report on inventory analytics: lost revenue due to out...
TRANSCRIPT
Report on Inventory Analytics:Reducing Lost Revenue Due To Out Of Stock
Inventory
Igor OfenbakhIgor Ofenbakh
Founder, WizData Systems, Inc.
The ChallengeThe Challenge
Not enough inventory
Too much inventoryy
Product/size mix not optimalnot optimal
Lost RevenueMaterials copyright WizData Systems, Inc. Proprietary Technology
Problem DefinitionProblem Definition
A chain of stores sell various products.Ideally Real world
There always are the right products in the right quantities at every store.
Products can be out of stock (OOS). After that moment a seller has Lost
Lost Sales are the potential sales after an item goes OOSh ld b d if h b h i i k
the right quantities at every store. After that moment a seller has Lost Sales.
that could be made if there were be enough units in stock.
For the seller, the following business questions arise:‐What is the amount of sales that were lost?‐ Is there a better distribution of products
between stores and items?between stores and items?
Materials copyright WizData Systems, Inc. Proprietary Technology
The BenefitsThe Benefits
Lost Sales estimation;Lost Sales estimation;
Size curves tied to local demand;
Proper distribution between stores;
I i l (4% 22%)Increase in sales (4% ‐ 22%);
Universal model.
Materials copyright WizData Systems, Inc. Proprietary Technology
ChallengesChallengesMaterials copyright WizData Systems, Inc. Proprietary Technology
Example 1.
Si l Li D d8Sale qt. One store, one item
Simple Linear Demand
7
66
7
8(25 units in stock)
Out of Stock
5
4
34
5Sold units
Lost sales3
2
12
3Lost sales
1
0
1
1 2 3 4 5 6 7Week
1 2 3 4 5 6 7
Materials copyright WizData Systems, Inc. Proprietary Technology
Example 2.
R l W ld D dReal World Demand
878
9Sale qt. One store, one item
(25 units in stock)Sold units
Lost Sales
67
567 Out of Stock
2 2
1234
? ?01
1 2 3 4 5 6 7Week
? ?
Materials copyright WizData Systems, Inc. Proprietary Technology
Example 3.Large quantities. Size proportionsSale qt.
680700
800Sale qt.
Large blouseSmall blouseLost Sales
400
502
396 398400
500
600 Lost Sales
303
206205
332261
200 196152200
300
400
15296
0
100
1 2 3 4 5 6 7 Week
Materials copyright WizData Systems, Inc. Proprietary Technology
Example 4.
R l ld Si tiReal world. Size proportionsReal sales in an individual store display large variances that do not
ll f di t ti l l ti hi
1214Sale qt. Large blouse
Small blouse
allow for direct proportional relationship.
11
9 9
12
98
10
12Small blouseLost Sales
76
5
8
46
8
32
4
2
4
? ?0
1 2 3 4 5 6 7Week
Materials copyright WizData Systems, Inc. Proprietary Technology
Aggregating by StoresAggregating by StoresEvery store has its own size distribution.
Store 1
L 36%
Store 2
4%
29%
48% 19%M
S14%
35%
36%15%
XS
Store 3 Store 4
19%
44%11%
13%
36%10%
26% 41%
Materials copyright WizData Systems, Inc. Proprietary Technology
Aggregating by StyleAggregating by Style
V NECK TOP PRINTED DRESS BLOUSE
Every item has its own size distribution.
44% 20%
V NECK TOP
L 38% 19%
PRINTED DRESS BLOUSE
9%
27%
44% 20%M
S13%
29%
19%
XS
LACE BACK CAMI BEADED TANK TOP
14%
37%14%
8%42%
33%
36% 16%
Materials copyright WizData Systems, Inc. Proprietary Technology
Sales by WeekDue to seasonality, sales change significantly from week to week.
436391
450500
Sale qt. PRINTED DRESS BLOUSE, All stores
373 391
299300350400
200163
107100150200250
050100
1 2 3 4 5 6 7Week
1 2 3 4 5 6 7
Materials copyright WizData Systems, Inc. Proprietary Technology
Challengesof Lost Sales Analysis
Small quantity of weekly sales per store, size andstyle. Hence week to week deviation of the realysales is significant ;
Demand is time dependent;
Different demand for various locations and styles;
Different size proportions for different stores andp pstyles.
Materials copyright WizData Systems, Inc. Proprietary Technology
SolutionSolutionMaterials copyright WizData Systems, Inc. Proprietary Technology
Problem FormulationProblem Formulation
There is a network of stores selling a product (e.g. Tshirts) of various styles and sizes in different locations.For every valid combination of Style, Location, Sizeand Week we have sales data in the following forms: g
X units sold or X units sold and OOS position was reached
After OOS there is no sales or demand information available.The problem is to find demand as a function of style,i d l ti f k f th id d i dsize and location for every week of the considered period.Once this function has been found, it provides an estimation of the Lost Sales for those weeks in which the store was OOS.
Materials copyright WizData Systems, Inc. Proprietary Technology
Sales Data vs.True Demand
The number of units sold in a store cannot be considered the True Demand especially in the case where the number of weekly sales is small.
bl l d h llIt is impossible to accurately predict that 1 unit will be sold during week 1, 4 units during week 2, 1 unit during week 3, etc.during week 3, etc. True Demand is a function (possibly fractional) of Style, Size, Location and Week
with a high degree of variance!
Materials copyright WizData Systems, Inc. Proprietary Technology
Example 5.S l D t T D dSales Data vs. True Demand
44
4.5
Sale qt. One store, one itemSold units
22.53
3.5
1 1
2
1
0 51
1.52
True Demand = 1.5 units per week
00
0.5
1 2 3 4 5 6Week
Materials copyright WizData Systems, Inc. Proprietary Technology
AssumptionsAssumptions
Materials copyright WizData Systems, Inc. Proprietary Technology
Local Size CurveSize distribution of a style is assumed to be a function
of style design and location.
Store #1 Store #2 Store #3L
M
S
XS
V NECK TOP
PRINTED DRESS BLOUSE
LACE BACK CAMI
Materials copyright WizData Systems, Inc. Proprietary Technology
Size CurveDoes Not ChangeDoes Not Change
Size curve of a style at a given location does not change in time.
Store #1 Store #2 Store #3L
M
S
XS
V NECK TOP
PRINTED DRESS BLOUSE
LACE BACK CAMI
Materials copyright WizData Systems, Inc. Proprietary Technology
Weekly SalesFor every style the weekly fractions of a total demand duringa considered period are the same for different locations.
22%25%
Pct of total sales
PRINTED DRESS BLOUSE, all stores
19% 20%
15%15%
20%
10%8%
6%10%
0%
5%
1 2 3 4 5 6 7Week
1 2 3 4 5 6 7
Materials copyright WizData Systems, Inc. Proprietary Technology
ResultsResults
As a result of the Lost Sales algorithm wegcalculated a True Demand function, whichallows us to:Find the Lost Sales in context of styles, locations and sizes;Analyze the structure of sales during the period;Find the optimal product distribution between locationslocations.
Materials copyright WizData Systems, Inc. Proprietary Technology
Results – Blouse
Sale qt.
Style: PRINTED DRESS BLOUSEColor: BLACK Size: ALL SIZES
200
250
q Size: ALL SIZESStore: ALL STORES
Real sales: 1512
150
Real sales: 1512Predicted sales: 1569Lost Sales: 57 (4%)
50
100 Predicted Sales
Real Sales
0
50
1 2 3 4 5 6 7 8 9 10 11 12 13 14Week
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Materials copyright WizData Systems, Inc. Proprietary Technology
Results – V Neck Top
Sale qt.
Style: V NECK TOPColor: IVORYSize: ALL SIZES
400
450
500
q Size: ALL SIZESStore: ALL STORES
Real sales: 2993
250
300
350
P di t d S l
Real sales: 2993Predicted sales: 3710Lost Sales: 717 (24%)
100
150
200Predicted Sales
Real Sales
0
50
100
1 2 3 4 5 6 7 8 9 10 11 12 13 14 Week1 2 3 4 5 6 7 8 9 10 11 12 13 14
Materials copyright WizData Systems, Inc. Proprietary Technology
Results ‐ V Neck Top
Sale qt.
Style: V NECK TOPColor: IVORYSize: ALL SIZES
4
5
Sale qt. Size: ALL SIZESStore: STORE #100
Real sales: 23
3Predicted Sales
Real sales: 23Predicted sales: 30Lost Sales: 7 (30%)
1
2Real Sales
0
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14Week
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Materials copyright WizData Systems, Inc. Proprietary Technology
ConclusionsConclusions
Lost Sales analysis is very difficult due to the small y ynumber of items sold and the very high variance of the data;
The proposed approach recognizes general regularities of the sales for different products and locations and provides the True Demand function;p ;
The amount of Lost Sales and the optimal distribution of products between storespcan be estimated by means of the
developed algorithm.
Materials copyright WizData Systems, Inc. Proprietary Technology
BenefitsBenefits
Lost Sales estimation can be used for future ost Sa es est at o ca be used o utu eplanning
Estimated Increase in sales (4% 22%);Estimated Increase in sales (4% ‐ 22%);
Smaller amount of merchandise sold at clearance prices
Immediate within the season responseImmediate within the season response
Our model can be used stand alone or integrated into existing systemsintegrated into existing systems
Materials copyright WizData Systems, Inc. Proprietary Technology
AuthorsIgor OfenbakhFounder WizData Systems IncFounder, WizData Systems, Inc.M.S. in Mathematics with concentration in operations research.
Nikolay KosarevNikolay KosarevPh.D. in MathematicsFields of interest: operations research, discrete optimization, artificial intelligenceartificial intelligence
Sergey Dobrovolskyh hPh.D. in Mathematics
Fields of interest: stability of chaotic and stochasticsystems, statistical modeling
Materials copyright WizData Systems, Inc. Proprietary Technology
Can we answer any questions?
You can reach Igor Ofenbakh at: 866‐949‐3282You can reach Igor Ofenbakh at: 866 949 3282or [email protected]
Materials copyright WizData Systems, Inc. Proprietary Technology