Case study: Keeping the MLR Queue Moving, From Triage to Review Built Into Creation

Since December 1992, a medicine approved under FDA's accelerated approval rules has had to show the agency its promotion before using it, unless FDA says otherwise. From 120 days after approval, 21 CFR 314.550 requires each promotional piece to be submitted at least 30 days before first use. The biologics rule, 21 CFR 601.45, is word for word the same. In the UK, the MHRA's vetting guidance, updated in June 2023, asks for at least three weeks' notice. It then takes material one piece at a time, and the next piece waits until comments on the last have come back.

Those clocks sit on top of ordinary promotional materials MLR review, in which medical, legal and regulatory reviewers check every piece before use. A queue that ignores them runs first in, first out. A reprint waits behind a launch campaign. A piece with a 30-day clock waits behind one with none. The committee reads both to the same depth. We built two products for pharma companies to deal with that. One is a dedicated MLR review orchestration and triaging platform. The other is a marketing asset creation and management platform with MLR review built in.

The situation: review queues fail on order as well as on reading

We built for pharma companies. The public figures on how long their review takes are all vendor-collected, so treat them as direction rather than benchmark. Veeva wrote in February 2022 that its data showed most biopharmas take three weeks on average to get new content to market. About six days of that is spent in draft before review starts, and about 15 days in review itself. Indegene's Future of MLR Review report puts the manual state at about 21 days, three rounds and $2,500 to $5,000 per asset. It cites IQVIA, MM+M, McKinsey and pilot data.

Recent AI products in MLR, such as Veeva's Falcon MLR, target the read itself: checks against approved labels and local rules. Our analysis of what agentic promotional review automates covers that ground. This case study is about the flow around the read, where four things break.

Everything gets the same review. A piece built from already-approved content and a piece carrying a new claim go to the same committee at the same depth. Low-risk pieces take reviewer time at the same rate as high-risk ones.

Deadlines are invisible to the queue. The 30-day clock, filing at first use and the regulator's vetting sequence live in people's heads and calendars. They do not set the order of work.

Recertification arrives on a calendar. Clause 8.5 of the 2024 ABPI Code requires material still in use to be recertified at intervals of no more than two years. That is a second stream of work that does not come from new campaigns.

Compliance comes in too late. Indegene's account of an agentic campaign project for a global pharma company names the pattern directly. Regulatory and MLR concerns arrived near the end of creative development, which increased rework and slowed approvals.

What we built

We built two products, both shipped, both for pharma companies.

A dedicated MLR review orchestration and triaging platform. Triage is the decision about what a submitted piece needs: how deep a review, from whom, and how urgently. Orchestration is what follows: getting the piece to those reviewers in the right order and keeping it moving until it is approved or sent back.

A marketing asset creation and management platform with MLR review built in. Here review is part of making and managing the asset, not a step that starts after hand-off.

The two belong together. The first manages what arrives at review. The second changes what arrives. A triage step can only sort the work in front of it, and a creation tool with review built in can only reduce the work that reaches the committee. Neither fixes the other half.

Whatever the implementation, any MLR triage has to know certain facts about a piece. The rules and the work itself put them on the table.

What it achieved, and what the numbers in this market measure

Both platforms are shipped. We publish no cycle-time, cost or volume figures in this case study. The figures already public in this market are worth reading closely, because none of them measures what triage changes.

The two library cases deserve a closer look. The UCB story is a testimonial from the company's head of US content clearance: speed to market fell "from weeks to days and hours". The page gives no baseline, sample or method. It tells you a sponsor credits its content system for faster launches, and nothing about which part of the change did the work.

The Indegene case is more useful for its diagnosis than its numbers. Its outcomes measure concept generation in a three-day workshop: about 100 concepts, of which 11 went forward to further evaluation. It describes an MLR pre-flight layer that flagged regulatory risks before formal review. It reports nothing about approval time, rounds or what the committee later found.

What to measure instead, by tier, on your own queue:

• Time waiting before first review, separate from time in review. Veeva's six days in draft is the kind of number that hides here.
• Rounds per asset and first-pass approval rate. Our analysis of MLR agent recall argued that rounds, not reading hours, are where the return shows.
• Under-triage: pieces sent down a light path that later drew major comments, a refusal or a regulator's objection. This is the error that matters.
• Over-triage: simple pieces sent to full committee. This costs time, not compliance.
• Deadline performance: 2253 filed at first use, accelerated approval pieces submitted at least 30 days ahead, and vetting submissions in the order the regulator asked.
• Recertification backlog: material in use that is past its two-year date.
• For review built into creation: the share of pieces that reach formal review with no new claims, and how their rounds compare.

What makes triage work, and what limits it

The errors are not symmetric. Sending a simple piece to full committee wastes time. Sending a risky piece down the fast lane can put a violation into the field. Thresholds should be set so that the second error is rare, and a sample of fast-lane decisions should get a full review to prove it.

Triage cannot route around the signature. Under Clause 8.1 of the ABPI Code, one person certifies the final form for the company. That person must not be the one who drew the material up. Clause 8.4 requires signatories to be notified in advance to the MHRA and the PMCPA. Routing can only send a piece to people who are allowed to sign it, and must keep author and certifier apart.

Some risk is only visible in the whole piece. FDA's enforcement in 2025 and 2026 has turned on net impression: what the whole piece conveys, not single claims. Our analysis of why claim-by-claim checking misses net impression explains the gap. A tier based only on claims can rate a TV spot as low risk because every sentence is approved. Format, channel and audience have to count as risk signals too.

The regulator's queue is not yours. MHRA reviews one piece at a time and recommends sending a key piece, such as the detail aid, first. FDA's rule fixes 30 days between submission and first use. Orchestration can sequence your side of those queues. It cannot shorten theirs.

Review built into creation stops at the final form. Building review into how assets are made reduces what is wrong when they reach the committee. Vendors are moving the same way: Veeva made its Quick Check Agent, which scans content against brand and compliance guidelines before MLR review, available in December 2025. None of it certifies anything. The ABPI Code requires the final form, with no further changes, to be certified. For dynamic content such as websites, the Code does not require every possible combination to be certified. It does require the context in which content appears to be considered.

The human in the loop needs evidence. If reviewers rely on what an earlier stage cleared, their override and review-time data are what show the control works. Our piece on qualifying the human in the loop sets out the two numbers to keep.

The rules it had to satisfy, position current at 12 October 2026

None of the US rules, the ABPI Code or the MHRA vetting guidance cited here mentions artificial intelligence. The obligations are the ordinary advertising rules, and they apply to whatever produced the piece.

Article 50 matters more to asset creation than to triage: where a tool generates content, the transparency duties for generative AI apply. The Commission says its code of practice is voluntary but the Article 50 obligations are legal ones.

What this means in practice

For the same result, work in this order, longest lead time first.

1. Measure the queue you have. Export a year of MLR submissions with timestamps. Split time waiting from time in review, and count rounds by asset type. This comes first because every later step is scored against it.
2. Write the triage rules with the people who sign. Define the tiers, what goes where, and who answers for a fast-lane decision. Check the routing against your notified signatory list and keep author and certifier apart.
3. Put the clocks into the queue as data. First-use dates for 2253 filing, 30-day dates for accelerated approval pieces, vetting sequences, and two-year recertification dates.
4. Decide what moves into creation. Reuse of approved content and checks while the asset is made can move upstream. Certification of the final form cannot.
5. Pilot on one brand in one market. Measure under-triage by giving a sample of fast-lane pieces a full review.
6. Only then add an AI reviewer. An agent that reads faster helps most once the right pieces reach it in the right order.

Three people own this. The head of promotional review owns the tiers and routing, the nominated signatories decide what may skip full committee, and regulatory affairs owns the clocks. If those three cannot agree a tier definition in the first meeting, the project is not ready to start.