AI systems can now model fermentation curves, forecast batch yields, and flag production anomalies in real time. According to the Stanford HAI AI Index 2026, practical AI adoption in physical manufacturing is accelerating faster than most predictions suggested, particularly in quality control and process monitoring. The one thing these systems still cannot do is stand in a cold workshop at 7am, taste a blend, and decide it needs another week.
That call belongs to the maker.
At Asterley Bros we have built and used AI tools in our South London production operation for a few years now. The demand planner we built in-house is the most consistently used tool across everything we have assembled. It ingests data from every platform we sell through and forecasts based on seasonal phasing and trends. It genuinely helps. But it does not touch the production decisions where judgement lives.
Why clarity and filtration is the hardest call in vermouth production
Every batch of Asterley Bros vermouth goes through a filtration stage. There is always a trade-off between clarity and flavour. Heavier filtration removes more particles and gives you a cleaner-looking product. It also strips flavour. Lighter filtration preserves the aromatic complexity but leaves the liquid with more visual variation batch to batch.
The right call depends on the specific batch, the botanicals used, how the maceration went, what the base wine brought to the blend, and a dozen other variables that interact differently every time. There is no single setting that produces the right outcome. The decision is made by tasting, looking, and applying experience from every previous batch that did or did not work.
No AI tool in production use today makes that call. Vinetur reported in June 2026 that even the most advanced AI production systems in the alcohol industry are focused on monitoring and optimisation, not on the final sensory judgement steps. The makers are still in the loop for those.
Flavour comes first. Clarity comes second. You build your processes based on the outcomes you want.
That is not a design choice born from distrust of technology. It is a practical observation about where the technology currently sits versus where the decisions actually live. The AI Index notes that AI struggles with the tacit knowledge experienced makers hold, and that finding is consistent with what we see in production. Tacit knowledge, accumulated by proximity to a specific product and specific equipment over years, is not easily encodable in a training dataset.
At a team of six people, every person in the Asterley Bros operation is close enough to the production process that this knowledge is shared by being in the room together. When James decides a batch needs more time, it is because he has made that blend before, felt what it tastes like when it is not ready, and recognises the same signal now. That recognition is not in a spreadsheet. It is in the person.
Where AI earns its place in small-batch production
This is not an argument against AI in production. The opposite. The demand planner we built is useful precisely because demand forecasting is a data problem. Sales patterns across Shopify, wholesale accounts, and our subscription service create a structured signal that a well-built model can read better than a human doing mental arithmetic across quarterly spreadsheets. Seasonal phasing, reorder triggers, procurement timing: all of this benefits from having a system that watches the data steadily rather than relying on the production team to check manually.
The same logic applies to fermentation monitoring, anomaly detection, and quality control pattern recognition. These are areas where AI adds value because they have the right shape: data-dense, pattern-driven, and not dependent on the kind of sensory experience that takes years to build.
| Production stage | AI contribution | Human lead? |
|---|---|---|
| Demand forecasting and procurement | Pattern recognition across multi-channel sales data, seasonal adjustment | Oversight and exception handling |
| Fermentation monitoring | Continuous data capture, anomaly alerts | Diagnosis and corrective action |
| Botanical selection and maceration timing | Reference data from past batches | Yes, primary |
| Blending decisions | Historical ratios and flavour modelling (limited) | Yes, primary |
| Filtration and clarity judgement | None in current production use | Yes, primary |
| Final release decision | Quality control data to support | Yes, primary |
The operational stuff nobody writes LinkedIn posts about
The division above is not clean in practice. Demand forecasting touches production planning which touches raw material availability which affects what we can make and when. The systems are connected. AI tools that handle the data layer well create space for the production team to focus on the sensory work rather than chasing administrative tasks.
That is the real value proposition for a small producer using AI. Not transformation. Not disruption. Time: specifically, the time of experienced people redirected from work that a system can do toward work that only they can do. The build-logs for how we approached this at Absolution Labs are worth reading if you are at a similar stage.
We built the demand planner because Rob and James spent too much time doing rough mental arithmetic across platform dashboards when a system could do it more reliably. The time that freed up went back into production attention. That is a genuinely useful trade. It does not change what craft is. It changes how much of it you can actually do in a week.
What the small-batch scale actually means for AI adoption
Large producers have the data volumes needed to train production-specific AI models. A major distillery running thousands of batches through consistent equipment generates the kind of repeatable signal that supports machine-learning approaches to sensory prediction. Asterley Bros does not operate at that scale, and neither do most of the small producers Absolution Labs works with.
At our scale, the AI tools that work are the ones built for the operational layer, not the production-decision layer. That is not a temporary gap waiting to be closed. It is a structural feature of small-batch production that the technology available to SMEs in 2026 does not change.
The question to ask, if you are a small producer looking at AI tools, is not "can AI help with my production?" but "which specific parts of my operation generate enough structured data that a model would be useful?" For most small-batch operations, that question points clearly to the operational and logistics layer, and away from the sensory and craft layer. Build there. The filtration call is still yours.
FAQ
Can AI decide when a spirit or vermouth batch is ready?
No current AI system can reliably make the final readiness call on a small-batch spirit. AI can model fermentation curves, flag anomalies, and predict likely outcomes from past data. But the decision about flavour balance, mouthfeel, and whether a blend is finished requires direct sensory assessment from an experienced maker. That tacit knowledge has not been encoded in any system the production drinks industry uses.
What parts of craft spirits production can AI help with?
AI adds the most value in the parts of production that generate data and follow patterns: inventory and demand forecasting, fermentation monitoring, botanical procurement timing, quality-control anomaly detection. These are real efficiency gains for a small team. The sensory and judgement-intensive stages, including blending, filtration decisions, and final release calls, still require human expertise.
How does Asterley Bros use AI in its production process?
The most consistently used tool in the Asterley Bros operation is an in-house demand planner that ingests sales data from every platform the business sells through and forecasts based on seasonal phasing and trends. It handles the data-heavy operational work so the team can focus on the sensory and formulation work that requires direct experience. Absolution Labs built the tool specifically for a small-batch drinks operation.
What is tacit knowledge in spirits production?
Tacit knowledge in spirits production is the accumulated sensory and contextual understanding that experienced makers develop over years of working with the same botanicals, equipment, and processes. It includes recognising the right clarity in a filtered spirit, knowing from smell when maceration is complete, and judging from mouthfeel whether a blend needs another week. This knowledge is difficult to codify because it is learned through experience, not instruction.
Will AI eventually replace human craft judgement in spirits production?
Not for small-batch craft producers in the foreseeable future. The data volumes needed to train production-specific models are only practical at large-scale operations. At the scale Asterley Bros operates, six people making spirits in a South London workshop, the batch-to-batch variation and the sensory complexity mean that human judgement remains the more reliable tool for final craft decisions. AI optimises the surrounding operational work. The two are complementary, not competitive.