Public Benefits Eligibility Software: Custom vs Off-the-Shelf

What Curam, Conduent, Deloitte and Salesforce actually do well

Merative Curam, Conduent, Deloitte and increasingly Salesforce Public Sector Solutions hold the integrated eligibility market, and they hold it because the job is enormous. An eligibility system covers application intake, evidence, verification, determination across several programs, benefit issuance, case maintenance, overpayment recovery, quality control sampling, federal reporting and worker workflow, under audit from more than one federal agency at once.

They also carry the federal plumbing. Interfaces to the Federal Data Services Hub, state wage records, corrections and vital records. Reporting formats that change on somebody else's schedule. The parts of the system that are identical in every state, which is most of it, are exactly the parts you should never pay to rebuild.

And they carry the political weight of continuity. An eligibility system pays benefits to households with no financial margin. A cutover that goes wrong is not a project delay, it is people without food assistance in a month with rent due. Vendors who have run that transition dozens of times price that experience into the contract, and it is worth what they charge.

So the recommendation is unambiguous and it costs us the largest possible engagement: do not replace the system of record. Modular procurement is now the sanctioned architecture precisely because the alternative failed repeatedly, and the right question is which modules deserve to be yours rather than whether the whole thing should be.

Where they stop: a determination that cannot explain itself

Picture the hearing. A hearing officer, a claimant, and an eligibility worker with a printout. The household was found ineligible for food assistance. The claimant says her hours were cut in March and she reported it. The screen shows a countable income figure, and nobody in the room can say how it was built: which pay stubs, over what period, with what conversion factor, against which deduction set, under the rules in force on the determination date rather than the rules in force today.

That is the central defect, and it is architectural rather than a reporting gap. The system was built to produce an answer. The explanation is the legal product, and it was never designed. Policy logic ends up spread across configuration, code and interface behaviour, so no single artifact represents what the rule was on a Tuesday in March.

The second wall is the household. One application can touch Medicaid, food assistance, cash assistance and child care subsidy, and each program constructs a household differently. Medicaid for most adults and children uses a tax based construction, food assistance uses who purchases and prepares food together, child care uses its own definition again. The same three people are one household in one program and two in another, with different countable income in each. Systems that force one household model and apply adjustments produce the subtle errors that surface in quality control samples.

The third is renewal. Federal rules require an attempt to renew Medicaid from information already available before asking the enrollee for anything. During the 2023 unwinding of the continuous enrollment condition, CMS identified states running that check at household level rather than per individual, which forced pauses in procedural terminations and reinstatements. That was not a policy misunderstanding. It was a data model treating the case as the unit when the rule operates on the person.

The fourth is paper. Processing standards are short and real: expedited food assistance runs on a seven day clock and ordinary applications on thirty days, and every day spent waiting on a pay stub comes out of that. Meanwhile documents arrive by mail, by fax, by drop box, by phone photograph and through a portal, and matching each one to the right case and the right open verification requirement is manual work performed by people who could be determining eligibility instead. No packaged product has solved that, because the channels and the local practice differ in every office.

The arithmetic: cost per determination against the cost to build

Government buyers rarely see a per seat price, so build the comparison from your own operating numbers instead.

Take a module at $350,000 with support and enhancement at 17 percent a year. Over five years that is roughly $630,000, or $126,000 annually. If your agency issues 900,000 determinations a year, the module costs about fourteen cents per determination. Now price the same capability as a change order from your incumbent, spread across the same five years, and add the release calendar you will wait in.

The crossover sits near 100,000 active cases. Below that, a module rarely clears against a change order, because your fixed development cost divides across too few determinations and the incumbent is amortising the same work across other states. Above it, the numbers move quickly, and each additional program under the same rules service costs almost nothing extra.

Two operating figures belong in the same calculation and almost never appear. The staff hours currently spent matching documents to cases and to open verification requirements, which is professional time doing clerical work against a thirty day processing standard and a seven day expedited one. And the cost of a lost hearing, which includes the corrective payment, the worker time, and the pattern it sets when the same defect produces the same loss forty times.

The cost to build a module, and what year two adds

Across Digital Heroes delivery experience, a modular component in this domain runs $200,000 to $600,000 over 6 to 12 months. That buys a versioned rules and determination trace service, or ex parte renewal automation with source by source logging, or a notice generation and archive engine, or document intake with verification matching, integrated against the system of record you already run.

Budget two lines your funding request should name explicitly. Data migration and historical rule reconstruction runs 10 to 25 percent of build cost, and the expensive half is not data, it is establishing what the rules were during each past period so that replay produces the right answer. Year two onward runs 15 to 20 percent of build cost annually, and in this domain that line is policy driven rather than technical: legislative changes, federal adjustments, waiver expiry and disaster rules all arrive on their own dates.

What raises the number: program count, since Medicaid alone is one shape and four programs are four rule sets and four household constructions. Any component touching federal tax information, because IRS Publication 1075 turns the environment, staffing and logging into a security program with its own timeline. Data sharing agreements, where the engineering is small and the agreement, test environment and production credentials are not. CMS streamlined modular certification, which changes what you build and not only what you document. And language and accessibility obligations, which for benefits are broader than a typical government portal and belong in scope from the first screen.

Four situations where building a module wins

Regulatory fit. Your determinations are appealable and the burden of showing the calculation sits with you. If a March determination cannot be reproduced in October under the rules that applied in March, you are losing hearings you should win. Ex parte renewal per individual is the second case, for the reason CMS made national.

Scale economics. Above roughly 100,000 active cases, owning the rules service costs less over five years than buying the same changes as vendor work, and the gap widens with every program you add.

A workflow that is your own. Policy logic is public property and it is yours. State options, waivers and local practice are the parts no vendor can generalise, and they are exactly the parts that should live in a versioned artifact your own analysts can read, change and replay against real cases before it goes live.

Integration sprawl. Count them: the federal data hub, state wage records, the Medicaid management information system, child care, corrections, vital records and your document channels. Once three or more must agree about one person, identity and evidence handling is the actual product.

How to decide in a week

Pull five adverse determinations from eight months ago, ideally ones that went to hearing, and try to reproduce each of them exactly under the rules in force on the determination date. Give the task to a policy analyst, not an engineer, and time it.

You are looking for three things. Whether the values consumed can be traced to their sources. Whether the rule version applied can be named. And whether the notice the household received says the same thing the system says. If those three outputs disagree, or if reproducing one determination takes more than an afternoon, you have found the module worth funding and you can write the justification straight from the exercise.

Run a second check the same week: sample fifty renewals and count how many completed without touching the household. That share is the single number worth designing the renewal component around, and most agencies have never measured it per individual.

Then commission a paid discovery phase of two to four weeks that ends with a signed product requirements document: the rule artifact structure, the trace format, integration inventory with the agreements each one needs, certification outcomes to be evidenced, acceptance criteria and a fixed price. Digital Heroes writes that before any code and the agency keeps it regardless of who builds. Contracts sign through our India LLP, US LLC or UK LTD so the work assigns under your own law, and on this kind of program we would tell you to walk away from any firm that treats code ownership as negotiable.

Book a 30-minute call with Digital Heroes and get a written plan and a fixed quote within 48 hours.