YOUR HOST FOR TODAYS WORKSHOP Kevin Anderson, Kevin R Anderson Consulting 29 year veteran of gravity models. Industry experience includes analyst at Howard L Green and Manager of Location Research BI-LO Supermarkets. Familiarity with Power:Site (DOS version of SItesPlus), SitesPlus, Model II, Business Analyst Huff Model, and SAM (Site check Research Groups proprietary C-Store, Fuel gravity/forecasting model). Kevin has an undergraduate degree in Business Administration, and a MBA from Northern Michigan University. Email: Kevin2441@Charter.net Cell: 864-884-3745 John Hefner III, Director of Facilities Research for Schnuck Markets, Inc 48-year grocery veteran in conjunction with 32 years in site research methodologies. He presently manages a three-person department at Schnuck’s Markets . John started out using the Locus model in 1981, followed by Model I & II, Power:Site and eventually Site Pro Plus. His expertise includes site location studies, metro markets analysis, acquisition strategies and due diligence for legal projects. John has an undergraduate degree in Accounting/Finance from the University of Missouri–St. Louis and a M.B.A. from Webster University. Email: jheffner@schnucks.com WHAT WILL WE LEARN TODAY: What are the basic types of Forecasting? How do other forms of forecasting come into play during the Forecasting Phase? Model definition and purpose? What inputs need to be set? What is the process? When is it a good forecast? How do I know? Some caveats when forecasting with gravity modeling. BEFORE WE START FORECASTING Make sure the model is balanced! Unbalanced models produce results that are suspect. Before starting the forecasting process look one more time at the balanced model. Collect all competitive changes and have them ready to input into the model. You can add them later, but it is always best to have your changes laid out before hand. Although it may not be possible to collect all changes, it will minimize the duplication of work if you collect it up front THINK Think about what you are doing and make a list of tactics to explore. Do not dwell on what you are doing; don’t over think the problems so much that you cannot make a decision. HOW OTHER METHODS OF SALES FORECASTING RELATE TO GRAVITY MODELING ANALOG (or Analogous Store). In models without sister stores or new markets - analogs can give us a basic framework to start on our understanding of the sales forecast. Comparing similar competitors with subject competitors can give us an indication on how our store should perform when setting the Power, Curve or other model parameters. REGRESSION (or Multiple Regression) Helps us explore correlations on what factors that make a site successful. We can take these factors and compare them to new locations to determine would be the sites with the highest probability of success, this benchmarking can be used in other areas of the organization. Information derived from these alternative ways of forecasting can be used to help the modeler increase the accuracy of the sales forecast within the Gravity Modeling environment. TYPES OF SALES FORECASTING IN GRAVITY MODELS Subject Stores and Competitive Stores. Subject Stores – Our or clients stores that are the target of the sales forecast These are the stores for which the model was built Formats and Sizes can be part of the investigation Competitive Stores – What is happening and when will happen New, remodels, expansions, closing and relocations that will happen to competitor stores (and sister stores not part of the analysis). Format and Sizes also can be explored. Alternative formats can also be explored. THE ORDER OF FORECASTING IN GRAVITY MODELING – SITESPLUS • First we start with a Balanced Model in the SitesPlus Project File. • This is a fully calibrated model for the Study Area and competitors. • Once model is fully calibrated, make sure to save an backup copy. • Second we open a new model for the competitive impacts. • Enter all competition that is projected for the model. • Enter the existing stores Loyalty Factors. • Test to see the relevance of the forecast. • Finally, open a new model for the subject site (s) • Enter the subject stores, and open all relevant competition. • Make sure all Loyalty Factors are still relevant from the previous model. • Double check the effects of the manual overrides. SITES PLUS FORM: ADDING MODELS Go and start another model within the project file’ Separate models for Competitive Changes and Subject Site Forecasting. By adding in the changes in separate models we can more effectively manage what our impacts are to sister stores and competitors Pressing this button will add a tab (to the left) Once tab appears fill out the form Filling out this field names the model This sets the opening date for the forecast Use these fields to add subtitles to the heading ADDING MODELS TO THE PROJECT FILE • ALWAYS start with a BALANCED model before Adding a New Model. • Press the Calculate and Balance Icons on the Menu Bar before adding a new model to make sure it is truly balanced. • ALWAYS start with the Competitive Changes Model. You can add all changes to this model, but only open the competitive changes. • Add all models to the Project File from the previous model. Otherwise the changes made will not be carried forward. Balanced Model Competitive Changes Model Forecast Model 1 THE STEPS INVOLVED FOR FORECASTING WITHIN SITES PLUS – LOYALTY FACTOR Once all changes are made and we are satisfied with the effects, click ok and you will exit the dialog Once al changes are made press calculate to see the effects Buds is a specialty market needs special treatment Epicurean Delight is a specialty store SOME NOTES ON LOYALTY FACTORS Add the Loyalty Factors If Needed Accessed through the Data ribbon at the top of the model. Determine which stores need loyalty curves. For rule of thumb; all Limited Assortment/Gourmet, Supercenters, Upscale stores should be reviewed for Loyalty curves, not conventional supermarkets. Some conventional supermarkets, however, may need a Loyalty Factor, after the first volume check, you may want to add them into the model. An example is a store that is getting impacted rather too severely, this may warrant a Loyalty Factor for that store Stores that typically do not get impacted by conventional stores opening should also get loyalty factors. Should be done immediately before Adding Stores. AND NOT IN THE BALANCED MODEL THE STEPS INVOLVED FOR FORECASTING WITHIN SITES PLUS – ADDING STORES First within the competitive changes model - enter the competitive changes within the model. New Stores. These are added via the Add New button on the Store Data Form. Remodels. These are typically added via the Rebuild function on Store Data Form but can also be added via the Add New button Expansions. Like Remodels use the Rebuild button on the Store Data Sheet, but can use the Add New button also. Conversions. With the Rebuild button we keep the existing data for the store before the proposed changes. All curve, manual overrides will be added to the ‘new’ record Add New command will not carry over all the curve and manual overrides. It maybe more appropriate for a new chain acquiring a store from a weaker competitor. STORES DATA ENTRY SHEET - SITESPLUS Insert New Stores Close Existing Stores Without Removing it from the model Remove the stores Make changes Remodels Relocations, & Expansions INPUTS NEEDED FOR THE SALES FORECAST. Typical store inputs needed for sales forecast: Location, size, curve, PWTA and Power. Plus any overrides also can be added to a stores. Later we will go over tips on how and what to put in the necessary field, but it is much the same for either type of forecast that you do – Subject Stores or Competitor Stores, Sales: $0 or $XXX Dependent on size and location Set by neighboring Stores; Sister Stores Chain Average or Completive Comparisons SPECIAL CONSIDERATIONS FOR EXISTING VOLUMES When making changes to stores that are in the Balanced Model and will receive a Remodel, Conversion or Expansion the analyst needs to decide whether to keep the existing (or from the Balanced Model) volume or leave the field null in the record . • Adding the existing volume to the store in the volume field • If store is on the periphery of the Study Area this option should be looked at for viability. It will keep huge swings in volume to a minimum. • Leaving volume field blank will allow more volume to be ‘captured’ from beyond. • Could inflate the sales forecast by assuming more volume could come from outside the Study Area. • If the store is in the middle of the trade area it make sense to do this because it keeps the sales from beyond (outside the study area) will not vary greatly (no more than 10% in total given the PWTA). THE EFFECTS OF PWTA AND OLD VOLUME ON FORECAST FOR RELO/EXPNS/REMODS/CONVS Scenario Keep existing Volume Leave Volume Field Blank Scenario Keep existing Volume Leave Volume Field Blank PWTA 85% 85% PWTA 35% 35% Old Volume Prior Outside Sales $200,000 $200,000 $30,000 $30,000 Old Volume Prior Outside Sales $200,000 $200,000 $130,000 $130,000 Projected Volume $250,000 $258,824 Variation Projected Volume $250,000 $628,571 Variation Post Outside Sales $30,000 $38,824 29% Post Outside Sales $30,000 $408,571 1260% Beware: If you keep the existing volume on the store it will increase the totals by that amount on the Projected Volume Tables THE STEPS INVOLVED FOR FORECASTING WITHIN SITES PLUS – SETTING THE CURVE Store Size (Total Area) 2,500 5,000 10,000 15,000 20,000 30,000 40,000 50,000 60,000 70,000 80,000 Store Size (Sales Area) 1,900 3,750 7,500 11,250 15,000 22,500 30,000 37,500 45,000 59,500 60,000 Suggested store curves based on store sizes. Source: SitesPlus v.2010,4 help file Consider a Curve of: 99.9 99 95 91 88 80 73 65 58 50 49 Curve relates to size, but also relates to location and the type of store. In models ALDI and Save-A-Lot typically have lower curves than the default. If a remodel: Keep the Curve the same and change it only if there are extenuating circumstances such as a total change in format. If expansion/relocation: Change the Curve to reflect the change in size. In new stores: Start with the Curve at default and adjust if there are sister stores nearby or the format demands it You can change (and probably should) the Curve if the physical nature of the roadways near the store changes (i.e. from a 5 lane roadway to a 4 lane median divided). THE STEPS INVOLVED FOR FORECASTING WITHIN SITES PLUS – SETTING THE CURVE (CONT) In large models or models were the Curve was adjusted for a number of stores beyond the default you will need to look at the stores that are going to be forecasted (whether Subject Stores or Competitors). • In cases of large models look at the like chains in the study area to see what the average Curve is by size. This will give you a good idea what the Curve should be in the forecast. • Some chains will be well represented in an area so their Curve’s maybe higher than defaults. • Likewise there may be few in the area and may necessitate a lower Curve. • Alternative Formats in the balanced model will have consistently lower curves than the default. • This should hold true in the forecasting models – Look at the average Curve for format/chain and apply in the forecast models. THE STEPS INVOLVED FOR FORECASTING WITHIN SITES PLUS – SETTING THE CURVE (CONT) • Special cases may exist with a Conventional store who’s image is discount and is one of only a few in an area. • The Curve may be lower than the calculated default. • If a new sister store comes to the area you may have to raise the Curve on both stores (to show realistic impacts). • Pay attention to the market share draws past sister stores or stronger competitors. • You maybe justified in having a higher (or lower) Curve than what is calculated based upon size. • Pay attention to any deviation beyond the default – are you reasons still valid for the deviation from default? Or should it be changed? THE STEPS INVOLVED FOR FORECASTING WITHIN SITES PLUS – SETTING THE PWTA Setting the PWTA is as critical as the other factors, but we just have more “indicators” as to the value. • For New and Relocated/Replacement and some Conversions stores – Look at the surrounding model stores to see what are the values of PWTA. • Exceptions arise with Alternative Formats, (i.e. limited assortment, supercenters, and upscale/gourmet). • Pay attention to the strength and placement of sister stores outside the Study Areas (if there are considerable sister stores outside the Study Area – a higher PWTA is warranted). • For Relocated/Remodels/Expansions – use the existing value from the balanced model. • May have to make adjustments in special circumstances, but those are few and far between. THE STEPS INVOLVED FOR FORECASTING WITHIN SITES PLUS – SETTING POWER One important thing to remember about Power is that it is based upon the relative position of stores within the model: all markets average Power (in the balanced model) will be 100 – this does not matter if it is an upscale or lower income area. So comparing models across markets the analyst needs to look at RELATIVE position of sister stores compared to competitors NOT ACTUAL! Look at sister stores within the model – what is the Power? Are they lower than they should be because of customer appeal? Higher? What is the size and the format of the stores? Usually Power is reduced for a larger store size especially in the replacement/expansion scenarios this is due to the fact that more of the sales for that store will be explained by the size. If the remodel expands product offering and increase or decrease of Power maybe warranted (i.e. adding more extensive perishable dept. may warrant higher Power; likewise going to a limited assortment should yield a lower Power). SETTING POWER – AN EXAMPLE Here is the average Power by chain in our sample model What should be the Power? Upscale/Service Supermarket Conventional EDLP Supermarkets Note: our chain is a EDLP operator and we put the site in at 120 – Where we right or wrong? And Why? THE STEPS INVOLVED FOR FORECASTING WITHIN SITES PLUS –MANUAL OVERRIDES With both Subject Stores and Competitive stores Manual Overrides should be used as sparingly as possible. Usually these overrides (Manual Power, Manual Curves , Sister Store, and Directional Curves) are used to explain something unusual about the store with relationship to the sectors or other stores within the model. When using the REBUILD button when adding a Expansion / Remodel or other capital expenditures be very careful about the manual overrides that automatically carry over to the ‘new’ record. You may want to change the overrides or delete them all together in order to more accurately explain the changes. Is the spotting you used to calibrate the balanced model still viable? MANUAL OVERRIDES - MANUAL POWER Store you want to affect change The Sector you want to manipulate The override Power Manual Power is more effective on the ‘close-in’ sectors (less than radius distance). They can go from 1-9999, but try to limit this value to a multiple of 4 from the default Power otherwise you may get some ‘different’ results. Higher than default, the more volume from the sector, likewise lower power than default lower amount of volume. When to Use? - When you want to explain more volume from a sector due to demographics or fit. MANUAL OVERRIDES - MANUAL CURVE Stores you want to affect a change The Sector you want to manipulate The override Curve Manual Curves work better beyond Radius Distance (or the ‘Outboard’ sectors). Values are 1-99. Lower than the balance model Curve for the store will increase volume, higher than default will decrease volume from a specific sector. Demographic fit is the best reason to use this similar to with Manual Power. When more than a few Manual Curves are used the analyst may want to consider a Directional Curve instead. MANUAL OVERRIDES - DIRECTIONAL CURVE The affected Store Compass Direction for the Curve The override for the store When to start the Curve When to end the Curve Directional Curves are similar to manual curves but will apply the same curve to all sectors impacted based upon compass directional and distance values. It is an easy way to apply Curves to a large number of sectors in a model to limit or expand the stores market share along a roadway or some other physical object. Sometimes they are used to explain Sister Store effect. Best Use – To “force’ volume directionally to explain access, demographics, and sister store effects MANUAL OVERRIDES - SISTER STORES Affected Stores Adjustment for strength of effect How the effect is calculated Sister Store Pairs command will reduce the outlets volume in sectors near sister stores from the same chain. Although the sales will ‘pull’ past each other, volume will decline at a faster rate than without the Sister Store command implemented. This effect can also be mimicked with the Directional Curve and increasing the default curve, but will not have the effect localized to the chain. With the levers and button selectors we can cut volume coming from sectors that are near the sister stores almost to nothing or have it pulling past theses stores with relative strength. MANUAL OVERRIDES - SOME RULES OF THUMB Use Directional Curves to limit or expand sales in a particular compass direction. Use if the site is located on a particularly strong roadway. You can stretch the market shares along the road so that you can more adequately explain what sectors are giving their volume to the site. Use default value to (i.e. what was used in the competition model) as your baseline. Below or above this baseline will increase or decrease volumes. Manual Powers and Curves should be used only to explain demographic fit and only if it is VERY necessary in the forecasted model. Sister Stores should be used for stores within the same chain. Some experimentation with the parameters should be done to get the desired effect. LARGE (AND METRO ) MODELS VS SMALLER MODELS • Large and Metro Models typically encompass multiple sites and a county or larger governmental unit. • Can do multiple what if scenarios to see the impacts to the store network. • Give us quick answers to questions such as; • What are the impacts from chain A coming to the area? • What are the effects of a alternative format upon our stores? What are the optimal storing strategy for an area . • Some compromises must be made when balancing (or calibrating) these models. • Will have to change default Curve for stores beyond what you would normally do; or add more manual overrides to get the model to balance. • It adds complexity to the models (i.e. Power of 100 on the north side of the model could be equivalent to a 90 on the south side of the model). • Take time to balance the model –on larger models it may take over a week to balance. LARGE MODELS VS SMALLER MODELS • Small Models are typically single or two site models • With a smaller model we have some advantages: • Able to keep the stores to default values within the model. • Designed only for one or two sites so lower number of sectors and stores to evaluate. • Easier and quicker to calibrate talking hours instead of days. • USUALLY the sales forecast is more accurate. • But we have some disadvantages: • If area is on the border or outside the model area we have to do a new model • If doing multiple models in an area it can be as time consuming as building a larger model. LARGE MODELS VS SMALLER MODELS • So which is better? • If you plan to review a large number of sites in and area or are trying to get and idea how the competitors market shares look over distance – Use the large model. • If you are going to be looking at a large number of sites over a given time frame (say 24 months) then use a large model. • If you only have a site or two in the area, use a smaller model (based upon the estimated Primary/Secondary trade areas). • Even if you expect to have a large number of sites in an area but will be spread out for the next 36 months – break them up to smaller models. • REALLY fragmented markets (Geographically or Demographically) are also candidates for smaller models. SALES FORECAST EVALUATION -VOLUME Volume of the subject site or competitive sites should be looked at with regards to the following: Does the volume make sense? Is it higher or lower than expected? What does our intuition say? What does the analog say? Is the store/format a match to the surrounding demographics? Is it the only store in the segment or are there others. How do they compare? How does the volume compare to the analog? An abnormally high volume could be an indicator of erroneous inputs into the model. Do these make sense? Is this a better site than what was anticipated by the analyst? SALES FORECAST EVALUATION – COMPETITIVE IMPACTS Another critical area to look at in evaluating a model is the competitive impacts. This is a check on the viability of the of the sales forecast and the effectiveness of the Loyalty Factors. Second look at the competitive impacts: Do the competitive impacts make sense? Are they to high – May have to Add/Adjust the store loyalty factor, curves or sister store effect. Are you impacting the supercenters too much? Are you having only limited impact upon a sister store? If they are too low – look at the other parameters in the model are they set right? The stores curve may need to be adjusted because of market changes. Sister store closing. Sister store opening Roadway change SALES EVALUATION – MARKET SHARES As with a balanced model; Market Shares should be evaluated with regard to Distance, Demographics , Natural barriers, Competitive Impacts and Sister store locations. How do the market shares over distance look? Are they falling like expected ? Are you pulling from a wrong demographic segment too strongly? Are there any unusual blips (market shares higher/lower than what you would expect)? Are you pulling strong market shares past sister stores? Are you pulling strong market shares past alternative formats or stronger competitors? If any of the previous items do not appear to be normal review the impacts. Could the inputs need adjustment? Might you need to add sisters stores or directional curves to adjust market shares? GIS – GEOGRAPHIC INFORMATION SYSTEMS Since its common adoption in the earlier adoption in the early 80’s it has helped sales forecaster get a handle on the demographics and physical geography of an area (and to produce colorful maps!) • Thematic mapping is invaluable for reviewing market shares on large and small models. • Running two themes together gives you a good idea of sister store effects. • Your competitive database can show you were stores are outside of the study area – allowing you to confirm or adjust the study area. Map showing market shares and tapestry clusters SPREADSHEETS AND STATISTICAL PROGRAMS Excel and SPSS help us conduct preliminary analysis on our competitive and customer database, along with refining our presentation materials for the Gravity Model output. Excel can take the sales forecast data from the gravity model and develop our ROI and ROIC calculations. Excel can also be used for adding additional information to the SITESPLUS report output. SPSS and other statistical programs help us determine our benchmarking with regards to customer typology and can be used to help us to determine what is exactly an acceptable market share from a sector for a store. Alteryx, Tableau and other programs help bring various data sources together for developing actionable presentation materials for the big data era. THE GROWTH CURVE Growth Curves are a good way to “Step” sales up over a period of time You can have more than one curve to use for forecasting. Having a Do Nothing curve aids in showing projected sales declining over time with respect of a lack of capital spend. New Store Curve can allow you to ramp up sales for a particular unit, especially if you include PHARMACY sales in the SitesPlus volumes. Above all else should use store data vs industry standard curves. THE CAVEATS Although Gravity Modeling is objective it is not totally objective forecasting tool. • Subjectivity is introduced by the analyst during balancing and forecasting phase – these can affect the accuracy of the forecast. • Subjectivity is also introduced during the data collection phase of the project which will also affect the sales forecast • Basic knowledge of stores (competitors and subject stores) is necessary. • Intimate knowledge probably not critical, but a knowledge of format and how the format will be accepted by customers is a necessity. • Are you using Total Sales vs Net Sales (AWV less Pharmacy, Gas, and other non food sales ), for the stores in the analysis? If you are using Total sales make sure the impacts make sense and the forecasted volume makes sense because Pharmacy sales mature at a much slower rate than grocery sales. • Knowledge of the study area (demographics , geography, etc.) • The macro factors within a model (sector radius, barriers, leakages and potential), should be set during the balancing phase and should not be changed during forecasting. • However, new changes (i.e. adding a bridge or cross point to a river or barrier) will change the dynamic of the market area. • WHEN TO CONDUCT THE POST MORTEM • Post Mortems are necessary for the continued success of the analyst, department, and company. • A look at the viability of the assumptions will help in future projects. • When reviewing the model; look at the inputs for the new stores (both competitors and subject stores). • Where were the errors? • Are the sales right for the wrong reasons or wrong for the right reasons • Once the Post Mortems are complete; publish out to the relevant individuals. • Make sure there are qualitative as quantitative analysis. • Steps for improving on the quality of the work. • Don’t just blame operations – take some responsibility. SUMMARY • Establish a process for the developing sales forecast • Look at all the data inputs for both competitor stores and subject site. • Make sure they are logical and can be defended to an outside person. • Review them with a colleague. • Review the market shares, volume and competitive impacts for anything out of the ordinary. • Make sure Alternative formats in the model behave in a way that is expected. • Make sure you start the forecasting process with a balanced model.