A team that prices a pursuit against $130 million is pricing against a proposal.

The enterprise line is smaller and easier to misread. Volume 5 requests $47.759 million for FY2027 against $43.070 million enacted, a 10.9 percent increase, for department-wide automation, decision support, and AI infrastructure and governance. [CC4] That is the governance and platform money, the layer that produces things like the AICES privacy assessment, and it is not VBA's alone. Adding it to the $130 million produces a number no VA document contains.

The routing layer has a date and no dollar figure in the retrieved record. The budget projects National Work Queue modernization for completion in 2027 and publishes the September 2025 baseline it means to improve: 13 days to assignment, 49.9 days in active development, 5.2 days in decision. [CC2] Devlin told the House the goal is to route claims more efficiently. [CC16] A modernized queue is a software and data-integration buy, and the public record does not yet show its vehicle.

3. The pilot whose record ends Friday

VA's privacy impact assessment for the Artificial Intelligence Claims Evaluation System, published in May, identifies the vendor as SteerBridge and the contract as 36C10X25P0051. [CC5] USAspending confirms the award to SteerBridge Strategies LLC on August 4, 2025, at $4,998,352.63 obligated with base plus options at the same figure, explicitly titled an ACE AI proof-of-concept pilot, with a current and potential end date of October 2, 2026. [CC6] The obligated amount equals the ceiling, so the pilot is fully funded and has no unexercised option in the public record.

What the pilot does, per the PIA, is read completed disability benefit questionnaires, interface with source systems, upload results to VBMS, and off-ramp any data that cannot be accurately extracted and validated to human review. [CC5] That is the evidence layer, applied to the exam output that VA's budget says accounts for the longest phase of a claim. The PIA is the only public description of the tool's boundary, and it is a proof of concept, so nothing in the record establishes results, scaled deployment, or a follow-on acquisition.

The capture read is a question, not an opportunity. When a fully funded proof of concept reaches its end date with no announced next phase, the decision belongs to the program office, in this case VBA's Medical Disability Examinations Office, with the contracting action at SAC Frederick. [CC5][CC6] The three outcomes are a no-cost extension to finish evaluation, a production follow-on that may or may not be competed, or a quiet close with lessons folded into ADS. The public record cannot distinguish among them as of September 28. What a team can do is prepare the artifacts the next phase will require, which are the five criteria in section 6, and track the award record for a modification after October 2.

4. The intake task order and the vehicle it outlived

The Deloitte intake work is confirmed at the task-order level and unclear at the vehicle level, and the distinction matters for anyone who has heard VICCS is coming up.

USAspending records task order 36C10D23N0026 to Deloitte Consulting LLP under parent 36C10E19D0016, with $48,282,895.46 obligated, $51,340,053.61 base plus all options, a current end date of March 27, 2027, and a potential end date of March 27, 2028. [CC7] Those are task-order figures. The parent record shows a base plus options amount of $900 million and an end date of March 8, 2025, which is earlier than the task order's own current end. [CC8] A task order that outlives its parent's ordering period is routine; the parent's ordering window closed and the order continues on its own period of performance. What it means for capture is that the public record does not show an open vehicle for new intake work, and no solicitation or confirmed recompete date is established. A prior figure of $25.6 million circulated for this task order is superseded by the USAspending obligation. [CC7]

Two dates are real. March 27, 2027 is the task order's current end, and the exercise or non-exercise of the final option toward March 2028 is the signal to watch. If VA intends to recompete intake, a sources-sought or RFI would need to precede that decision by enough months to run a competition, which puts any such notice in the first half of FY2027. If VA instead intends to fold intake into the $425.5 million planning requirement through another route, the signed modifications and the acquisition strategy behind that requirement are the documents that would show it, and neither is public. [CC2]

5. Who holds ADS today is not in the public record

ADS is the highest-value layer and the one whose development ownership this analysis could not confirm. The lead that circulates is Agile Six, and the record does not support it as stated. Purchase order 36C10B24P0010 to Agile Six covered ABD applications, ended April 15, 2025, and carried a reported value of $4,230,661. [CC9] Automated benefits delivery applications are adjacent to ADS, and a purchase order for them does not establish ownership of the enterprise ADS engine.

The successor lead was tested and does not hold. A newer Agile Six order, 36C10B25N00010003, carries the description "Cross Benefits Crew," and its parent is the SPRUCE IDIQ, 36C10B25D0001, with a different named recipient on the parent record. [CC10][CC11] Nothing in either record says ADS. This is retained as an acquisition-office inquiry lead and nothing more, and a capture plan that names Agile Six as the ADS incumbent is naming a firm the public record does not place there.

The practical consequence: the layer receiving the largest proposed FY2027 investment has no publicly identifiable prime, which means either the work is performed under a vehicle whose task descriptions do not use the program name, or it is substantially in-house, or the record has not caught up. Each of those has a different capture answer, and the first step for any of them is a FOIA or a direct inquiry to VBA's Office of Business Integration for the current ADS development and support contract numbers.

6. Five acceptance criteria VA has already written down

VA has not published a solicitation for the next round of claims automation. It has published two Inspector General reports, one privacy assessment, one congressional statement, and one budget volume that together describe what it will accept, and a team that reads those as requirements is ahead of the team waiting for a SOW.

First, output traceable to source. VA's own standard for the TERA memo automation is that generated content uses verified data with links to source documents. [CC18] A summary that cannot be clicked back to the page it came from does not meet the bar VA set for itself.

Second, quality scored separately for automated work. The 2023 OIG review of the hypertension project found 16 of 60 sampled automated claims inaccurate and directed VBA to compare automated and traditional work side by side; VA now reports ADS claims at 96.4 percent against 95.5 percent in its quality review. [CC12][CC2] A vendor that cannot supply its own per-code accuracy series, separated from the national average, will be asked for it.

Third, no final decision by the tool. Orifici's statement that a trained claims processor makes the final decision on every disability compensation claim is the program's stated architecture, and the April 2026 OIG finding on survivor benefits is what happens when a program departs from it. [CC18][CC13] The criterion is a decision point with a named human role, not a slide that says human in the loop.

Fourth, an operational exception queue. The AICES PIA describes off-ramping data that cannot be accurately extracted and validated to human review. [CC5] Before scaling, a receiving office needs the exception queue defined, a review owner named, source-document access for the reviewer, and a workload estimate for that reviewer. A team that shows up with those four items has answered the question the PIA raises.

Fifth, accuracy by diagnostic code. ADS expands one code at a time, from 26 to 202 and now toward 262. [CC2] A model's aggregate accuracy across all conditions is a marketing number; the number VA can use is accuracy for the 60 codes it plans to add next, which the budget says require more complex logic.

These are an analyst's reading of the record, not a published evaluation scheme, and a real solicitation may weight them differently. They are also the five things GAO's July testimony, which reported 15 of 43 disability compensation recommendations still open and flagged data quality and workforce expertise, would have a team ready to defend. [CC14]

7. The counterargument, stated fairly

Three objections deserve the page.

The first is that overtime, not technology, explains most of the gain, and that the FY2027 request is a bet on replacing a lever that worked with one that is unproven at the margin. Devlin credited overtime first and automation second, the budget carries $67 million against overtime, and the claims-examining workforce shrank 14 percent while output rose. [CC16][CC2][CC15] That reading is consistent with the record. The answer is that the $130 million line itself is the administration's statement that overtime is not the plan, which makes the automation buy more likely, not less. [CC3]

The second is that the accuracy story is weaker than the headline. The dashboard reports 93.94 percent issue-level accuracy over twelve months against a 96 percent target and 83.53 percent at the claim level, and the non-rating inventory carried 291,804 actions older than 125 days in June with no backlog measure of its own. [CC19][CC20] A capture team should assume that the next automation buy will be scored on whether it moves those numbers, and that quality evidence will weigh more than speed evidence.

The third is that the rating backlog has drifted up, from roughly 70,300 in late July to 73,868 on September 19, and that a buyer under pressure may reach for surge staffing again. [CC1][CC21] The FY2027 budget projects about 75,000 as a sustainment level, so the drift is near plan, and proximity to a projection is evidence about the projection. [CC2] The drift is small against the decline and worth watching for the direction of the next quarter's spending.

8. What to do in the next ninety days

Track 36C10X25P0051 on USAspending after October 2 for a modification, a new order to the same vendor, or silence, and read each as the section 3 decision tree describes. [CC6]

Send the ADS ownership inquiry to VBA's Office of Business Integration now, in writing, asking for the contract numbers and acquisition office supporting Automated Decision Support development and sustainment in FY2026. The answer is a matter of public record and closes the largest gap in this brief.

Build the five artifacts in section 6 before any solicitation appears: a source-linking demonstration, a per-code accuracy series, a decision-point diagram with named human roles, an exception-queue operating model, and a validation plan for complex-logic codes. VA has described each of them in a public document.

Watch the FY2027 appropriations process for the $130 million and $47.8 million lines. Under the continuing resolution through December 11, neither can start, and the House and Senate marks will show whether the automation line survives at the requested level. [CC17]

For the intake work, calendar March 27, 2027 and watch the first half of FY2027 for a sources-sought or RFI, and do not treat the expired parent vehicle as open. [CC7][CC8]

For teams selling into the Department of War side of the same claim, the executive order's October 8 deadline for records transfer is the metric event, and VBA's evidence supply chain is the model the receiving agency already runs. The measure to propose is the share of service treatment records arriving complete within the window. [CC22]

Mary

Mission Meets Tech Premium

---

The views expressed in this newsletter are my own and do not represent the official position of any organization. This content is for informational purposes only.

---

Sources

[CC1] Veterans Benefits Administration, "Monday Morning Workload Report," national rating bundle for the week ending September 19, 2026. Source for: 664,263 pending, 73,868 over 125 days, 3,009,832 FYTD completions, and the 75.57-day FYTD average days to complete. https://www.benefits.va.gov/REPORTS/mmwr/2026/MMWR-09-19-2026.xlsx

[CC2] U.S. Department of Veterans Affairs, "FY 2027 Budget Submission, Volume 3: Benefits and Burial Programs and Departmental Administration," April 2026. Source for: ADS scope and FY2027 expansion plans (printed p. 69), the September 2025 phase baselines and NWQ modernization projection (printed p. 76), the VCIP $425.5 million 2027 budget-planning requirement (printed p. 72), the $118.784 million C&P Claims Modernization request (printed p. 68), the $67.0 million overtime allocation, VCIP volumes, and the approximately 75,000-claim FY2027 backlog projection. https://department.va.gov/wp-content/uploads/2026/04/Volume-3.pdf

[CC3] Office of Management and Budget, "Budget of the U.S. Government, Fiscal Year 2027," Department of Veterans Affairs chapter, printed p. 56. Source for: $130 million for VBA automation and AI investments in claims processing, described as reducing dependence on surge staffing and extra labor costs. Proposal, not enacted. https://www.govinfo.gov/content/pkg/BUDGET-2027-BUD/pdf/BUDGET-2027-BUD-18.pdf

[CC4] U.S. Department of Veterans Affairs, "FY 2027 Budget Submission, Volume 5," printed p. 119. Source for: $47.759 million requested for FY2027 against $43.070 million enacted for FY2026, a $4.689 million or 10.9 percent increase, for department-wide automation, decision support, and AI infrastructure and governance. https://department.va.gov/wp-content/uploads/2026/04/Volume-5.pdf

[CC5] U.S. Department of Veterans Affairs, "Artificial Intelligence Claims Evaluation System (AICES) Privacy Impact Assessment," FY2026, May 2026. Source for: identification of SteerBridge and contract 36C10X25P0051; the tool's source-system interfaces, upload of completed DBQs to VBMS, and off-ramp of unextractable data to human review; and the program office. https://department.va.gov/privacy/wp-content/uploads/sites/5/2026/05/FY26ArtificialIntelligenceClaimsEvaluationSystemAICESPIA.pdf

[CC6] USAspending.gov, award CONT_AWD_36C10X25P0051, SteerBridge Strategies LLC. Source for: $4,998,352.63 obligated and base plus options, award August 4, 2025, current and potential end October 2, 2026, and the ACE AI proof-of-concept pilot title. https://api.usaspending.gov/api/v2/awards/CONT_AWD_36C10X25P0051_3600_-NONE-_-NONE-/

[CC7] USAspending.gov, award CONT_AWD_36C10D23N0026 under parent 36C10E19D0016, Deloitte Consulting LLP. Source for: $48,282,895.46 obligated, $51,340,053.61 base plus all options, current end March 27, 2027, potential end March 27, 2028. https://api.usaspending.gov/api/v2/awards/CONT_AWD_36C10D23N0026_3600_36C10E19D0016_3600/

[CC8] USAspending.gov, parent award CONT_IDV_36C10E19D0016, Deloitte Consulting LLP. Source for: the $900 million base plus options amount and the March 8, 2025 end date. https://api.usaspending.gov/api/v2/awards/CONT_IDV_36C10E19D0016_3600/

[CC9] HigherGov, contract record 36C10B24P0010, Agile Six Applications. Source for: the ABD applications purchase order, end date April 15, 2025, and reported value of $4,230,661. Secondary aggregation of federal award data. https://www.highergov.com/contract/36C10B24P0010/

[CC10] USAspending.gov, award CONT_AWD_36C10B25N00010003 under parent 36C10B25D0001. Source for: the "Cross Benefits Crew" description. https://api.usaspending.gov/api/v2/awards/CONT_AWD_36C10B25N00010003_3600_36C10B25D0001_3600/

[CC11] USAspending.gov, parent award CONT_IDV_36C10B25D0001. Source for: identification of the parent as the SPRUCE IDIQ with a different named recipient. https://api.usaspending.gov/api/v2/awards/CONT_IDV_36C10B25D0001_3600/

[CC12] VA Office of Inspector General, "Improvements Needed in VBA's Claims Automation Project," 2023. Source for: 16 of 60 sampled automated hypertension claims inaccurate and the recommendation to compare automated and traditional processing. https://www.vaoig.gov/reports/review/improvements-needed-vbas-claims-automation-project

[CC13] VA Office of Inspector General, "Review of Automated Decisions for Veterans' Service-Connected Death Claims," April 2026. Source for: DIC decisions automated without human involvement, about 2 percent legal error, and approximately $2.7 million in improper payments; recommendations open as of the report date. https://www.vaoig.gov/reports/review/review-automated-decisions-veterans-service-connected-death-claims

[CC14] OrangeSlices AI, "VA expands AI use as GAO highlights risks in disability claims modernization," July 2026. Source for: summary of GAO's July 13, 2026 testimony reporting 43 disability compensation recommendations with 15 open and flagging data quality and workforce expertise. Secondary; the GAO testimony itself is the primary record. https://orangeslices.ai/va-expands-ai-use-as-gao-highlights-risks-in-disability-claims-modernization/

[CC15] Federal Hiring Data, "VA disability claims, AI, and the workforce in 2026." Source for: the reduction of 2,950 claims examiners, about 14 percent, September 2024 to June 2026. Third-party analysis of OPM data. https://www.federalhiringdata.com/articles/va-disability-claims-ai-workforce-2026

[CC16] Margarita Devlin, Principal Deputy Under Secretary for Benefits, statement before the House Committee on Veterans' Affairs, April 15, 2026. Source for: attribution of gains to the targeted use of overtime and automation assist tools, and the NWQ routing goal. https://www.congress.gov/119/meeting/house/119134/witnesses/HHRG-119-VR00-Wstate-DevlinM-20260415.pdf

[CC17] Congressional Research Service, R49353, on the Continuing Appropriations Act, 2027, H.R. 6500, signed September 2, 2026 as P.L. 119-103. Source for: funding through December 11, 2026. https://www.congress.gov/crs-product/R49353

[CC18] Robert Orifici, Veterans Benefits Administration, statement before the House Committee on Veterans' Affairs, Subcommittee on Disability Assistance and Memorial Affairs, July 13, 2026. Source for: TERA memo automation using verified data with links to source documents, and the statement that a trained claims processor makes the final decision on every disability compensation claim. https://docs.house.gov/meetings/VR/VR11/20260713/119443/HHRG-119-VR11-Wstate-OrificiR-20260713.pdf

[CC19] Veterans Benefits Administration, "Detailed Claims Data," dashboard marked updated September 21, 2026. Source for: 93.94 percent twelve-month issue-based accuracy, 83.53 percent twelve-month claim-based accuracy, and the 96 percent target. https://www.benefits.va.gov/reports/detailed_claims_data.asp

[CC20] U.S. Department of Veterans Affairs, response to questions for the record, House Committee on Veterans' Affairs hearing of April 15, 2026, posted July 15, 2026. Source for: 291,804 non-rating actions older than 125 days as of June 2026 and the absence of a non-rating backlog measure. https://docs.house.gov/meetings/VR/VR00/20260415/119134/HHRG-119-VR00-20260415-QFR002.pdf

[CC21] Veteran Benefit Desk, "VA Claims Backlog Tracker," reading of September 19, 2026. Source for: the increase of 3,576 since July 25, 2026, from which the late-July level is derived. Third-party tracker computed from VBA weekly reports. https://veteranbenefitdesk.com/va-claims-backlog

[CC22] Executive Order 14426, September 8, 2026, 91 FR 58003. Source for: the 30-day records transfer requirement and the October 8, 2026 deadline.

Sources verified as of September 28, 2026.