Measure Verification of Benefits First: The 90-Day Baseline That Settles the AI Argument in a 60-Bed Program

Which process would you bet payroll on?

Suppose someone hands you $12,000 and a rule: you may point it at exactly one process in your program, you must be able to show the board in 90 days whether it worked, and if you cannot show it you stop. Most owners answer with marketing copy or scheduling. Both are wrong, and for the same reason — you cannot cleanly attribute the result.

The right answer in a substance use treatment program is verification of benefits. It is measurable to the minute, it happens dozens of times a week, it has a documented error cost, and it sits directly on top of the two numbers your business runs on: how fast a person gets to a bed, and how much of the bill you actually collect.

Why verification of benefits and not the group notes?

Deloitte's State of AI in the Enterprise 2026 found 84% of organisations investing in AI report positive returns. The interesting part is not the percentage, it is what the winners have in common. They picked one messy process with a lot of repetition, kept a human reviewing every output, captured a baseline before switching anything on, and only widened scope after the numbers moved. That is it. There is no fifth ingredient.

Measuring first means writing down what a process costs you today — in minutes, in errors and in dollars — before any software touches it, so that the comparison in 90 days is arithmetic instead of an opinion. Verification of benefits fits that shape better than anything else on your org chart. Group notes are clinical and slower to judge. Marketing is contaminated by season and by whatever your competitor is bidding. A VOB either matched the payer's actual adjudication or it did not.

What does one VOB actually cost you today?

Watch an admissions coordinator do one. She has a name, a date of birth and a photo of a member card taken at arm's length. She goes to Availity or the payer's portal, and if the plan is not in there she calls the number on the back of the card and sits in an IVR queue. She is pulling deductible met, out-of-pocket maximum, coinsurance, whether residential, PHP and IOP are covered benefits, whether preauthorization is required and by whom, whether out-of-network benefits exist, and whether the plan is self-funded — because a self-funded plan changes the entire single case agreement conversation.

Then she writes it into the CRM, builds the estimate the family is quoted, and moves to the next one. Fast plans take six minutes. A self-funded plan with a carved-out behavioral health administrator and a hold queue takes forty. The average is the number that matters and almost nobody measures it.

The error cost is worse than the time cost. A misquoted out-of-pocket — you told the family $3,000, the actual exposure is $11,500 — shows up on day nine as an angry phone call, and it ends in one of two places: an AMA discharge with an empty bed, or a write-off in your aging report that your billing company quietly stops chasing at 120 days.

flowchart TD
  A["Pull 90 days of completed VOBs from the CRM"] --> B["Write down the baseline: minutes per VOB, misquote rate, write-offs"]
  B --> C["Assistant drafts each new VOB, coordinator still verifies every field"]
  C --> D{"Did the draft match the portal on all six fields?"}
  D -->|No| E["Coordinator corrects it and logs the miss by payer"]
  E --> C
  D -->|Yes| F["Week 4: compare the same three numbers to baseline"]
  F --> G{"Moved more than normal month-to-month swing?"}
  G -->|No| B
  G -->|Yes| H["Widen to preauth submissions, human sign-off stays"]

What five numbers do you write down before you switch anything on?

Capture these for 90 days of history, before anyone demos anything to you. First, average minutes per completed VOB, split into portal-only and phone-required — your coordinators can time-log two weeks of this honestly if you tell them it is not a performance review. Second, elapsed time from first inquiry call to VOB complete, which lives as timestamps in your CRM. Third, the misquote rate: admissions where the quoted patient responsibility differed from the adjudicated amount by more than 20%, which you get by joining the CRM estimate to the posted claims in your billing system.

Fourth, dollars written off in the last four quarters attributable to benefit misquotes, tagged by hand if your write-off reasons are sloppy — most are. Fifth, inquiry-to-admit conversion split by how long the VOB took, because the correlation between a same-hour VOB and an admission is the number that usually surprises the owner most.

Those five numbers are the whole project. If you never buy any software, you have still learned something worth the two weeks.

What does the 84% actually have in common?

The shape of the working version, on a Tuesday: the inquiry comes in at 9:12am, and the assistant reads the member card photo, checks the payer portal, and returns a completed benefit summary into the CRM with every field marked as either confirmed from the portal or unconfirmed. The admissions coordinator opens it at 9:16, verifies the four fields that drive the estimate, fixes what is wrong, and marks it approved. Her job changed from data entry to checking. She still owns the number that gets quoted to the family.

Every correction she makes goes on an exception log, tagged by payer. Four weeks in, that log tells you something operationally useful whether or not you keep the tool: which three payers produce most of your errors. Programs consistently find it is the same handful — a regional Blues plan with a carved-out behavioral administrator, one national plan's self-funded book, and whatever state Medicaid managed care organization you deal with least often.

Does the arithmetic clear the bar?

Illustrative for a 60-bed residential program with a PHP and IOP track. Use your own numbers.

That is roughly $7,650 a month, about $92,000 a year, and none of it depends on believing a vendor. Every line is a number you already have or can start logging on Monday. If after 90 days the misquote rate has not moved, you stop — and that is a successful project too, because it cost you two weeks of measurement instead of a year of assumption.

Where does the human stay?

The coordinator signs the estimate. Always. A benefits portal is a snapshot that can be wrong, plans change mid-month, and the family is making a decision about money they may be borrowing. A person's name goes on that number.

Single case agreement negotiations stay human, because they are a conversation with a network representative about a specific clinical situation, and what moves the number is a relationship somebody on your team has built. Preauthorization submissions can be drafted by the assistant but reviewed by your UR coordinator before they go, because what you assert about medical necessity at intake follows the chart for the whole stay.

And the compliance layer applies from minute one. A prospective patient's benefit information is protected under both HIPAA and 42 CFR Part 2, whose alignment rule's compliance date came in February 2026. Signed business associate agreement, no training on your data, documented access controls — before the first live VOB, not after.

Frequently asked questions

Our billing is outsourced. Does that change the project?

It makes the baseline harder to pull and more important to have. Ask your billing partner for the last four quarters of adjustments with reason codes, and for the claims data you need to compare quoted patient responsibility against adjudicated amounts. If they cannot produce it in two weeks, that itself is a finding.

Why 90 days rather than 30?

Because claims lag. A VOB done in April does not show up as a misquote until the claim adjudicates and the patient statement goes out. Thirty days measures the time saving only; ninety measures the money.

Won't our January volume distort the comparison?

Yes. Deductibles reset on 1 January, holiday-delayed admissions land in the first two weeks, and out-of-pocket exposure in Q1 looks nothing like Q4. Compare against the same quarter last year, or run the baseline and the test in the same season.

Do we need a full-time person to run this?

No. It is a defined project for your admissions director and whoever owns the CRM, roughly a day a week for a quarter. The failure mode is not the workload — it is nobody owning the measurement, so the review meeting turns into opinions in month four.

If the same measure-first discipline is applied to the phone line rather than the benefit check, the baseline is abandoned calls and inquiry-to-admit conversion by hour. CallSphere builds AI voice and chat agents that answer phone lines and web chat 24/7, capture the caller's details and book the callback — and the same rule applies: capture your abandonment rate before you turn anything on, so you can tell in 90 days whether it moved.