Sensory panel screening before you commit: a Sri Lankan R&D guide
By the Silk Foods Ceylon Editorial Team
Buyer’s snapshot
- ISO 4120:2021 puts a triangle test at 24 to 30 assessors when the question is whether two formulations are perceptibly different, and roughly 60 when the question is whether they are close enough to swap.
- ISO 11136:2014 sets a hard floor on the consumer side. A consumer sample conforming to the standard is never fewer than 60, and 100 where the laboratory has no basis for a power calculation.
- A panel answers a narrow question well. It will not tell you whether the product sells, what it can be priced at, or whether it holds for nine months on a Sri Lankan shelf.
- Screening two or three formulations properly adds two to four weeks before a recipe locks. The timeline table below shows where those weeks go and which of them can run in parallel.
Most Sri Lankan brand owners arrive at a contract manufacturer with one recipe they believe in and two they are willing to consider. The research conversation that follows usually turns on a question nobody asks out loud: how does anyone actually know which of the three is better?
The office tasting most brands run does not know. Six colleagues around a table, samples served in their own containers, everyone aware of which one the founder made, is not a screening method. It is a consensus ritual with a founder at the head of it. A sensory panel exists to replace that with a number and a stated risk level.
What a sensory panel is doing when you have two or three formulations
Sensory testing splits into three different questions with three different methods. A discrimination test asks whether two products are perceptibly different. A descriptive test asks what the differences are, in named attributes. A hedonic test asks which one consumers like more. ISO publishes a separate standard for each, and the three are not interchangeable.
Getting the order wrong is the common and expensive mistake. A brand that runs a 100-person liking test between two formulations nobody can reliably tell apart has bought a very precise measurement of noise. The cheap test comes first.
| The question being asked | Method | Who tastes | Typical numbers |
| Are these two formulations perceptibly different at all? | Triangle test, ISO 4120:2021 | Screened assessors, familiar with the mechanics of the test | 24 to 30 for a difference test, roughly 60 for a similarity test |
| What exactly is different, and by how much? | Descriptive profiling | Assessors selected and trained under ISO 8586:2023 | A small trained panel, run repeatedly over sessions |
| Which one do buyers actually prefer? | Hedonic test, ISO 11136:2014 | Untrained consumers drawn from the target population | Never fewer than 60, and 100 by default |
The sequencing that saves money runs top to bottom. If a triangle test says two variants are not perceptibly different, the decision is already made on other grounds: whichever formulation is cheaper, more stable, or easier to source wins by default, and the liking test never needs to be commissioned. That single result is often the most valuable thing a screening round produces.
How many tasters do you actually need?
ISO 4120:2021 states that when testing for a difference, typical numbers of assessors are between 24 and 30, and that testing for no meaningful difference needs roughly twice as many, about 60, for equivalent sensitivity. With 24 assessors, 13 correct identifications are required before you can conclude that a perceptible difference exists at a 5 per cent risk level.
The mechanic behind that number is worth understanding, because it is what makes the test honest. In a triangle test the assessor gets three coded samples. Two are identical, one is different. The task is to pick the odd one out, and it is a forced choice: the standard does not allow an assessor to report no difference. Chance alone therefore produces one correct answer in three. On a 24-person panel, pure guessing delivers about 8 correct answers before anyone has tasted anything meaningful. Thirteen is the threshold that clears that noise.
Run the same arithmetic on the office tasting. At six assessors, ISO 4120 requires 5 correct answers out of 6 to reach a 5 per cent risk level, and its table gives no attainable value at all at a 0,1 per cent level. A six-person panel can only produce a usable result if it is nearly unanimous, and even then the confidence interval around the finding is wide enough to drive a truck through.
There is a legitimate workaround for a small operation, and the standard spells it out. Where enough assessors are not available, replicate evaluations can be used to reach the required number of total evaluations: 10 assessors evaluating 3 triads each produces 30 evaluations. That is valid for a difference test. It is explicitly not valid for a similarity test, where the standard says treating those 30 evaluations as independent does not hold. If the question is whether a cheaper ingredient can be swapped in without anyone noticing, the shortcut is unavailable and the panel has to be real.
The consumer side has firmer floors. ISO 11136:2014 states that a consumer sample conforming to the standard shall never be fewer than 60, and fixes the minimum at 100 where the laboratory does not have the information needed for a power calculation. Segmenting raises it further: where a segmentation is decided before the test, the minimum is 60 consumers per segment. A brand that wants to read Colombo separately from the regional market is committing to at least 120 respondents before the first questionnaire is printed.
What the panel can tell you, and what it cannot
A discrimination test returns one piece of information: whether a difference is perceptible at a stated risk level. A hedonic test returns a liking score and a preference split within a defined group. Neither predicts repeat purchase, price tolerance, or shelf-life stability. Those are separate studies, with separate methods and separate costs.
What a properly run panel does tell you is narrow and genuinely useful. Whether the change you made is detectable at all, which matters most when an ingredient is being substituted for cost or supply reasons. The direction and size of a preference within a target group. Which attributes carry the difference, when a trained descriptive panel is involved.
What it does not tell you is the part brands routinely over-read. A liking score is measured in a booth with no price on the pack, no competing product on the shelf, and no second purchase three weeks later.
It says nothing about whether the shelf life holds, which is a stability question answered by accelerated and real-time stability work, and nothing about regulatory acceptance, which runs on composition, safety and labelling evidence through the SLSI submission rather than on taste data.
One further limit sits in the standard itself and catches brands that test the same product repeatedly. ISO 11136 warns against professionalising a consumer pool and recommends an interval of at least three months between individual tests on the same product. A brand that runs the same 80 people through four rounds of the same recipe in six weeks is measuring familiarity, not liking.
How many weeks does a screening round add to a launch?
Two to four weeks is the realistic answer for in-house discrimination screening between two or three bench variants. A commissioned consumer test at ISO 11136 numbers adds one to two weeks on recruitment alone, and that time is spent before anyone tastes anything.
| Step | What actually happens | Elapsed |
| Brief and formulation targets agreed | Decision rule written down, variants defined, the question the panel will answer fixed in advance | Week 0 |
| Bench samples produced | Two or three variants made at bench scale under the same process conditions, so the only difference is the intended one | Weeks 1 to 2 |
| Screening panel run | Assessors recruited and screened, samples coded and rotated, session run, results analysed against the table | Weeks 2 to 3 |
| Result read and acted on | One variant selected, or a fourth commissioned if the answer is that nothing is distinguishable | Week 3 |
| Pilot batch at line scale | The selected recipe run on the production line, then screened again against the bench sample it came from | Weeks 4 to 5 |
| Optional consumer hedonic read | Runs in parallel from week 3 if the budget carries it, at 60 to 100 respondents | Weeks 3 to 5 |
The last row matters more than it looks. The consumer read can run alongside pilot production rather than after it, which is how a brand gets an independent liking number without adding two weeks to the critical path. It only works if the pilot recipe is locked before the consumer test starts, so the thing being measured is the thing that will be made.
The research team at the Matale facility sees the same pattern on first briefs from local brand owners. A founder arrives with a home-kitchen recipe and a firm opinion about it, and the first bench batch at pilot scale does not taste like the kitchen version. The recipe has not changed. The heat load, the hold time and the fill temperature have. The useful panel at that moment is not the one that ranks three flavours. It is the triangle test that asks whether the pilot batch is perceptibly different from the kitchen sample the founder signed off, because that question can be settled in a week and the argument it replaces routinely runs four to six.
The same dynamic shows up in detail in the work of reformulating a kitchen recipe for retort at 1,500 jars and in the retort iteration cycle on a vegan cheese spread.
Setting the room and the samples so the answer means something
ISO 8589:2007 covers the physical conditions. A test room is a testing area separated from a preparation area, with controlled temperature and, where it affects the evaluation, controlled humidity. On lighting the standard is specific: a correlated colour temperature of 6 500 K gives a neutral light similar to northern daylight, and 5 000 K to 5 500 K with a high colour rendering index approximates noon daylight. Booths exist to stop assessors reading each other.
Most of what goes wrong in a Sri Lankan screening session is not the room, though. It is the sample handling.
Codes come first. Three-digit random codes on every vessel, never A, B and C, because a letter carries an implied ranking and assessors respond to it. Presentation order comes next. ISO 4120 lists the six possible sequences of two products and instructs that they be distributed at random in groups of six assessors, so that serving position does not quietly become the variable under test. Then the serving conditions: same vessel, same volume, same temperature, same interval out of chilled storage, for every sample in the set.
Assessor selection is the part small operations skip. ISO 8586:2023 sets out recruiting, pre-screening, training and validating sensory assessors, including screening for colour vision, odour recognition, texture discrimination and descriptive ability. A panel drawn from whoever was free at 3pm has an unknown proportion of people who cannot detect the attribute in question at all, and that proportion is buried inside the result.
Buyer's checklist: before you book a screening session
|
Where does a Sri Lankan brand actually run the panel?
Three venues cover almost every local project, and they are not substitutes for each other. The choice usually comes down to which question is live and whether an independent result is needed for anyone outside the company.
| Venue | Good for | Cannot do | Practical note |
| In-house at the manufacturer | Fast discrimination screening during the iteration cycle, between bench variants and against a pilot batch | An independent consumer liking read at ISO 11136 numbers | Samples do not travel or age between production and tasting, which removes a real source of error |
| An accredited third-party laboratory | Independent sensory analysis, and the shelf-life and thermal process work that usually runs alongside it | The scale of a full consumer study, and same-week turnaround during rapid iteration | The Industrial Technology Institute offers sensory analysis via a trained taste panel maintained on site, alongside real-time and accelerated shelf-life testing |
| A commissioned consumer test | Liking and preference reads at the numbers ISO 11136 requires, segmented where the brief calls for it | Attribute-level diagnosis of why one variant wins | Recruitment is the cost and the calendar. Budget one to two weeks before the first session |
For a first product-development project the sequence that costs least is usually in-house triangle testing through the iteration rounds, then one independent read on the final two variants. That keeps the expensive test for the decision that actually needs it. Where the manufacturer has research capability on the same property, the round trip from Colombo to Matale is a day, and samples can be tasted in person rather than shipped and re-tasted three days later.
The budgeting side of this is worked through in more detail in the research and development iteration budget for a Sri Lankan launch, and the flavour-selection case appears in the coconut jam flavour variant work and the kombucha flavour pivots.
When to skip the panel entirely
There are four situations where a sensory panel is the wrong line item, and a brand owner is better served knowing them before commissioning one.
The first is having only one formulation. A panel compares. With nothing to compare against, the money buys a liking score with no reference point, and it is better spent on stability work that tells you whether the product survives to the end of its claimed shelf life.
The second is a change the brand has already decided to accept. If an ingredient is being substituted for supply reasons and the substitution is going ahead regardless of a perceptible difference, the discrimination test has no decision attached to it. Go straight to stability and labelling.
The third is an undefined decision rule. A panel cannot supply the rule that reads its own result. Deciding after the fact what counts as a win is how a result gets argued into whatever the founder wanted at the start, which is the exact failure mode the panel was bought to prevent.
The fourth is scale. At a first-run size of 1,500 jars for spreads, 1,250 bottles for beverages or 180 bottles for capsules, a 100-respondent consumer test is a material line item against the batch itself. Silk Foods Ceylon (SFC) structures those first-run minimums for a brand that has already validated its concept and is ready to commit to a commercial run. For a founder still deciding what the product is, the honest advice is two more rounds at bench scale before anyone books a panel.
Frequently asked questions
How many people do you need for a triangle test?
ISO 4120:2021 gives 24 to 30 assessors as typical for a difference test. At 24 assessors, 13 correct identifications are needed to conclude that a perceptible difference exists at a 5 per cent risk level. Testing for similarity, meaning close enough to swap, needs roughly 60 for equivalent sensitivity.
Is a six-person office tasting enough to choose a formulation?
Not as evidence. ISO 4120:2021 requires 5 correct answers out of 6 assessors to reach a 5 per cent risk level, and its table offers no attainable value at a 0,1 per cent level. A six-person result needs near-unanimity to say anything, and still carries a very wide margin of error.
Does a sensory panel result count towards SLSI clearance?
No. Sensory results are internal product-development evidence. Sri Lanka Standards Institution (SLSI) clearance and Sri Lanka Food Act labelling compliance are separate submissions built on composition, safety and labelling data. Plan the panel inside the research calendar, not the regulatory one, because the two do not substitute for each other.
Can Silk Foods Ceylon run sensory screening as part of a product development project?
Yes. Research and new product development, listed as Co-Development on the Silk Foods Ceylon brochure, covers bench iterations and blind screening between variants before a recipe locks. The Matale facility is BRCGS- and FSSC 22000 V6-audited, and samples can be tasted in person on the property.
How Silk Foods Ceylon can help
For local FMCG brands choosing between two or three formulations, Silk Foods Ceylon (SFC) runs research and new product development from a cellular-manufacturing facility in Matale, with an in-house team that works through two to four sample iterations per variant before a recipe locks. Blind screening between variants, and between a pilot batch and the bench sample it came from, sits inside the engagement rather than after artwork has been committed, which is where most product-development projects lose four to six weeks. First-run minimums are 1,500 jars for spreads, 1,250 bottles for beverages and 180 bottles for capsules, so the screening work is sized against a batch a brand can actually commit to. The facility is BRCGS- and FSSC 22000 V6-audited, with SLSI clearance and Sri Lanka Food Act labelling support built into a standard engagement.
To brief a formulation project, email b2b@esilkroute.com.lk or call +94 76 441 0389 / +94 76 918 5744.
Sources
International Organization for Standardization, ISO 4120:2021, Sensory analysis, Methodology, Triangle test, third edition March 2021, clause 6.2 on typical assessor numbers of 24 to 30 for a difference test and approximately 60 for similarity, clause 7 on the six presentation sequences and the forced-choice rule, and Table A.1 on the minimum correct responses required at each risk level. iso.org (retrieved 29 August 2026).
International Organization for Standardization, ISO 11136:2014 with Amendment 1:2020, Sensory analysis, Methodology, General guidance for conducting hedonic tests with consumers in a controlled area, clause 6.2 on the consumer sample never being fewer than 60, the fixed minimum of 100 without a power calculation and the 60 per segment rule, and clause 5.2.4 on the recommended three-month interval between tests on the same product. Confirmed current on review in 2025. iso.org (retrieved 29 August 2026).
International Organization for Standardization, ISO 8586:2023, Sensory analysis, Selection and training of sensory assessors, on recruiting, pre-screening, training and validating assessors, including screening for colour vision, odour recognition, texture discrimination and descriptive ability. iso.org (retrieved 29 August 2026).
International Organization for Standardization, ISO 8589:2007, Sensory analysis, General guidance for the design of test rooms, on the separation of testing and preparation areas, controlled temperature and humidity, and lighting at a correlated colour temperature of 6 500 K for neutral daylight or 5 000 K to 5 500 K at a high colour rendering index. iso.org (retrieved 29 August 2026).
Industrial Technology Institute, Food Technology Section, Sri Lanka, listing sensory analysis via a trained taste panel maintained at the institute, shelf-life testing of packaged food by real-time and accelerated methods, thermal process validation of canned and bottled foods, and new product development services. iti.lk (retrieved 29 August 2026).
Sri Lanka Export Development Board, export performance for 2025, reporting food and beverage exports of USD 583.89 million, up 24.23 per cent on the prior year, on a 40.4 per cent increase in processed food exports. srilankabusiness.com (retrieved 29 August 2026).