7 minute read

TL;DR: The Pentagon has rebranded Advana as the War Data Platform, and Accenture Federal Services just won a task order worth up to $821 million to integrate it. The new name is not the story. A Department of War Inspector General report from May found that 78 of 387 source systems were connected to the platform without data sharing agreements, and the office that runs the platform disagreed with 10 of the 12 recommendations it received. That is the story. You cannot contract your way to trustworthy data, and every AI capability the Department builds on this foundation inherits whatever is underneath it.

What Actually Happened

On July 9, DefenseScoop reported that Accenture Federal Services had been selected for a five-year task order worth up to $821 million to provide core integration support for the War Data Platform. The award ran through the General Services Administration’s Alliant 2 vehicle, and Accenture beat four other bidders. Accenture, the Department, and GSA all declined to say much about it.

The War Data Platform is what Advana became. On January 12, 2026, Secretary Pete Hegseth signed a memo titled “Transforming Advana to Accelerate Artificial Intelligence and Enhance Auditability.” It directed the Chief Digital and Artificial Intelligence Officer to break Advana into three pieces: the War Data Platform for standardized data integration, Advana for Financial Management for audit remediation, and WDP Application Services to migrate everything else onto the new architecture. CDAO owes the Secretary’s office a status update every 45 days until the platform reaches full operational capability.

If you have not tracked Advana, the short version: it started inside the Comptroller’s shop as a way to pull together data from the Department’s 400-plus business systems. In 2021, Booz Allen Hamilton won a five-year, $647 million contract to expand it. CDAO later took it over, and it became the Department’s default answer to the question of where the data lives.

So: a platform with years of history, a new name, a new org chart, and a new integrator.

From $15 Billion to $821 Million

Here is the part I would put on a chart if I were briefing this.

In the fall of 2024, Department officials floated a follow-on contracting plan worth as much as $15 billion. In July 2025, the Pentagon canceled the draft solicitation for the Advancing Artificial Intelligence Multiple Award Contract and put the program on hold. Now, a year later, what has actually landed is an $821 million integration task order.

I am not going to pretend $821 million is small. It is not. But the distance between a $15 billion vision and an $821 million integration award tells you something honest about where this program sits. The Department did not buy the grand unified data enterprise. It bought integration support for the layer underneath it, which is a more modest and, frankly, more sensible thing to buy.

The ceiling is also just a ceiling. A five-year task order worth “up to” $821 million is authority to spend, not money spent. I have watched too many people read a ceiling number as a program budget and then wonder why the capability never showed up.

The Report Nobody Is Quoting

Now the part that matters.

On May 7, 2026, the Department of War Inspector General issued report DOWIG-2026-079, “Audit of Interface Controls Over the DoW’s Advancing Analytics Repository.” I would encourage anyone selling into this program to read it before they write another line of their pitch.

The findings, in the IG’s numbers:

• 78 of 387 source systems were connected without data sharing agreements in place.
• 26 of 29 privileged users were granted access without proper verification.
• There was no effective process for monitoring user access.
• Validation checks for data accuracy were missing.
• There was no notification system to tell anyone when an interface error occurred.

The IG concluded that the office had “limited assurance” that data transferred into the platform met accuracy and completeness standards, and warned that deficient data could lead Department leaders “to make misinformed decisions affecting operations.”

The IG made 12 recommendations. The Office of the CDAO agreed with 2 and disagreed with 10, including the recommendation to immediately put the missing data sharing agreements in place.

Separately, the fiscal 2025 SOC-1 audit on Advana came back adverse, and officials said they were working to remediate the control deficiencies it identified.

Sit with that fifth finding for a minute. No notification when an interface error occurred means data could stop flowing, or flow wrong, and the system would not tell anyone. Every dashboard downstream would keep rendering. Confidently.

An Integrator Cannot Fix Governance

None of the IG’s findings are integration problems. Every one of them is a governance problem.

A data sharing agreement is not code. It is an agreement between two organizations about what the data means, who owns it, how good it has to be, and who gets called when it breaks. An integrator can build you the pipe. An integrator cannot make a program office in a different chain of command agree to be accountable for what flows through it.

This is the pattern I have watched repeat for years. Platform money is easy to defend in a budget drill because it produces a demo. Data governance money is nearly impossible to defend because it produces a spreadsheet of agreements and a slower yes. So the platform gets funded, the governance does not, and eighteen months later somebody writes the report we just read.

The $821 million will buy real engineering. It will not buy 78 signatures.

The 80 Percent Doctrine Meets Authoritative Data

Cameron Stanley took over as CDAO in January 2026, arriving from a national security technology role at Amazon Web Services. His stated posture is a shift from strategic enablement to disciplined delivery: fewer piecemeal efforts, results in months instead of two to three years, monthly progress reporting against defined metrics.

At the Potomac Officers Club AI Summit on March 18, he described the method plainly: “Put it in front of warfighters, have them give you feedback … and then iterate again.” The idea is not to chase perfection before fielding. Get it useful, then let contact with real users make it better.

I agree with that for the application layer. It is the right instinct, and it is how good software actually gets built.

It does not transfer to the data layer.

If a user interface is 80 percent right, the warfighter can see the other 20 percent and tell you about it, aggressively and with prejudice. That feedback loop works because the error is visible. If the underlying data is 80 percent right, nobody can see which 80 percent. The dashboard renders. The model returns an answer. The confidence interval looks reasonable. The error is silent, and it propagates into every decision made on top of it.

Stanley has said the WDP team wants to “accelerate the identification of true authoritative data” for warfighting systems. That is exactly the right goal, and I want to be fair here: he is pointed at the right problem. But authoritative is not a technical property you can iterate into existence. Authoritative means somebody with a name and a chain of command has said this is the source of truth and I own it. That is what those 78 agreements are. That is the thing the IG asked for and the office disagreed with.

This is my lane, so let me say it in test and evaluation terms: you cannot test your way to data quality at the end. T&E on an AI capability tells you what the model does with the data it was given. It does not tell you whether the data was right in the first place. If you want to know that, you have to instrument the pipeline and put somebody accountable at the other end of it. Nobody demos that. Everybody needs it.

So What

For the people I talk to most, here is what I would actually do about this.

• If you are selling into WDP: read DOWIG-2026-079 before you write another word of your capture plan. The IG just published the government’s own list of what is broken. Your discriminator is not another pipeline. It is helping them close 78 data sharing agreements and stand up error notification. That is unglamorous, and it is the actual requirement.
• If you are a program office feeding this platform: you are one of the 387. Find out whether you are one of the 78. If you are, that is a finding with your name on it, and it is cheaper to fix now than after the next audit.
• If you are building AI on top of this: your model inherits every gap underneath it. Ask where your data came from, who attests to it, and what happens when an interface breaks silently. If nobody can answer, you do not have a model risk. You have a data risk wearing a model costume.
• If you are watching the money: track obligations, not the ceiling. The $821 million is authority. Watch what actually gets placed against it over the next four quarters, because that will tell you whether this is a real program or a holding pattern with a new name.

Renaming Advana to the War Data Platform was the easy part. The Department has been saying data is the ammunition since the 2020 data strategy. The IG just pointed out that we are not inspecting the rounds.

Tags: AI