Food Formulation Process: From Product Brief to Stable Prototype (2026)

Introduction

Most failed prototypes do not fail because the recipe was weak. They fail because nobody wrote down what “good” meant before the first batch, so every tasting moved the target. A gated method fixes that: each step ends with a pass condition that is set before the trial and decided by a test, not by opinion.

This guide sets out the six steps Food Research Lab formulators use to take a brief to a stable prototype, the gate that closes each step, and the decisions that sit behind them. For the wider journey from idea to launch, see our food product development process guide; this article covers the formulation stage in depth.

What Food Formulation Really Involves

Formulation vs recipe development

A recipe lists ingredients and steps. A formulation is an engineered formula: ingredient weights to a fixed basis, process yields, and targets for pH, water activity, texture, cost and claims, written so another team can reproduce it at scale. The difference shows up the first time a batch is made by someone else, on other equipment, with a new lot of raw material.

Where formulation sits in product development

In food product development, formulation sits between the approved concept and the pilot batch. Stage-gate models describe what happens in each stage; the method below adds the targets, tests, documents and go/no-go decisions that sit between a brief and a prototype that holds its quality over its shelf life.

Step 1: Product Brief, Target Product Profile and Specification

Translating a brief into measurable targets

Turn every adjective in the brief into a number and a test. “Soft” becomes a hardness range on a texture analyser; “long life” becomes a shelf-life target with a storage condition and an end point. Hold the result in a specification sheet as a target product profile (TPP), give each field a tolerance (for example, pH 4.2 ± 0.1), and start the bill of materials (BOM).

Composition, pH, water activity, cost and claims

pH and water activity (a_w) decide the safety category and much of the stability strategy, so set them first. US rules use pH 4.6 and a_w 0.85 as boundaries: low-acid foods have a finished equilibrium pH above 4.6 and a_w above 0.85, acidified foods are low-acid foods with added acid that reach pH 4.6 or below at a_w above 0.85, and acid foods have a natural pH of 4.6 or below.^[3] Moving a product across one of these lines changes its process controls, so decide the category at the brief, not after tasting. Claims work the same way: a protein or sugar claim sets a composition threshold that the formula must meet with margin, after process losses.

Table 1. Target product profile (TPP) template

Gate 1: a signed specification in which every target has a number, a tolerance and a test.

Step 2: Ingredient System and Functionality

Choose ingredients by function first and supplier second, because each function decides texture, stability or cost. Hydrocolloids, for example, are used to thicken, gel and stabilise emulsions and dispersions, while most plant-protein gels are sensitive to pH, temperature and ionic strength, which is why a protein swap often needs a hydrocolloid or process change alongside it.^[1]

Table 2. Ingredient functionality matrix (FRL view unless cited)

Model cost of goods (COGS) from the start: cost-in-use per kg of finished product = Σ(inclusion fraction × price per kg) ÷ yield fraction. Dividing by yield matters, because a formula that loses 10% of its mass in baking costs about 11% more per kg sold than its raw-material sum suggests. Use supplier quotes and file each supplier specification. For a worked view of mapping functions to ingredients, see how brands map ingredient functionality in NPD.

Gate 2: functions mapped, first bill of materials costed and supplier specifications on file.

Step 3: Process Parameters, Benchtop Trials and DOE

Designing trials with DOE and controls

Change one factor at a time only when you are confident the factors do not interact. Protein and hydrocolloid levels, or sugar and acid, usually do, so use a design of experiments (DOE). A full factorial tests every combination: k factors at n levels need n^k runs, so cost rises quickly with each added factor.^[2] A two-level design estimates main effects and interactions (y = b₀ + Σb_ix_i + Σb_ijx_ix_j), and centre points show whether the response is curved. Plackett–Burman designs screen many factors in few runs; central composite and Box–Behnken designs fit curved responses for optimisation, and a central composite design needs F + 2v + n_c runs (factorial points, two axial points per factor, and centre points).^[2] Randomise the run order, replicate, and block by session so day-to-day drift does not look like a treatment effect.^[2]

Table 3. Illustrative 2² factorial layout with centre points (design only, no results)

Run the six trials in a randomised order. If the centre points sit well away from the average of the four corners, the response is curved and a central composite design is the next step.

Documenting versions, weights and yields

Give every batch a version number, weigh every ingredient, record the process yield and keep an unchanged control in each session. Measure pH and water activity before anyone tastes, so a version that has drifted out of its safety category is rejected on data, not after a panel likes it.

Gate 3: a version that meets the water activity, pH and texture targets in replicated runs, with the control.

Step 4: Sensory and Analytical Evaluation

Texture profile analysis (TPA) compresses a sample twice, imitating two bites, and reports hardness, cohesiveness, springiness, gumminess, chewiness, adhesiveness and resilience.^[4] In the standard definitions, hardness is the peak force of the first compression, cohesiveness is the ratio of the second to the first compression area, and chewiness is hardness × cohesiveness × springiness. Report the probe, compression level and speed with every result, because TPA values are only comparable under the same settings.

The 9-point hedonic scale is the most widely used scale for consumer acceptance testing; in one study, the number of scale categories did not change acceptance scores for attentive assessors, but attentiveness to the task did affect how the scales were used.^[5] Panel size follows training: about five to ten trained panellists, 25 semi-trained, and at least 100 untrained consumers.^[2] For method choice, see our note on food sensory evaluation.

Set the pass criteria before the panel sits. For the protein bar example below, the example targets are a_w ≤ 0.65, day-30 hardness ≤ 2,200 g and a mean hedonic score of at least 7.0. Every trial goes in the iteration log.

Table 4. Iteration log, protein bar example (illustrative values, not test results)

Note how v1 meets the hardness target on day 0 but fails it by day 30: a release-day texture test alone would have passed it.

Gate 4: results meet the specification, or the log names the next change.

Step 5: Stability and Shelf-Life Prediction

Prototype development ends when the product holds its safety, sensory and physical properties across its intended life, not when it tastes right on day 0. Hardening of high-protein bars is a typical failure, and it is a formulation problem as much as a moisture problem. In an accelerated study at 32 °C, bars made with whey protein isolate started at about 3.4 N and reached almost 100 N by day 37, while bars made with partially hydrolysed whey protein stayed at 10–15 N; the authors linked hardening to phase separation between protein and carbohydrate during storage.^[6] The decision this supports: if a bar hardens, review the protein and sugar or polyol system before simply adding moisture, which raises water activity.

Plan shelf-life work with end points and a margin. UK food-safety guidance advises storing test product under reasonable worst-case conditions, testing beyond the intended shelf life to show a margin of error, and testing chilled products at around 8 °C rather than assuming a consumer’s fridge stays below 5 °C. It also notes that accelerated shelf-life testing, which raises temperature and humidity to speed deterioration, may suit long-life ambient products better than standard durability studies.^[7] In FRL practice, accelerated runs screen versions and real-time runs confirm the claim. Methods, timelines and cost drivers are in our guide to real-time vs accelerated shelf-life testing.

Gate 5: end points, storage conditions and margin agreed before the shelf-life claim is set.

Step 6: Regulatory Pre-Screen and Scale-Up Readiness

Pre-screen the formula against the target market’s rules before it is frozen, because a late allergen, additive or approval problem forces reformulation.

Table 5. Regulatory pre-screen checkpoints

Then prove the formula on a pilot batch. Scale-up is not arithmetic: mixing, shear and heat transfer change with batch size. In one study of grape products for older adults, pilot-scale hardness rose significantly in one texture grade but fell in the other two, and pH and sugar acidity differed between lab and pilot scale at every grade.^[10] Because the direction of change is not predictable, re-run texture, pH and water activity on the pilot batch against the same specification.

Gate 6: pre-screen cleared, and a pilot batch reproduces the targets.

Inside the Lab Example: A Cold-Pressed Juice from Brief to Stability Plan

FRL Inside the Lab: choosing a thermal treatment for a cold-pressed juice

Brief: develop a cold-pressed juice prototype and compare thermal treatments for shelf life without losing fresh taste.

Approach: a tropical, citrus and berry blend was standardised for °Brix, titratable acidity and pH, adjusted to 3.5 ± 0.1. Three treatments were compared: LTLT (63 °C, 30 min), HTST (72 °C, 15 s) and UHT (135–140 °C, 2–5 s), with storage at 4 °C and 25 °C and a 15-person hedonic panel on a 9-point scale.

• HTST was preferred: it balanced microbial safety with retention of fresh taste.
• Freshly prepared juice scored 8.3 for appearance, 8.0 for flavour, 7.5 for mouthfeel and 8.0 overall.
• The processed juice was stable for 21–28 days refrigerated with no microbial growth reported; ambient storage was limited to 3–5 days.

Source: FRL Inside the Lab, Cold-Pressed Juice (15 October 2025). The page does not give fruit ratios, microbial counts, the ambient temperature or which treatment each shelf-life figure refers to.

The example maps onto Steps 1, 3, 4 and 5: a fixed pH target at the brief, a structured comparison of three process conditions, a hedonic panel against a 9-point scale, and storage at chilled and ambient conditions. It was a treatment comparison, not a replicated DOE. The 3–5 day ambient limit is the kind of result that should feed back into the brief: it tells the brand that this product needs a cold chain.

Common Pitfalls and the Brief-to-Prototype Checklist

Most failures in the food formulation process come from skipped controls, not weak recipes. Common pitfalls (FRL view) and their fixes:

• Tasting before targets exist. Without a target product profile, “good” shifts with each tasting. Fix: set pH, a_w, hardness and hedonic targets first.
• Changing several factors at once. You cannot tell which change moved the result. Fix: use a DOE with randomised runs, replicates and centre points.
• Dropping the control. Raw-material or panel drift then looks like improvement. Fix: re-run the control every session.
• Testing stability late. Hardening and moisture migration only show over time. Fix: start the shelf-life plan with the first passing prototype.
• Treating scale-up as arithmetic. Mixing, shear and heat transfer change at pilot scale. Fix: repeat texture, pH and a_w on a pilot batch.
• Checking regulation late. A late allergen or approval problem forces reformulation. Fix: run the pre-screen before the formula is fixed.

Brief-to-prototype checklist

1. Brief and specification. Gate: every target has a number, a tolerance and a test.
2. Ingredient system. Gate: functions mapped, bill of materials costed and supplier specifications filed.
3. Benchtop trials and DOE. Gate: a_w, pH and texture met in replicates, with the control.
4. Sensory and analytical. Gate: results meet the specification, or the log names the next change.
5. Stability plan. Gate: conditions, end points and margin agreed before the claim is set.
6. Regulatory and scale-up. Gate: pre-screen cleared and a pilot batch reproduces the targets.

Formulator’s note: write the gate before the trial

A gate set after the data tends to fit the data. Write down the metric, the test method, the threshold and tolerance, the number of replicates, and what you will do on a pass or a fail. For example: “a_w ≤ 0.65 on three replicate batches; if not met, change one factor and re-run with the control.” A gate written this way decides the trial, so a near-miss cannot be argued into a pass.

Conclusion

A prototype is ready to move on when it meets a written specification in replicated runs, holds that specification over a stability study with a margin, clears the target market’s rules and reproduces on a pilot batch. Each of the six gates exists to catch one failure early: a moving target, an untested ingredient interaction, a release-day pass that fails at day 30, or a texture shift at scale. Writing the gates before the trials is the cheapest way to avoid rework later.

For Brands & Formulators: Turning Your Brief into a Stable Prototype

If you have a product brief and a target market, FRL’s formulation team can run this gated method with you, from specification to pilot batch.

• Specification: we turn the brief into a target product profile with tolerances and test methods.
• Trial plan: we design benchtop trials and DOE around the factors that matter for your product, with controls and replicates.
• Stability and readiness: we set the shelf-life plan and margin, pre-screen for India, US or EU rules and confirm the targets on a pilot batch.

Start with FRL’s new food product development service, add a shelf-life study for the finished prototype, or contact the team to scope the work.

1. Gao, Y., Liu, R., & Liang, H. (2024). Food hydrocolloids: Structure, properties, and applications. Foods, 13(7), Article 1077. https://doi.org/10.3390/foods13071077 (Cited for hydrocolloid functions and the sensitivity of plant-protein gels to pH, temperature and ionic strength.)
2. Prasad, G., Paul, M. M., Gopinath, P. P., Krishnasree, V., & Somanath, G. (2025). Designs for sensory experiments and product optimisation: A comprehensive review. Journal of Sustainable Technology in Agriculture, 1(2). https://doi.org/10.65287/josta.202510.B068 (Cited for factorial run counts, Plackett–Burman, central composite and Box–Behnken designs, randomisation, replication and blocking, and panel sizes by training level.)
3. Definitions, 21 C.F.R. § 114.3 (2026). https://www.ecfr.gov/current/title-21/chapter-I/subchapter-B/part-114 (Cited for the US pH 4.6 and water activity 0.85 category boundaries.)
4. Mabrouki, S., Brugiapaglia, A., Glorio Patrucco, S., Tassone, S., & Barbera, S. (2023). Texture profile analysis of homogenized meat and plant-based patties. International Journal of Food Properties, 26(2), 2757–2771. https://doi.org/10.1080/10942912.2023.2255758 (Cited for the set of texture profile analysis parameters.)
5. Addo-Preko, E., Amissah, J. G. N., & Adjei, M. Y. B. (2023). The relevance of the number of categories in the hedonic scale to the Ghanaian consumer in acceptance testing. Frontiers in Food Science and Technology, 3, Article 1071216. https://doi.org/10.3389/frfst.2023.1071216 (Cited for the 9-point hedonic scale as the most widely used acceptance scale and the effect of scale length and attentiveness.)
6. McMahon, D. J., Adams, S. L., & McManus, W. R. (2009). Hardening of high-protein nutrition bars and sugar/polyol–protein phase separation. Journal of Food Science, 74(6), 312–321. https://doi.org/10.1111/j.1750-3841.2009.01225.x (Cited for protein-bar hardening during storage at 32 °C and its link to protein–carbohydrate phase separation.)
7. Food Standards Scotland. (2025). Shelf-life guidance: Guidance for food businesses for the setting of product shelf-life (Version 1.2). https://www.foodstandards.gov.scot/downloads/Shelf-life_guidance_2025.pdf (Cited for worst-case storage, testing beyond the intended shelf life, 8 °C chilled testing and accelerated shelf-life testing.)
8. Food Safety and Standards Authority of India. (2022). Food Safety and Standards (Approval for Non-Specified Food and Food Ingredients) Regulations, 2017 (Compendium, version of 17 October 2022). https://www.fssai.gov.in/upload/uploadfiles/files/Compendium_FSS_NFS_FA_17_10_2022.pdf (Cited for the prior-approval requirement in regulation 3(1) and the definition of non-specified food.)
9. U.S. Food and Drug Administration. (2025). Questions and answers regarding food allergens, including the food allergen labeling requirements of the Federal Food, Drug, and Cosmetic Act (Edition 5): Guidance for industry. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/guidance-industry-questions-and-answers-regarding-food-allergen-labeling-edition-5 (Cited for US major food allergen labelling, including sesame.)
10. Kim, Y. J., Ryu, J. H., Lee, J. Y., Kim, Y. S., Kim, D., & Kwon, Y. (2024). What is the difference between laboratory-scale and pilot-scale grape-based products for older adults with chewing difficulties? Foods, 13(23), Article 3844. https://doi.org/10.3390/foods13233844 (Cited for differences in hardness, pH and sugar acidity between laboratory-scale and pilot-scale production.)
11. European Parliament and Council of the European Union. (2011). Regulation (EU) No 1169/2011 on the provision of food information to consumers. Official Journal of the European Union, L 304, 18–63. https://eur-lex.europa.eu/eli/reg/2011/1169/oj/eng (Cited for the Annex II allergen list and the Article 21 emphasis requirement.)