Jewelry demand forecasting: a practical guide for retailers

By Elie Dufeu, CTO & Co-Founder, Metreecs. Published 23 October 2026.

Jewelry demand forecasting means estimating how many units of each piece and variant (metal, stone, size, length) will sell in each store and online, week by week, so that buying and production orders land before the gifting peaks they are meant to serve. It works when the forecast separates gift, self-purchase, and bridal demand, handles pieces that sell a few units a year, and plans far enough ahead to cover casting and stone lead times.

Claire buys for a jewelry retailer with 22 stores. Last year she placed her holiday order for gold chain necklaces in August, using the previous December plus 10%, split evenly across three lengths. By the second week of December the 45 cm chain had sold out in most stores while the 40 cm and 50 cm lengths sat in the cases, and the manufacturer could not recast before Christmas. (Claire is an illustrative example, but the pattern is familiar to any jewelry buyer.) The season total was roughly right, and the lost sales came from the split by length.

Most jewelry planners already know the holiday season decides the year. What is harder is turning that knowledge into a forecast at piece level that a buyer can order against months ahead. This guide to jewelry demand forecasting covers why demand in the category behaves the way it does, a six-step method, how to handle new collections, and how lead times and gold prices change the plan. If you want to test the method on your own range, you can see a piece-level jewelry forecast built on your sales history.

Key Takeaways

• The US Census Bureau found that nearly 19.8% of 2016 annual sales at US jewelry stores ($32.9 billion) were made in December alone, so a forecast error in one month carries a fifth of the year.
• The World Gold Council reports that global gold jewelry demand fell 18% in volume to 1,542 tonnes in 2025 while its value rose 18% to a record $172 billion, which means units and revenue forecasts can move in opposite directions.
• McKinsey found that early adopters of AI-enabled supply chain management improved inventory levels by 35% and service levels by 65% compared with slower competitors.
• Safety stock grows with the square root of lead time, so a piece on a 9-week casting and setting lead time needs 1.5 times the buffer of the same piece on a 4-week lead time.

What is jewelry demand forecasting?

Jewelry demand forecasting is the process of predicting unit sales for each jewelry piece and variant, by location and channel, over a planning horizon that covers supplier and workshop lead times. It combines sales history, the gifting calendar, launch plans, promotions, and price changes, and feeds purchase orders, production runs, and store allocation.

The unit of the forecast is the hard part. A ring in three metals and seven sizes is 21 forecasts, and many of those variants sell one or two units a year in a given store. A category total ("rings up 6% for the holidays") hides the variant that will run out, which is why AI demand planning software for retail works at piece, variant, and location level and then rolls up.

This article stays on the forecast itself. For the stock side (segmenting pieces by role, allocating across stores, and clearing aged inventory) see our guide to inventory optimization for jewelry retailers.

Why jewelry demand is hard to forecast

Five features of the category push a standard forecast off course, and each needs its own treatment.

Demand comes in a few short gift peaks. The Census Bureau figure above puts December at almost a fifth of annual sales, and Valentine's Day and Mother's Day add smaller spikes on top. A forecast built on monthly averages smooths those peaks away, and the order placed against it arrives too thin for December and too heavy for January.

Purchase motives have different curves. Gift buying peaks hard around holidays and leans toward mid-price pieces that are easy to choose. Self-purchase is steadier through the year and reacts more to new launches and social media. Bridal follows engagement and wedding seasons and depends heavily on sizes. Mixing all three into one history produces a curve that fits none of them.

Many pieces sell rarely, and each unit is expensive. Fine jewelry often has products that sell a handful of units a year per store, with a high unit cost. That is intermittent demand, and the usual forecasting methods treat the weeks of zero sales badly, either forecasting a steady trickle that never happens or overreacting to one sale. Methods built for intermittent demand, which forecast the size of a sale and the gap between sales separately, handle this pattern better.

Gold prices change what customers buy. The World Gold Council attributes the 2025 volume drop to affordability, after gold hit several all-time highs during the year. When the metal price climbs, customers trade down in weight, move to lower karats or to silver, or buy fewer pieces at the same budget, so a unit forecast built on last year's mix overstates heavy gold pieces.

Lead times are long and uneven. Casting, stone sourcing, setting, and finishing add up, and imported pieces add transit on top. Orders for the holiday season are usually committed months before the first gift shopper walks in.

The buyers we work with often underestimate how much of their forecast error comes from the split across variants rather than from the total. The category number for the season can be close while the sizes, lengths, and metals inside it are badly wrong.

How to build a jewelry demand forecast in six steps

This jewelry demand forecasting method works for a chain of stores or a direct-to-consumer brand. It assumes weekly sales history at piece and variant level, ideally two years or more.

1. Clean the history. Mark the weeks where a variant was out of stock and estimate what it would have sold, otherwise the forecast learns that a sold-out size has low demand. Remove one-off events such as a large corporate gifting order.
2. Split demand by motive where you can. Use product attributes (bridal collection, gift sets, price band) and channel data to separate bridal, gift, and self-purchase demand. Even a rough split by collection improves the seasonal curve.
3. Forecast slow pieces at family level, then split down. Group pieces that sell in a similar pattern (for example, all solitaire pendants in one price band) and forecast the family. Then allocate the family forecast to variants using their recent share of sales.
4. Apply a seasonal profile per family. Estimate the weekly shape of demand around each gifting event from pooled history, rather than from each piece's own thin history. Move the profile when the calendar shifts, for example when Thanksgiving falls late and the holiday window shortens.
5. Add known events and price changes. Launches, campaigns, store openings, and price increases tied to the metal price all go in as explicit adjustments with a written reason, so they can be reviewed after the season.
6. Turn the forecast into orders against the lead time. Order each variant to cover forecast demand over the lead time plus the review period, plus safety stock, minus stock on hand and on order. Review forecast error and bias every month by family.

The table below matches the forecasting approach to the type of jewelry product.

Forecasting new jewelry collections without sales history

New collections are the hardest case for jewelry demand forecasting: they have no history, so the forecast has to borrow it. Pick past launches that match on price band, product type, metal, and launch timing, and use their first weeks of sales as the curve for the new pieces. Group them into a low, base, and high curve rather than averaging them into one line. Our guide to forecasting new product introductions covers the analog method in more depth.

The commercial decision is how much of the buy to commit before any data arrives. A common approach is to commit the first production run to the base curve, hold an agreed slot with the workshop or supplier for a repeat, and compare weekly sell-through with the three curves for the first four to six weeks.

Mateo runs planning at a demi-fine jewelry brand that sells online and through 30 wholesale doors. For a birthstone collection launched in September, he forecast each of the 12 stones from analog launches instead of splitting the total evenly. The analogs said two months would carry far more of the gift demand ahead of the holidays, so he weighted the first run toward those stones and kept a repeat slot open. Three weeks in, two stones were tracking the high curve, and the repeat went to those two instead of being spread across all 12. (Mateo is an illustrative example.)

One mistake we repeatedly see is a launch forecast that stays frozen after the first week of sales. The first weeks of data are the most valuable information the planner will get on a new collection, and the forecast should be re-run every week until the curve settles.

Plan jewelry demand forecasting around lead times and gold prices

The horizon for jewelry demand forecasting has to be at least as long as the lead time plus the time between two orders. If casting, setting, and delivery take nine weeks and you order every four weeks, a variant needs a reliable forecast 13 weeks out. For holiday stock that often means committing in late summer, which is why the seasonal profile in step 4 matters so much.

Lead time also sets the size of the buffer. With the standard formula, safety stock equals a service factor times the standard deviation of weekly demand times the square root of the lead time in weeks. Take a pendant with a weekly demand standard deviation of 4 units and a 95% service target (factor 1.65):

• On a 4-week lead time: 1.65 × 4 × 2 = about 13 units of safety stock.
• On a 9-week lead time: 1.65 × 4 × 3 = about 20 units of safety stock.

(These figures are our own illustrative arithmetic.) The longer lead time adds seven units of buffer on a single piece, and for an expensive piece that is a real amount of capital. Shortening the lead time or forecasting more accurately over it are the two ways to reduce that buffer without giving up availability.

Gold prices belong in the forecast as a demand scenario as well as a cost line. The World Gold Council data shows what happens when the metal price rises fast: fewer tonnes sold, more money spent. In practice, a planner can keep the revenue plan and shift the unit mix toward lighter pieces, lower karats, and silver, then watch the sell-through by weight band each week.

Anjali plans gold jewelry for a retailer with 15 stores. When the metal price rose sharply ahead of the spring wedding season, she ran two scenarios for her bridal sets: last year's mix, and a mix shifted 20% toward lighter designs at the same price points. Weekly sales by weight band followed the second scenario within a month, so her repeat orders went to the lighter sets and the heavy sets were not reordered. (Anjali is an illustrative example.) For peak-season tactics beyond jewelry, see how to manage seasonal peaks without overstocking.

Where jewelry demand forecasting falls short

A forecast narrows uncertainty without removing it, and three situations keep a gap open in this category.

• Viral pieces. A design that takes off on social media can outrun any model trained on history. The defense is a fast weekly re-forecast and a supplier able to produce a quick repeat.
• One-of-a-kind and high-jewelry pieces. A unique piece does not have a forecast in any useful sense. Plan these as an assortment and capital decision, and forecast the demand for the price band rather than the piece.
• Sharp swings in the metal price. A forecast can include price scenarios, but it cannot predict the price itself. Agree in advance which scenario triggers a change in the buy.

Across Metreecs' work with jewelry retailers, the forecasts that hold up are the ones reviewed by variant family every month, with bias tracked separately from error. A forecast that runs consistently high on one family is a bias problem that compounds quietly over a season, and it is easier to fix than random error once it is visible.

FAQ

How do jewelry retailers forecast demand for the holiday season?

Start from at least two years of weekly sales at piece and variant level, with stockout weeks corrected. Build a seasonal profile for each product family from pooled history, then apply it to the base forecast and add planned campaigns. Order early enough to cover casting and delivery lead times, and keep a repeat slot for the pieces that sell fastest in November.

How do you forecast jewelry pieces that only sell a few times a year?

Forecast them as a family of similar pieces rather than one by one, then split the family forecast to each variant using its recent share of sales. Use methods built for intermittent demand, and set safety stock with care, because each extra unit of an expensive piece ties up capital.

How do gold prices affect jewelry demand forecasting?

Higher gold prices tend to reduce the number of pieces sold while total spending can hold up or rise, as the World Gold Council reported for 2025. Forecast units and value separately, run a scenario with a lighter mix, and track sell-through by weight band so you see the shift early.

How far ahead should a jewelry retailer forecast?

At least as far as the lead time plus the review period. For pieces with a nine-week production lead time and monthly orders, that is about 13 weeks of reliable weekly forecast, and longer for the holiday buy, which is often committed in late summer.

Can a small jewelry brand forecast demand without a data team?

Yes, if the data is clean at piece and variant level. Jewelry demand forecasting at a small brand can start with family-level forecasts, a seasonal profile per family, and a monthly review of error and bias. Software helps once the number of variants makes the spreadsheet too slow to update every week.

Conclusion

Jewelry demand forecasting works when it follows how the category sells: a forecast per piece and variant, seasonal profiles built for gift peaks, family-level methods for slow pieces, and a horizon that covers the real lead time. A good first check is the split inside your last holiday season, because the variants that sold out early usually show where the forecast needs work. To see how this looks on your own range, book a demo with your jewelry sales data.

Sources

• US Census Bureau, "Measuring the Holiday Shopping Season" (November 2017): share of 2016 US jewelry store sales made in December.
• World Gold Council, Gold Demand Trends Full Year 2025, Jewellery: 2025 jewelry demand in tonnes and value, and the effect of record gold prices on affordability.
• McKinsey, "Succeeding in the AI supply-chain revolution": inventory and service level gains for early adopters of AI-enabled supply chain management.