R&D data gets lost because it's recorded for one reader. A formulator writes up a trial for themselves, at that moment, to answer the question they had that week. Reusing it means a different scientist asking a different question years later, and almost nothing in a typical lab's records was built for that reader.
A technical lead described the result on a recent call: "I call it the Easter egg hunt every time I have to go and search through our system." The comparison is more exact than it sounds. Easter eggs are hidden by someone who knows exactly where they are. The hunt is only hard for everyone else.
Every record is written for its author
Look at a typical trial spreadsheet. The rows are the samples one person made. The columns are the tests that mattered for one project. A header reads "Visc 25C," and only its author knows which viscometer, spindle and speed produced the number. Yellow cells mean "retest." Sample codes like AB-114-3 point to a notebook page on a shelf in another building.
Every one of those choices is efficient for the person who made it. Together they turn the file into a puzzle for anyone else. The missing context was never written down because the author never needed it written down.
The context fades for the author too. In a 2016 Nature survey of 1,576 researchers, more than 70% said they had failed to reproduce another scientist's experiments, and more than half said they had failed to reproduce their own. The survey covered academic research rather than industrial formulation, but the mechanism carries over: a result outlives the memory of how it was produced.
Finding the file is the easy part
Suppose a customer asks a coatings team whether they can switch to a lower-VOC coalescent. A senior formulator remembers trying exactly that in 2021. Answering the customer means reassembling five pieces of evidence:
- The formulation, in a spreadsheet that now exists as v7 and v8, with no note on which one was tested
- The raw material lot, recorded in the ERP under a supplier code
- The application and cure conditions, written in a paper notebook
- The adhesion and scrub results, exported from instrument software on a lab PC
- The conclusion, in a slide deck emailed to a manager who has since left
Each piece uses its own identifier, and none points to the others. The formulator reconnects them by hand, and one broken link makes the whole result untrustworthy. Here the lot record shows the supplier changed its manufacturing process in 2022, so nobody can say whether the 2021 result still holds. The team reruns the trial. It takes two weeks and answers a question the company had already paid to answer once.
That's the distinction most data projects miss. A result is reusable only when someone can find it, understand it and trust that it still applies. Search tools help with the first and do nothing for the other two.
Failed experiments disappear first
Successful work gets promoted. It moves into product specifications, tech transfer packages and final reports, and each move adds context. Failed work stays where it was made, in a spreadsheet in a folder called "old trials" or "don't use."
For reuse, that's backwards. A recorded failure is what stops the next scientist repeating it. In experimental design, negative results mark the edges of the space worth exploring. A predictive model trained on a lab's history learns only from what was recorded, so a history made mostly of successes teaches it a design space that looks friendlier than the real one.
Why the usual fixes don't last
Most labs have tried at least one of these:
- Folder rules and naming conventions. They ask the author to spend time now for a stranger's benefit later, so compliance fades within a few busy quarters. A folder tree also holds one hierarchy, while scientists' questions cut across it by material, by property or by customer.
- Enterprise search. It indexes the words inside files. The answer to "which formulations with more than 2% of this dispersant passed adhesion?" lives in the relationships between files, which an index of text can't see.
- Moving everything to one shared location. The files move, and their missing context stays missing.
- An ELN used as digital paper. The narrative becomes searchable while the results sit in attached spreadsheets, back where they started.
Each fix asks someone to add structure after the work is done, and that step is the first to go when deadlines tighten.
What findable actually requires
Academic data science already has a working answer: the FAIR Guiding Principles, published in Scientific Data in 2016. They say data should be findable, accessible, interoperable and reusable, by machines as well as people. In an industrial formulation lab, that comes down to four things:
- Every formulation, sample and material lot has a persistent identifier that other records point to.
- Results are stored as values with units, test method and conditions attached, instead of as cells in someone's layout.
- Relationships are captured as the work happens. The result knows its sample, the sample knows its formulation, and the formulation knows its lots.
- Failed trials are recorded with the same structure as successful ones.
The third point decides whether the other three survive. The reusable record has to be a byproduct of doing the work. If recording a trial properly takes longer than the old spreadsheet did, scientists will go back to the spreadsheet, and they'll be right to.
Run the reconstruction test
Pick one decision your team made in the last two years, such as a raw material approval or a reformulation. Ask someone who wasn't involved to rebuild the evidence behind it: the formulations, lots, conditions and results, including the trials that failed. Note how many systems they open, how many people they have to ask and where the trail breaks.
The place where it breaks is where to start, and it's a far better first project than migrating everything.
What this means for R&D leaders
- A lab that reruns trials it can't trust has less real capacity than its experiment count suggests. Count the repeats as well as the hours spent searching.
- The records your team keeps this year are what next year's hires inherit. Ask what they could rebuild from them.
- When you evaluate software, bring your messiest real spreadsheet and the reconstruction test to the demo. One buyer put the standard simply: "It has to work on our data."
- Judge any AI or predictive tool by the history it learns from, failures included.
Context captured where the work happens
Uncountable's Research & Development suite records each experiment with its ingredients, process parameters and results, and each result with its test conditions, so scientists can pick up past work from the ELN instead of a folder tree. Instrument data becomes structured results on the experiment that produced it, so results stay findable after the project ends. Bodie lets scientists ask for past work in plain language, across the experiments, data and documents in Uncountable that they can access. Rogers Corporation reported 40% less time searching for past experiments and compiling reports with Uncountable.
.png)
.png)
.png)
