The demand planning problem for a small spirits producer is not a scaled-down version of the enterprise problem. It is a structurally different problem, and most of the tools on the market are not built for it. At Asterley Bros, we land squarely in that gap: a team of six, pronounced seasonal demand driven by the aperitivo and gifting calendar, botanical ingredients with lead times of three to five months from specialist suppliers, and SKU counts you can list on one hand. When we looked at what the mainstream demand planning market was selling, the answer was clear before we had even opened a pricing page: we were going to build.
The mismatch between craft producers and commercial forecasting tools
Most demand forecasting products on the market are designed around assumptions that do not hold for small batch spirits producers. They assume frequent, high-volume reorder cycles. They are optimised for businesses running dozens or hundreds of SKUs across national distribution networks. Their statistical models are calibrated on data volumes that a six-person London vermouth operation will not accumulate for years, if ever.
Vintaflow's 2025 guide to demand forecasting for craft breweries documents exactly this tension: the forecasting approaches suited to craft producers tend to be lighter, more qualitative, and more dependent on operational knowledge than the models commercial software wants to run. The tools that claim to serve the sector often do so by stripping features out of enterprise products rather than rethinking the problem from the ground up.
For us, the result of that mismatch is not just inconvenience. With real stakes and thin margins at every production run, a forecast that is wrong in the wrong direction means either tying up working capital in excess stock or turning away orders we cannot fill. The cost of a poor fit is not abstract.
What demand planning actually looks like for Asterley
Our demand picture has three main drivers: the on-trade (bars and restaurants in London and nationally), direct-to-consumer gifting, and wholesale accounts that spike around key aperitivo moments in the calendar, particularly late spring through summer and again at Christmas. None of these are flat or predictable in the way a fast-moving consumer goods model would expect.
On top of that, our key botanicals carry lead times that mean a production decision in July has implications for what we can make in October. We source from a small number of trusted suppliers, and we are not always their largest customer. Availability is not guaranteed. A demand model that cannot hold that operational reality in the same frame as the sales forecast is not a demand model for our business; it is a sales spreadsheet with a fancier interface.
I want to be careful here not to overstate the complexity: our SKU range is deliberately contained, and that is a structural advantage when it comes to forecasting. But the interplay between seasonal demand, long procurement lead times, production batch sizes, and a finite team capacity is structurally hard to capture in a generic tool.
The build-versus-buy comparison at our scale
The table below sets out how the key considerations compare for a craft spirits producer at our scale. This is not a universal verdict on build versus buy; it is a map of the specific trade-offs that shaped our decision.
| Consideration | Off-the-shelf tool | Custom build |
|---|---|---|
| Fit to business model | Generic; requires adapting your process to the tool's assumptions | Built around your specific SKUs, lead times, and seasonal drivers from day one |
| Upfront cost | Low to moderate (SaaS monthly fee) | Time investment; no seat licence |
| Ongoing cost | Recurring licence, often priced for larger operators | Maintenance time; no external dependency |
| Adaptability | Dependent on vendor roadmap and support responsiveness | Amend, adapt, and evolve same day |
| Data ownership | Data typically held in vendor infrastructure | Full ownership and portability |
| Integration with other internal tools | API access varies; often requires workarounds | Designed to connect directly to your own systems |
| Key-person risk | Low (vendor maintains the product) | Higher; requires internal documentation discipline |
Six months of deep AI integration: what the build has become
The demand planning build does not sit in isolation. Over the past six months, we have been building AI integrations across the whole business from our South London workshop: operational meeting summaries, client reports, recipe development tracking, supply chain monitoring, sourcing decisions, pricing models, and production scheduling. Demand planning sits inside that wider system, pulling from the same data layer and feeding into the same operational cadence.
InfinitySky's 2026 report on AI automation in the drinks industry notes that adoption among craft producers is accelerating, but that the majority of small operators are still reaching for off-the-shelf automation rather than building custom workflows. We are running a different experiment: treating every layer of the business as something we can instrument, model, and improve on our own terms.
The advantage of that approach compounds. When the demand forecast connects directly to the procurement schedule, which connects to the production plan, which connects to the cash flow model, the whole system gets more useful with every data point. A generic tool gives you a forecast; a connected internal system gives you an operating picture.
What building actually requires from a small team
The honest version of this section is that building requires a specific kind of resource: time and technical appetite, not necessarily a large engineering team. We are six people doing two jobs each, which is itself a constraint that shaped the architecture of what we built. Everything had to be maintainable by people who are also making vermouth, running sales calls, and managing logistics.
Tools like the AI workflow automation options covered by Elementum AI's 2026 overview of AI workflow tools have brought the floor down considerably. The infrastructure required to connect internal data, run models, and surface outputs in usable formats no longer demands a full engineering function. What it still demands is clarity about what you are trying to model, and enough data discipline to feed the system consistently.
We documented our logic, built outputs into tools the whole team uses, and made sure the forecast is not a black box that only one person can open. That distributes the understanding and reduces the key-person fragility that is the legitimate risk of the build path.
Where we are now and what comes next
The current system covers seasonal demand projection, botanical procurement scheduling, and batch-size optimisation across our core range. It is not finished, because the right answer to that framing is that it will never be finished: it grows with us, changes when our range changes, and reflects decisions we made rather than defaults some product team set two years ago.
We have written more about the broader shape of this work over at Absolution Labs, including some of the specific tooling choices we made and how they fit together. The short version is that for a business at our scale, with our level of technical curiosity, building was the only option that made sense in 2026. Not because buying is always wrong, but because what was available to buy was not built for us.
The work in the margins is cumulative. Each connection we build between parts of the business creates a little more operating clarity, a little more lead time on problems before they become expensive. For a small spirits company with real stakes and thin margins at every production run, that clarity is not a luxury.
Frequently asked questions
Why do off-the-shelf demand planning tools struggle for small craft spirits producers?
Most commercial demand forecasting products are designed for businesses running high-SKU, high-frequency retail or distribution operations. For a craft spirits producer with a handful of SKUs, pronounced seasonal demand, and botanical lead times measured in months, the underlying models assume data volumes and reorder cadences that simply do not apply. The tool ends up fighting the business rather than serving it.
What does 'building' demand planning actually mean for a six-person drinks company?
In our case it means writing custom logic in Python, connecting it to our own operational data, and running forecasts in a workflow we control end-to-end. No seat licences, no waiting on a vendor's roadmap, no support tickets to change a field label. When the model needs adjusting for a new product or a supplier delay, we adjust it that afternoon.
Is building your own forecasting tools realistic without a dedicated data team?
It depends heavily on what you are actually trying to model. For a small spirits producer with a contained SKU range and a relatively stable set of input variables, a bespoke forecast does not need to be a machine-learning research project. The bigger requirement is disciplined data collection over time, which is something any organised small team can manage with the right internal tooling.
Which commercial demand forecasting tools did you evaluate before deciding to build?
We looked at several options aimed at the food and beverage sector, including tools positioned specifically at craft producers. The common pattern was pricing designed for operators far larger than us, with feature sets that reflected those larger businesses' needs. Vintaflow's guide to demand forecasting for craft breweries gives a useful overview of where the mainstream options sit and what assumptions they carry.
What is the biggest operational risk of building rather than buying?
The honest answer is key-person dependency. If the tool lives primarily in one person's head or one person's codebase, staff changes become a real fragility. We have tried to mitigate this by documenting the logic thoroughly and building outputs into shared workflows that the whole team touches regularly, so understanding is distributed rather than siloed.