Reducing Process Mass Intensity in API Manufacturing with AI

Every kilogram of API carries hundreds of kilograms of materials

A kilogram of active pharmaceutical ingredient is never just a kilogram. Behind it sit reagents, catalysts, water and, above all, solvents: often hundreds of kilograms for each kilogram of product. Most of that mass is bought, handled, heated, recovered or incinerated, and then thrown away.

That is why Process Mass Intensity (PMI) has become the pharmaceutical industry's headline green chemistry metric. It is simple to calculate, hard to game, and it tracks cost almost as closely as environmental impact. A lower PMI usually means fewer raw materials, smaller equipment, less waste treatment and a smaller carbon footprint.

The gains can be dramatic. When Merck redesigned the manufacturing route for its antiviral letermovir, the new process cut PMI by 73% and raw material costs by 93%, while raising overall yield by more than 60%. Sustainability and margin moved in the same direction.

This guide is for process chemists and CMC leads who want to make those gains routine rather than exceptional. It covers:

• what PMI measures, and how to calculate it without the usual mistakes
• where the mass actually goes in an API process
• realistic benchmarks by drug modality
• seven proven levers, from quick wins to a new route
• why improvement usually stalls before commercial manufacture
• how AI-driven optimisation and autonomous labs change the economics

It builds on our guide to eco-design for industrial processes, where PMI is introduced as one of the core metrics.

What PMI measures, and how to calculate it

The ACS Green Chemistry Institute Pharmaceutical Roundtable uses PMI to benchmark the greenness of a process by the total mass of materials used to produce a given mass of product. It counts reactants, reagents, catalysts and solvents, in the reaction and in purification. Water used in the process counts too.

The ideal PMI is 1: every kilogram in becomes product. PMI is tied directly to Roger Sheldon's E-factor, the mass of waste per mass of product, through a simple relation: PMI = E-factor + 1. The industry generally prefers PMI because inputs are easier to measure accurately than waste.

Three details trip teams up:

1. Count the whole route, not the last step. The Roundtable defines the process as "all steps of a synthetic path from commonly available materials" through to the final bulk API. A clean final step can hide a wasteful intermediate.
2. Handle convergent routes properly. When fragments are made in separate branches and coupled later, each branch's mass must be carried into the final total. The Roundtable's Convergent PMI Calculator allows multiple branches for exactly this reason.
3. Be consistent about water and recycling. Whether recovered solvent is credited, and how process water is counted, changes the number. Fix the convention at the start of a project and keep it.

PMI has one important limit: it measures mass, not impact. A kilogram of water and a kilogram of a halogenated solvent count the same. Use PMI to drive efficiency, and pair it with solvent hazard scoring and carbon intensity, as described in our guide to early-stage LCA.

Where the mass actually goes

Ask a chemist where an API's mass goes and they will usually point to the reagents. The data says otherwise. Roundtable benchmarking of its member companies' processes found this composition:

Solvents and water make up 86% of the mass. The chemistry itself, the molecules that actually react, is less than a tenth. Sheldon cites a similar finding: solvents make up 80–90% of the non-aqueous mass used to make APIs.

This changes where to look. Most solvent mass is not in the reaction flask. It is in the work-up: extractions, washes, solvent swaps, chromatography and crystallisation. A step with a perfect stoichiometric reaction and five washes can have a far higher PMI than a messy reaction with a clean isolation.

So reducing PMI is mainly a solvent and work-up problem, and only secondarily a chemistry problem. That does not make the chemistry irrelevant: route choice decides how many isolations you need in the first place. But it means every process team should know exactly which operations carry the solvent mass in each step.

Benchmarks: what "good" looks like

PMI varies enormously by modality. When the Roundtable's greener peptide team assessed 40 synthetic peptide processes, it compared them with other drug types:

Two takeaways matter for strategy:

• Small molecules have a well-understood range. A small-molecule route well above roughly 300 kg/kg is a strong candidate for redesign, and every reduction translates directly into cost.
• Peptides are the next frontier. With more than 100 peptide drugs approved, the Roundtable concludes: "This assessment highlights the need for more environmentally friendly processes in peptide manufacturing." As peptide therapeutics scale up, their PMI becomes a material sustainability issue for the whole industry.

PMI also changes across development phases. Early routes are built for speed to the clinic, not efficiency, and PMI typically falls as a route is optimised toward commercial manufacture. Track it by phase, so a falling number is visible as progress and a flat one as a warning.

The most useful benchmark, though, is predictive. The Roundtable's PMI Prediction Calculator estimates probable PMI ranges for a proposed route, before any lab work begins. Using it at route selection, rather than after, is eco-design in practice.

Seven levers to reduce PMI

The levers below run roughly from quick wins within an existing route to changes that need a new one. The first ones are cheaper to pull; the last ones usually move PMI further.

1. Run reactions more concentrated

Reaction volume is often set by habit, or by the first lab run. Many reactions tolerate much higher concentration once heat transfer and mixing are understood. Halving the reaction volume halves the solvent mass for that step, and cuts the work-up that follows.

2. Rethink the work-up

Since most solvent mass sits in extractions, washes and swaps, this is usually the richest seam. Challenge every wash: which impurity does it remove, and is there evidence it is needed? Replace solvent swaps by distillation with a solvent that suits both steps. Use crystallisation instead of chromatography wherever the impurity profile allows.

3. Choose solvents that can be recovered and reused

A solvent that can be recovered in high purity and recycled cuts its effective contribution to PMI, depending on your recycling convention. A single solvent across several steps makes recovery far simpler than a mixture of different ones.

4. Telescope steps

Every isolation adds work-up solvent, drying energy and yield loss. Carrying an intermediate in solution into the next step removes that whole cycle. Pfizer's redesign of the sertraline process is the classic example: a three-step sequence was streamlined to a single step, ethanol replaced four solvents (methylene chloride, tetrahydrofuran, toluene and hexane), and overall yield doubled.

5. Use catalysis instead of stoichiometric reagents

Stoichiometric reagents, protecting groups and resolutions add mass that never reaches the product. A chiral resolution throws away at least half the material by design. Merck's letermovir process replaced a late-stage chiral resolution with an asymmetric, organocatalysed cyclisation. The old route had an overall yield of 10% and used nine solvents.

6. Move suitable steps to continuous flow

Flow chemistry allows higher concentrations, better heat transfer and the safe use of reactive conditions that would be impractical in batch. It can also enable in-line work-up and telescoping. It is not a universal answer, but for hazardous or highly exothermic steps it often unlocks both safety and PMI gains.

7. Redesign the route

The largest gains come from changing the route itself: fewer steps, convergent rather than linear assembly, and fewer protecting groups. It is also the most expensive lever, and the hardest to pull once regulatory filings are in. The letermovir redesign shows the scale of the prize:

Why PMI improvements stall

If the levers are well known, why do so many processes reach commercial manufacture with a PMI that could have been far lower? Three structural problems explain most of it.

Yield is optimised first, and PMI is checked last. Most development campaigns optimise one objective: yield, or yield and purity. PMI is calculated afterwards, as a report. By then, conditions have been chosen that maximise yield, often with excess reagent, dilute conditions and extra washes "to be safe". Those choices quietly inflate PMI.

One-variable-at-a-time optimisation cannot see the trade-offs. Concentration, solvent, temperature, equivalents and wash volumes interact. Changing them one at a time misses conditions that improve yield and PMI together, and makes testing those interactions too expensive. Classic Design of Experiments helps, but its cost grows quickly with each added variable and objective.

The route locks in before PMI matters. Early routes are built for speed to the clinic. Once a route has made material for clinical trials and appears in regulatory filings, changing it means comparability work and regulatory risk. The best opportunity to reduce PMI, route redesign, closes just as the business case for it becomes clear.

The result is a familiar pattern. Teams know the process is wasteful, can name the levers that would fix it, and still do not have the time or the experimental budget to pull them before the route freezes. The bottleneck is not chemistry knowledge. It is experimental throughput, and the habit of treating PMI as a result rather than a target.

How AI-driven optimisation cuts PMI without trading yield

Each of those bottlenecks is a data and throughput problem. That is exactly where AI and autonomous labs change the picture.

Make PMI an objective, not a report

Multi-objective Bayesian optimisation treats PMI, or its close cousin the E-factor, as a target alongside yield and purity. Instead of one "best" condition, it maps the Pareto front: the set of conditions where no objective can improve without another getting worse. At the University of Cambridge, Alexei Lapkin's group has used this approach to optimise flow reactions for space-time yield and E-factor together, revealing the trade-off explicitly rather than discovering it after the fact.

For a CMC team, that turns a vague debate ("can we cut the washes without losing purity?") into a chart. Leadership can see what each kilogram of solvent buys in yield or purity, and choose deliberately.

Explore many variables in few experiments

Bayesian optimisers pick each experiment based on everything learned so far. In one automated campaign published in Chemical Science, the optimiser reached its target in about 23 experiments, out of more than 12,000 possible combinations:

That efficiency is what makes it affordable to add work-up volumes, solvent choice and concentration to the search: the very variables that drive PMI.

Close the loop with autonomous execution

An autonomous lab runs, analyses and logs those experiments without waiting on a chemist for each cycle. Crucially, it records every input mass, so PMI is calculated for every run automatically, not reconstructed from notebooks at the end of a campaign.

Move the decision earlier

The biggest prize is timing. Models trained on chemical and process data can estimate PMI and impact for candidate routes before they are run, and suggest solvents and conditions likely to perform well. That brings route-level decisions, the lever with the largest effect, forward to a point where they are still cheap to make.

This is what DecaPlants is building: DecaOS, a process reasoning model working with self-driving labs, that embeds eco-design, including mass efficiency, into every step of process development. The aim is a process that is lean by design, rather than one made lean through rework.

A routine for measuring PMI through development

PMI only falls if someone is watching it while the choices are still open. This routine works whatever tools you use:

• Predict before you start. Estimate the PMI range for each candidate route with the Roundtable's PMI Prediction Calculator or an equivalent model, and use it in route selection.
• Set a PMI target per step at kickoff, next to the yield and purity targets.
• Log every input mass on every run, including work-up solvents, washes and water. Capture it in your electronic lab notebook (ELN), not a separate spreadsheet.
• Compute step and cumulative PMI automatically, using the convergent method for branched routes.
• Break PMI down by operation: reaction, extraction, washes, swap, isolation. The hotspot is usually obvious once it is visible.
• Report PMI at every stage gate, with the trend since the last gate.
• Pair PMI with impact metrics. Add solvent hazard and carbon intensity, so a mass reduction does not simply shift the burden elsewhere. Our guide to early-stage LCA shows how.
• Revisit the route before it freezes. Before material is made for pivotal trials, ask explicitly whether a route change would cut PMI enough to justify the regulatory work.

Frequently asked questions

Lean by design

PMI is one of the rare sustainability metrics that a CFO and a sustainability lead can agree on. Every kilogram removed is a kilogram not bought, not handled, not recovered and not incinerated. The letermovir and sertraline cases show that the biggest reductions come with better yields and lower costs, not at their expense.

What separates those cases from the norm is not better chemistry knowledge. It is the ability to explore enough options early enough, and to treat PMI as a target from the first experiment. That is precisely what multi-objective optimisation and autonomous labs make routine.

To see where PMI fits among the other eco-design metrics, read our guide to eco-design for industrial processes. To measure impact beyond mass, see early-stage LCA. And if you want to cut PMI on your next route, talk to a process engineer or explore DecaOS on our Tech page.

Sources

1. ACS GCI Pharmaceutical Roundtable, Process Mass Intensity Metric
2. ACS GCI Pharmaceutical Roundtable, Convergent PMI Calculator presentation and Convergent Process Mass Intensity Calculator
3. ACS GCI Pharmaceutical Roundtable, Process Mass Intensity Prediction Calculator
4. ACS GCI Pharmaceutical Roundtable, Peptides focus area
5. R. A. Sheldon, The E factor 25 years on: the rise of green chemistry and sustainability, Green Chemistry (2017)
6. US EPA, Green Chemistry Challenge: 2017 Greener Synthetic Pathways Award (Merck, letermovir)
7. US EPA, Presidential Green Chemistry Challenge: 2002 Greener Synthetic Pathways Award (Pfizer, sertraline)
8. A. A. Lapkin, Multi-Objective Bayesian Optimisation for Sustainable Reactions Development, AIChE Annual Meeting (2024)
9. O. Schilter et al., Combining Bayesian optimization and automation to simultaneously optimize reaction conditions and routes, Chemical Science (2024)