Find the process worth taking forward.

Feed strategy is only one potential difference. Material source, grade, lot attributes, preparation, storage and age can influence the response. So can vessel geometry, scale, equipment configuration and analytical method changes.

Use linked run records to construct the comparison population. Keep exclusions and missing values explicit. Scale correlations and models retain their assumptions, calibration data and applicability; a prediction at a receiving scale is not observed performance at that scale.

Preserve the actual execution. Keep the candidate version, material lots, prepared solutions, equipment, set points, actual signals, additions, interventions, observations, samples, pools, yields, holds and exceptions together. Do not edit the protocol retrospectively to make an exploratory run look as though it followed the original plan.

A sample links the result to a known run, vessel, unit operation and time. Its preparation, custody, method, acquisition and review explain what the measurement represents. A result copied into a summary table without those relationships loses information needed for comparison.

Give Neil the process target, supplied procedures and permitted run records. It can compare the conditions, identify missing context and prepare a proposed experiment with linked samples, measurements and review work. Ask it to author the template and verification cases as part of the package.

In this example, the proposal crosses the two observed feed levels with the two observed temperatures at 2 L, keeps the material basis and analytical method consistent and captures titre and viability alongside the actual conditions. Scientists review the experimental design and execute the work. The example does not claim that Neil ran an experiment or established a control strategy.

Everything in the middle column is preparation. Nothing Neil drafts runs until a scientist has reviewed it, and the people who own the study still make its decisions. In line with the EU’s draft GMP Annex 22 on AI, Neil is not used in GMP execution. It helps set up configuration, which your team verifies and releases under change control. Customer data is not used to train AI models, and the requester’s permissions govern what Neil can read. See how Neil fits Annex 22 and more about Neil.

Define what the process must achieve. Product quality, yield, throughput, facility, scale, containment, material constraints, cycle time and development stage establish the target. Connect candidate processes to those requirements so that an attractive local result does not obscure a failure elsewhere in the process.

Model the unit operations, parameters and expected outputs as reusable, versioned records. Upstream development compares feed, culture conditions and actual run profiles without losing vessel, material or sample context. Downstream development follows recovery and quality across unit operations, pools and fractions, to locate a loss before choosing what to change. Analytical development keeps method versions, standards and sample preparation attached to the result used in a process decision. Process characterisation connects the study question, parameter ranges, responses and analysis to the proposed process understanding.

Each team keeps its own specialist assessment while sharing the material, sample and method relationships. A method change or a missing sample then shows up in the comparison rather than after the decision.

Enforcement is configurable. During early development, scientists capture data in a structured framework without being constrained by it: exploratory changes are allowed, and the actual conditions and reasons remain visible. As the process matures, controls tighten. Exploratory and controlled studies can use different permissions and review paths.

Your organisation defines the intended use, record controls, review and verification required for each stage, and qualifies the configured workflow. Compliance depends on that assessment and your procedures, not on the software alone.

Failed experiments remain part of the process evidence. Their factors, actual execution, results and conclusions can support risk assessments, operating boundaries and investigations, and help avoid repeating dead ends. ICH Q8 makes the same point: changes during development are opportunities to gain knowledge, and knowledge from experiments giving unexpected results can also be useful.¹

A knowledge claim can concern a parameter effect, hold, material attribute or scale model. Retain the supporting runs, conflicting observations, analysis, uncertainty, scope and reviewer conclusion. New evidence can challenge the claim without erasing the basis for earlier decisions.

Process characterisation draws on that evidence to assess relationships and a proposed control strategy.¹ Merely collecting development data does not mean characterisation is complete. Characterisation then quantifies how parameters and materials affect performance and quality attributes, working from the same candidate, runs, parameters and samples in Seal.

Candidate selection, scale-up, characterisation and transfer each need a defined readiness decision. Record open risks, conditions, owners and further work. A design space can be developed at any scale, but its relevance to production scale has to be justified and the risks of scale-up discussed.² An unresolved scale question is part of that case, not a footnote to it.

Technology transfer exists to carry product and process knowledge between development and manufacturing, and that knowledge forms the basis for the manufacturing process, control strategy and process validation approach.³ The receiving team starts with the selected process, parameters, materials, equipment needs and evidence links. It still authors and approves a recipe suitable for its facility and validation state. Shared records reduce reconstruction; they do not make a receiving process qualified by inheritance.

Carry the selected process into tech transfer and MSAT manufacturing support, with the original development evidence still attached.

Instrument software remains responsible for acquisition and control. Seal IoT collects completed exports from an approved folder, keeps the original file linked to every derived result and maps selected results into the run, so a value can always be traced to the acquisition behind it. For a chromatography run, a volume axis is not a time axis and a reduced export is not the original acquisition; those choices are recorded rather than hidden.

Legacy development data can be imported and linked to process definitions. New development work is structured from the start, and legacy data provides historical context. See all integrations.

Choose a decision in front of the team: which candidate to take forward, whether a result holds at scale or what the next study should separate. Bring the runs, samples and results behind it, and the instrument exports where you have them.

Together, define the comparison and the output you need, inspect whether each finding is supported by its source records and review the proposed next study with the scientists who would run it. Expand to the next discipline once that first decision is useful.