Stop Buying Throughput. Start Buying Capacity, Outcomes and Accountability.

As AI takes over the routine delivery layer of government health work, agencies should rethink what they’re paying for. 

By Josh DeLung   |   VP, Population Health, Altarum

For most of the last two decades, the government health market has rewarded contractors for one capability above all others: the ability to produce a lot of work, fast, at scale. Process the dataset. Draft the toolkit. Stand up the training logistics. Turn around the quarterly report. Throughput was the product, and the firms that could put the most qualified people against the most tasks tended to win the most work.

That era is closing, and artificial intelligence (AI) is the reason.

The routine delivery layer of public health contracting, including data cleaning and processing, first-draft content, the logistics behind technical assistance (TA), and standardized reporting, is exactly the kind of work AI now does faster and cheaper than a staffed team can. I’m not speculating about some future state. It’s already happening in the work my colleagues and I deliver. The capacity to generate output is becoming a commodity. And when something becomes a commodity, you stop paying a premium for it.

So here’s the uncomfortable question every government health agency at the federal, state, or local level should be asking: if AI can absorb the throughput, what exactly are you buying a contractor for?

The answer is the part AI can’t supply. AI can draft a TA curriculum; it can’t tell you whether a rural health department will adopt it, or what must change in a clinic’s workflow for the guidance to stick. AI can summarize an evaluation; it can’t sit with a state Medicaid director, weigh the political and operational constraints, and decide which finding is worth acting on first. AI can produce a communications plan in seconds; it can’t read the room in a community that has good reason not to trust the message. That judgment, grounded in domain expertise, implementation science, and a focus on capacity building for the agency, with clearly defined outcomes, rather than just deliverables, is harder to buy than throughput. Real value comes from bringing the lived experience, accountability, and tools to help agencies carry on long after the contractor is gone. TA experts who will weigh tradeoffs and understand the downstream impact of policy scenarios have a stake in the public health outcomes of agencies’ work that AI doesn’t. AI will hand you an analysis; it has nothing riding on what you do with it. That accountability, a partner on the hook for whether the work improves health and adds capacity for the agency long-term, is the scarce resource now.

Health agencies should stop buying throughput and start buying partners. The distinction matters, and it should show up in how agencies write requirements and evaluate bids. Evaluators need to look for contractors that bring strategic judgment to scope the right work in the first place. On a training and TA contract, agencies have long paid for sessions delivered and people reached. But the metric that matters is whether the assistance changed practice, whether a program is doing something differently with measurably better health outcomes for constituents a year later. On communications support, the buy should not be about more content, channels, or impressions. The differentiator is the strategy underneath it, the ability to translate science to behavior-changing storytelling, and the human judgment about audience, trust, and timing that no model can generate on its own. And on data and analytics work, the dashboard is no longer the hard part. Knowing which question is worth answering, and what the agency should do with the answer, is.

In every one of those cases, production can be accelerated by AI. The judgment about what to produce, for whom, and to what end still needs a human owner. The organizations worth partnering with are the ones that pair the two on purpose: a person accountable for the strategy and invested in the outcomes, with AI accelerating the work on both sides of it.

That’s the model we’ve built at Altarum: mixed-methods research and implementation science to understand what’s actually happening on the ground, flowing into the discipline of TA that’s treated as a craft rather than a logistics line item. Responsible AI does the heavy lifting on administrative production so our people spend their time on the important human-to-human work that measurably improves health outcomes and builds agency capacity to sustain the effort over time. We’re not there to secure vendor lock-in; what government agencies and TA recipients need is the ability to improve health outcomes long after we’re gone.

We also know if AI is doing the routine work, human work must be genuinely better: sharper analysis, harder questions, real accountability for whether an intervention worked. That accountability matters most when the news is bad: a partner worth paying will tell a program office an initiative isn’t working, and should be redirected before it burns another year of budget, even when the easier move is to renew quietly and keep the deliverables flowing. That isn’t integrity for its own sake. It’s what protects the mission and people’s health.

The opportunity for agencies is bigger than cost savings. For years, the structure of government health contracting has quietly rewarded activity over impact — reports filed, sessions held, deliverables shipped. AI removes the excuse. When the routine layer costs a fraction of what it used to, the budget and the attention can move to the part that changes outcomes: the strategy, the implementation, the judgment about what works for this population, in this place, right now.

Government must partner with organizations that demonstrate an ability to drive outcomes-focused capacity building instead of entrenched dependency. That’s the future of government contractor health services worth buying, and it’s exactly why Altarum has set an audacious goal to positively impact the health of 100 million Americans by 2036.