Every fall, it's the same scene. You sit in a conference room, someone clicks to a slide that says "Projected Medical Trend: 8.5% to 10.0%," and everyone nods. Budgets get locked. Plans get set. What the slide doesn't show is that that forecast is systematically inflated by broken data systems, not by bad math.
I've spent years inside these models, and I can tell you the dirty secret. Medical trend forecasts aren't predicting the future. They're extrapolating a backward-looking, flawed past through a lens of administrative lag. The result is a forecast with a built-in upward bias, year after year. And almost nobody is talking about it.
Three structural distortions hide inside every trend model you've seen, and each one has a fix.
1. The Ghost Utilization of Service Code Churn
Traditional trend models see a CPT code and assume it means real utilization. If code frequency goes up, the model screams "inflation!" But the cause is usually something else.
Billing systems have changed. Under value-based care and reference-based pricing, providers now have every incentive to unbundle services. A routine office visit in 2019 was one code: 99213. The same visit in 2024 is the 99213 plus a remote monitoring code, plus a behavioral health screener, plus a visit-complexity add-on that Medicare began paying in 2024. Same patient. Same visit. Three times the codes.
- The systems problem: The model sees a 20% increase in "service units" and calls it utilization inflation.
- The trend skew: This is billing disaggregation, not inflation, and it can inflate your forecast by a percentage point or two every year.
You're budgeting for phantom visits that never happened. That's noise, not trend.
2. The Actuarial Echo Chamber of Rx Carve-Outs
The biggest driver of medical trend today is GLP-1s and gene therapies. Your forecast is using old data to measure them.
Most large-group trend models rely on claims data that's 6 to 9 months old. In that window, GLP-1 use keeps climbing: the drugs grew from 6.9% of employers' annual claims in 2023 to 10.5% by 2025, per the International Foundation of Employee Benefit Plans. But the rebate and discount data from your PBM hasn't caught up yet. Say the model sees $1,200 in gross drug cost per member. It has no idea about the $600 in anticipated rebates that won't land until months after the budget closes.
- The systems problem: The forecast projects gross cost, not net cost. Rebates arrive long after the budget is set.
- The trend skew: This creates a phantom inflation point that vanishes once the rebates finally land. You're forecasting cash flow, not true cost.
The gap is not hypothetical. Over 2012-2017, U.S. drug prices rose 12% a year at list and only 3% at net once rebates were counted, according to an NBER working paper on pharmaceutical rebates.
3. The Asymmetric Regret Bias
This last one is the most uncomfortable, and the most accurate. Carriers and TPAs don't build forecasts to predict reality. They build forecasts to set renewal rates and reserve levels.
The incentives run one way. If a TPA forecasts 8% trend and actual comes in at 6%, everyone cheers: "We managed risk well!" But if they forecast 6% and actual hits 8%, the client is furious. They might switch carriers. The fear of being wrong on the low side is far stronger than the fear of being wrong on the high side.
- The systems problem: That asymmetry, what economists call regret aversion, gets baked into the model as a conservatism buffer hidden in the standard deviation assumptions.
- The trend skew: This buffer is a risk management hedge dressed up as a forecast, not a prediction. You're paying for someone else's fear.
What to Ask Your Consultant Next Time
Don't ask "What is your trend forecast?" That question invites a number that looks authoritative but is structurally inflated. Instead, ask two questions:
- "What is your model's assumed data lag?" If they can't give you a specific number of months, they're guessing.
- "What percentage of your trend increase comes from billing code expansion versus true unit cost growth?" If they don't know, they haven't cleaned the data.
If they can't answer both, they're extrapolating, not forecasting.
Who Can See the Data
One caveat decides whether these questions get real answers: data access. Fully insured employers don't hold their own claims. The carrier produces the renewal and the trend number, and the employer can't independently check the code mix, the rebate timing, or the data lag. You can ask both questions, but if the answer is a carrier deck, you're still taking the number on faith.
Self-funded employers sit on the other side. Their TPA passes through claims-level detail, so the employer, or a consultant with a data license, can test the model's assumptions directly: the paid runout, the share of unit growth that comes from new codes, and the net Rx cost after rebates settle. That's the only setting where "show me net trend, not gross" is a demand someone can meet.
None of this needs a bigger budget. It needs a contract that gives you your own data and a partner who will explain the lag instead of burying it. If you can't get that, treat any forecast as a number to budget against with a cushion, not as a truth.
The Real Fix
Medical trend is a system output, not a natural disaster. And systems can be fixed.
The future of benefits cost management is collapsing data latency, not predicting trend better. Strip away the administrative noise by modeling episodes of care rather than individual service lines. Episodes group the visit, the lab, and the follow-up into one unit, so splitting a single visit into three codes stops looking like three times the utilization. Pull real-time Rx rebate data into your cost projections. Demand that your partner show you net trend, not gross.
Stop budgeting against a forecast that whispers 8%. Start budgeting against the data truth. Your bottom line will thank you.