Get in touch
Email
jia.eq.liu@gmail.com
LinkedIn
@jia-eq-liu
Case study · Enterprise science

Benchling Bioanalysis

Bioanalytical labs often work across disconnected systems, turning work that should take minutes into days or even weeks. We partnered with Merck to build something that could fix that, using Benchling's existing Bioprocess platform as the starting point.

Could a platform built for a different scientific workflow handle this one? On-site research and design sprints showed where the existing model fit and where we needed something new: a status lifecycle for lab tests, and a way to design physical plate layouts.

Role
Design lead
Team
Four designers, five engineering teams
Timeline
2024–2025
Outcomes

10X

efficiency gain

1.1M

samples supported

3

new customer opportunities

Source: axendia.com

Clinical testing is slow by design, and slower in practice

Developing a clinical test is slow, labor-intensive and costly in materials. Scientists refine a test until it’s understood and fully documented, then run it across thousands of samples under strict re-test rules.

To manage that complexity, labs adopted or custom-built their own systems. Those systems kept the work compliant but slowed it down, stretching days of work into weeks.

  1. Experiments lived in documents Their data-capture tools were too hard for scientists to configure, so early experiments were written up in documents, not as structured data.
  2. Results were cut off from the design Tests ran on plates, on instruments with their own software, so linking a result back to the experiment that produced it was hard.
  3. Finalizing a test meant starting over There was no clean way to go from experimenting to a locked-down test ready for a clinical trial. A downstream team had to rebuild much of the work by hand.

One platform from first experiment to clinical run

Bioanalytical covers assay development and sample testing in one product, built on Bioprocess.

  1. 1Structured from the start Scientists map their workflow as steps and define each step’s materials, parameters and equipment, so experiments are captured as structured data instead of documents.
  2. 2Plates connected to results Plates are designed inside the experiment, and instrument data flows back to the wells and samples it came from.
  3. 3Run the study, see results as they land Samples and assay runs are tracked in one place, and results show up in a dashboard as testing runs.

1Structured from the start Scientists map their workflow as steps and define each step’s materials, parameters and equipment, so experiments are captured as structured data instead of documents.

2Plates connected to results Plates are designed inside the experiment, and instrument data flows back to the wells and samples it came from.

3Run the study, see results as they land Samples and assay runs are tracked in one place, and results show up in a dashboard as testing runs.

Mapping their science to our model

I acted as a translator between Merck and Benchling, turning dense scientific language into concepts our team could build on. The core problem was that our Bioprocess product had been built for users and workflows which looked similar at a high level, but were different in critical ways. We had to get the domain modeling right before we designing screens.

It also wasn't within scope to completely revisit this approach, as we'd made a commitment to launch with Merck, so I pushed to investigate the key risks alongside my PM and engineering counterpart. This led to a lot of hands-on work for two areas: plate design and test orders.

Designing plate-based experiments

Bioprocess plans experiments in a table where each column is one condition, run in one bioreactor. Bioanalytical runs on plates, where one condition can fill a block of wells, span several plates, or share a plate with others.

This meant our experiment planning UI couldn't support the complex spatial layouts required for plate-based experiments.

Invest in plate-based experiment design vs. shipping in increments

The question became: where would scientists design plates and how much should we invest in getting there? The table was wired into everything downstream, from experiment creation to run execution and analysis. Changing it meant a heavy engineering lift.

I led a design spike with engineers and PMs from the procedures and plates teams to weigh two options.

Planning · BioprocessPlanning · PlatesRebuiltWorksheet · BioprocessWorksheet · PlatesAnalysisTwo versions of planning and worksheets to build and maintain
Option A
Plates in experiment plan
Build plate design into experiment planning, so scientists lay out plates while they plan.
Pros
  • Scientists plan conditions and plates in one place
  • The plan matches the physical run from the start
Cons
  • Scientists wait longer for plates, with a less refined plate designer
  • Engineering forks the procedures code into two versions
PlanningWorksheetAnalysisLater, if neededSame planning table; plates designed in the worksheet
Option B ✓ Shipped
Plates in worksheets
Build plate design into worksheets, so scientists lay out plates when they run the experiment.
Pros
  • Scientists get plates sooner, with a more refined plate designer
  • Engineering reuses the planning table, with no fork
Cons
  • Scientists plan conditions and plates in two places
  • The plan and the plate layout meet a few steps apart

Reuse the planning table, invest in the plate designer

We chose option B. Plates were always going to need to live in worksheets, and that work was already in our estimates. Pulling plate design upstream into planning meant forking the procedures code for a single way of planning, and I didn’t think that was worth it compared with making the plate designer itself better, or other work our engineers could take on.

Choosing B also freed us to invest in other improvements to the plate design UI itself. We didn’t put a date on the second increment, but nothing in the first one closed it off: plate design could still move into planning later.

A layers panel listing well roles, contents with well counts, schema fields and parameters, with an Add layers action

Managing plate layers

Scientists could layer more onto a plate than we first planned. I designed a layer panel that stays consistent as layers are added.

Searching for plate contents, with options to search for an entity, add from a worklist, or pick material inputs

Scanning samples in

On site, I saw how much scientists relied on barcode scanning, so we explored bringing it into the plate designer itself.

A transfer table with wells, contents, initial concentration, units and transfer volume for each selected well

Filling plates

Plates hold many samples and quantities at once, so I pushed us to redesign filling: select wells, then set concentrations and volumes in one table.

One sample, many tests

Another place our data model broke was tests. Benchling already had samples and results, and customers used both heavily, but the test connecting them had never been modeled; it was only implied.

Scientists need more than that. Each sample needs tests assigned on arrival, tracked through results, and sometimes re-run until the results hold up. Some customers approximated this with a custom entity linked to the sample, though many didn’t bother, and someone then had to read that link and know what to do. The system never assigned, tracked or re-ran anything. The closest thing we had was our workflow engine, which already assigns and tracks tasks.

Extending the workflow engine versus building a test object was a heated debate. Long term, workflows could probably be redesigned to cover these cases, but they had already been redesigned once, and we’d seen duplicative objects blur boundaries before. For tests, the engine fell short in three ways:

  • It pointed the wrong way. A task knows its samples, but a sample doesn’t know its tasks. Scientists start from a sample, and in a task view they couldn’t see their samples or results.
  • Re-testing needed custom logic to read result values and create new tasks. Doable, but more work than a test object.
  • It was too heavy. We were already seeing performance problems with high-volume, small tasks, and a task per sample would multiply them.

I argued for the test object. It was a faster, cleaner path to an experience that has been standard in the industry for decades, and it avoided dependencies on the workflow engine that we couldn’t resolve before launch. It also kept the scope small: a test object with a status lifecycle was enough for scientists to find their samples and get results almost instantly, instead of waiting. Ideally, one workflow engine with the right abstraction would support these use cases, but we weren’t there.

A samples table listing each sample with its tests, status, dilution and result, including a failed test and its pending re-test at a higher dilution
What scientists expect

Samples with their tests, status and results. If a result is out of range, a new test is assigned at a higher dilution.

A workflows table listing tasks by name, schema, task group, status and assignee, with no samples or results
What workflows gave them

Tasks, with no samples or results. Where are my samples? How do I re-test them?

Making the data flow across the product

A test object touches every surface it flows through, so much of the work was keeping them consistent. I focused on three areas, doing the design work and connecting the teams involved so everything fit together.

Defining tests and dilutions

Benchling configures objects at the schema, template or instance level, so the first decision was where tests should live. I defined tests at the schema level so they can be reused and composed, and tied dilutions and re-test conditions to the assay procedure, since they vary by assay. These feed the analysis template that processes results.

The Tests and Dilutions tab of an assay procedure, choosing a test from a list, each with its result schema

Viewing tests with their samples

Scientists think in samples and the tests run on them, so I wanted a view of containers with their tests. Search was built around single objects, and our data model couldn’t support that relationship directly. We inverted it: a list of test orders, with the container as a column. Merck found this workable in testing, though it remains a compromise.

A list of test orders with their container, assay, status and dilution, with filters for assay template, test status, sample container and dilution factor

Ordering re-tests from results

Worksheets had no place for analysis results and no way to trigger actions outside themselves. We added results to the worksheet and a way to order re-tests from them. Out-of-range results suggest a new dilution, higher for high results and lower for low ones, and scientists review and create the re-tests in one step.

An Order re-tests dialog listing each sample and test with the previous and new dilution, and a container for each

Shipped, validated, and expanded

After months of work and weeks of user acceptance testing, we saw Merck go live with a public announcement and 400 users onboarded.

Time to end-to-end dataDaysMinutes
Legacy systems in use60
Scientists on the platform0400

Other teams at Merck started asking about Benchling, and with no sales effort Bioanalytical picked up two more prospects through word of mouth. One deal got far enough along to validate the approach, and was on track to close by the time I left.

Usage backed it up: scientists ran more experiments per user over time. Together with what we heard from them directly, it looked like real adoption. They’d made it part of their day-to-day.

Scaling bioanalysis across teams

Lessons learned

Make the invisible decisions visible

The hardest calls on this project were abstract relationships with implications nobody could see yet, like which way a link points or where a test gets defined. I think design’s power is making those implications visible, through diagrams, lo-fi mocks or working prototypes.

On complex software, that means helping the whole team see how the pieces fit, sometimes from a bird’s-eye view and sometimes at the level of a single interaction. The skill is switching between the two to find where things are most unclear.

Make bets you can build on

Product development is a series of informed bets, sometimes with one-way doors. With unlimited engineering time, a good team can design the perfect experience, but that's seldom the situation.

So as design lead, I try to ground our bets in real evidence from customers and the industry, then think through what each option commits us to: if we do A, what follows, and if B, what follows? Which one lets us come back and improve it?

Working in increments, starting with the first, guards against over-investing and pushes us to think long term when everyone is under pressure just to get it done. On this project that meant plates in worksheets first and a slightly hacky test-order list, each leaving a bigger redesign possible later.