Forecasting and market research

1
Forecasting and market research

• PBIRG Pre-Conference Education Series

2
Outline of the presentation

• Concepts of market potential and forecasting
• Importance of forecasting
• Overview of sales forecasting methods
• Market research and sales forecasting
• Intentions
• Conjoint analysis
• SKIMs approach to forecasting
• Case study

3
Concepts of market potential and forecasting

• Potential Capable of being but not yet in
  existence the inherent ability or capacity for
  growth, development, or coming into being.
• Forecast To estimate or calculate in advance,
  especially to predict by analysis of data, with
  varying degrees of uncertainty.

?
100
Forecasting market potential
Forecasting realistic sales
Internal factors (100 awareness, 100
distribution, etc.) External factors (formulary,
reimbursement, generics, parallel trade, etc.)
What chunk of my market potential can I
realistically achieve?
Assumptions regarding internal and external
factors
Virtually impossible to achieve
4
Outline of the presentation

• Concepts of market potential and forecasting
• Importance of forecasting
• Overview of sales forecasting methods
• Market research and sales forecasting
• Intentions
• Conjoint analysis
• SKIMs approach to forecasting
• Case study

5
Importance of forecasting

• The markets are often volatile and unpredictable.
  Why do we even bother to forecast? Is it just a
  theoretical exercise?
• Strategic objective Justify investment decisions
  necessary in order to capture the largest chunk
  of the market potential pie possible (exploit the
  market opportunities to the extent possible)
• Tactical objective Provide guidance for the
  business process and various business functions
• Marketing
• Concerned with the success of individual products
  and product lines
• Annual plans of marketing efforts for new and
  existing products
• Marketing plans take into account projected
  product changes, promotional efforts, pricing,
  etc.
• Business development
• Concerned with strategic decisions (go/no go,
  in-licensing, etc.)
• Sales
• Concerned with setting goals for the individual
  members of the sales force and motivating them to
  exceed these goals
• Production supply chain management
• Concerned with ensuring sufficient supply meeting
  the demand and minimizing losses due to
  out-of-stock conditions
• Finance
• Concerned with projecting cost and profit levels
  and capital needs

6
Outline of the presentation

• Concepts of market potential and forecasting
• Importance of forecasting
• Overview of sales forecasting methods
• Market research and sales forecasting
• Intentions
• Conjoint analysis
• SKIMs approach to forecasting
• Case study

7
Overview of sales forecasting methods
Rely on intuitive judgments, opinions, and
probabilities.

• Judgmental models
• Jury of executive opinion
• Delphi method
• Judgmental bootstrapping
• Intentions
• Conjoint analysis
• Technological forecasting models
• Curve fitting
• Analogous data
• Time series extrapolation models
• Moving averages
• Exponential smoothing
• Decomposition models
• Box-Jenkins
• Causal models
• Regression analysis
• Leading indicator analysis
• Input-Output Models

Predict how others will behave
Predict own behavior
Appropriate for very new technologies and very
long-range forecasting.
Based only on past data. Use patterns, changes,
disturbances, etc. in the data to forecast the
future.
Based on relationship between predictable factors
and outcomes.
8
Judgmental models Jury of executive opinion

• Basic idea
• Combines input from key information sources.
• Simple procedure forecast meetings beginning
  with initial estimates of the forecast variables
  until consensus is reached.
• Uses the solid understanding of the participants
  who are assumed to have relevant knowledge.
• Comments
• Useful in situations where past data do not
  exist, causal relationships have not been
  identified, or some major change has occurred in
  the forecasting context which is not accounted
  for by other techniques.
• Subject to severe judgmental biases (the tendency
  to see things a certain way) that can result in
  poor forecasts.
• Mixed evidence as to the validity.

9
Judgmental models Delphi method

• Basic idea
• Opinions expressed in group settings can be
  influenced by the dominant positions of some
  participants, etc.
• The Delphi method attempts to lessen these
  potential biases resulting from group dynamics
  when making judgmental forecasts
• The procedure is handled by a coordinator and a
  panel of experts is selected.
• Initial set of questions sent to all the
  participants (for example, revenue estimates,
  likelihood that these estimates will be realized,
  minimum and maximum estimates and reasons).
• The coordinator tabulates the results and returns
  them to each participant.
• The process continues until little or no change
  occurs.
• Comments
• Eliminates need for group meetings.
• Alleviates the bias inherent in group meetings.
• Participants can change their mind anonymously.
• Time consuming.

10
Judgmental models Judgmental bootstrapping

• Basic idea
• Converts subjective judgments into objective
  procedures.
• Experts are first asked to make predictions for a
  series of conditions (for example, they could
  make forecasts for next years sales in various
  geographical regions). This process is then
  converted to a set of rules by regressing the
  forecasts against the information used by the
  forecaster.
• Comments
• Low cost procedure for making forecasts.
• Even though bootstrapping models always provide
  an improvement in accuracy in comparison with
  purely judgmental forecasts, these improvements
  are typically modest.

11
Technological forecasting models Curve fitting

• Basic idea
• A set of techniques particularly appropriate for
  very new technologies with little or no data or
  very long-range forecasting.
• Highly exploratory.
• With a few data points, subjective extrapolation
  of the trend in the data so that a forecast can
  be made (linear, S-shaped).
• Theoretical limits of the technology suggest the
  bending point of the trend.
• Comments
• Large errors likely.

12
Technological forecasting models Analogous data

• Basic idea
• Data about another product on the market or one
  that existed at an earlier time used to forecast
  a new products expected growth pattern.
• Critical to establish a logical connection
  between the sales of the two products (for
  example, the products serve a similar need).
• Trace out the expected pattern for the new
  product, taking into consideration environmental
  factors and market conditions that may uniquely
  affect the new products growth pattern.
• Comments
• One of the oldest forecasting techniques.
• Useful only if the analogy holds.

13
Time series extrapolation models

• Basic idea
• Extrapolate historic data into the future.
• The premise is that there is some underlying
  pattern in the value of the variable being
  forecast.
• Different techniques available depending on the
  type of the trend line (moving average,
  exponential smoothing, decomposition models,
  Box-Jenkins models).
• Comments
• Requires availability of data from a large number
  of consecutive past periods.

14
Causal modelsRegression analysis

• Basic idea
• Changes in the value of a main variable (for
  example, the sales of a product) are closely
  associated with changes in some other variables
  (for example, the cost of the product). So if
  future values of these other variables can be
  estimated (cost), they can be used to forecast
  the main variable (sales).
• The simplest regression analysis model YabX,
  where Y is the dependent variable (sales) and X
  is the independent variable (cost).
• If there are several past concurrent observations
  of Y and X, regression analysis allows to
  calculate the values of a and b.
• Comments
• Requires availability of data from a large number
  of consecutive past periods.

15
Causal models Leading indicator analysis

• Basic idea
• Leading indicators are industrial and economic
  statistics from which an indication of the value
  or direction of another variable might be
  obtained. For example, you may find that money
  supply indicates (leads) the future level of
  consumer spending.
• Having identified a likely leading indicator, the
  degree of lead-lag (for example, in terms of
  months or years) is explored. A simple way to
  identify whether a relationship exists and the
  degree of lead-lag is to graph the values of both
  variables over a number of time periods and then
  line up their peaks and troughs. The amount of
  adjustment needed indicates the length of time
  the leading indicator leads the variable of
  interest.
• Comments
• Difficult to apply in practice because the
  lead-lag relationships tend to be volatile.
• Regression analysis can be used to identify the
  nature of the relationship.

16
Causal modelsInput-output models

• Basic idea
• End products are typically comprised of several
  components and as a result their sales are
  interrelated.
• Procedure
• identify and quantify the interrelationships
• construct an input-output table that represents
  these interrelationships
• Comments
• Many data points are needed to get a good
  estimate of the input-output ratio.

17
Overview of sales forecasting methods
Rely on intuitive judgments, opinions, and
probabilities.

• Judgmental models
• Jury of executive opinion
• Delphi method
• Judgmental bootstrapping
• Intentions
• Conjoint analysis
• Technological forecasting models
• Curve fitting
• Analogous data
• Time series models
• Moving averages
• Exponential smoothing
• Decomposition models
• Box-Jenkins
• Causal models
• Regression analysis
• Leading indicator analysis
• Input-Output Models

Predict how others will behave
PREMISE OF MARKET RESEARCH
Predict own behavior
Appropriate for very new technologies and very
long-range forecasting.
Based only on past data. Use patterns, changes,
disturbances, etc. in the data to forecast the
future.
Based on relationship between predictable factors
and outcomes.
18
Selecting the appropriate technique

• There is rarely one best technique for any given
  forecasting situation.
• Rule In general, selection of an appropriate
  forecasting technique can be guided by the focal
  products stage in its life cycle.

Maturity
Sales
Decline
Growth
Pre-launch
Post-launch
Intentions Conjoint analysis
Time
Judgmental Technological forecasting
Time series
Causal
19
Outline of the presentation

• Concepts of market potential and forecasting
• Importance of forecasting
• Overview of sales forecasting methods
• Market research and sales forecasting
• Intentions
• Conjoint analysis
• SKIMs approach to forecasting
• Case study

20
Judgmental modelsIntentions

• Basic idea
• Straightforward questioning.
• Respondents are asked to predict how they would
  behave in various situations / under various
  assumptions.
• Especially suitable for new product forecasts
  when sales data are not available.
• Used when there is a limited number of possible
  scenarios, the product profile is fixed and no
  changes with regard to the competitors products
  are expected.
• Comments
• In contrary to the other forecasting methods,
  extra calculations are necessary to arrive at a
  forecast.

21
Judgmental models IntentionsFixed product
profile(s)
Please now review these three different product
profiles (Drug X, Drug Y, and Drug Z). Lets
assume that you have 100 patients with diabetes
who you are about to prescribe drug treatment to
and all of these drugs are available and perform
according to these specifications. How many of
these patients would you prescribe Drug X, Drug Y
and Drug Z?   Drug X _____ patients Drug Y
_____ patients Drug Z _____ patients
100 patients
Provide a good description of the target patient!
22
Judgmental models IntentionsGabor Granger
questioning (1)

• Basic idea
• A technique developed by two economists, Gabor
  and Granger, in the 1960s.
• Consumers are asked to say if they would buy a
  product at a particular price. The price is then
  changed and consumers again say if they would buy
  or not.
• From the results we can work out what the optimum
  price is for each individual and by taking a
  sample of respondents we can work out what levels
  of demand could be expected at each price point
  across the market as a whole.

23
Judgmental models IntentionsGabor Granger
questioning (2)

• In pharmaceutical market research we apply an
  adapted version of the original Gabor Granger
  technique.
• The end user of a medication does not have an
  opportunity to decide for himself / herself which
  product to take. The physician is the key
  decision maker.
• We could ask the physicians whether they would
  prescribe a drug at a particular price or not and
  repeat the question for other price points.
  However, the results would not be very revealing.
  
• Instead, we ask to what proportion of their
  (target) patients they would prescribe a given
  drug at a number of various price points.
• It is an extension of the straightforward
  questioning using a product profile(s), with
  variable prices.

24
Judgmental modelsConjoint analysis (discrete
choice)

• Basic idea
• Instead of presenting the respondents with a
  fixed product proposition or propositions, we
  present them with a large number of product
  profiles composed of product features (in the
  form of a choice task).
• By making a number of choices, the respondents
  reveal their sensitivities to the product
  features (utilities) and we can predict which
  products they would choose (given various
  assumptions).
• Especially suitable for new product forecasts
  when sales data are not available.
• Used when changes to the product profile are
  expected but hard to predict, when the
  competitive field is expected to change and the
  number of possible scenarios is large.
• Comments
• In contrary to the other forecasting methods and
  similar to intentions data, extra calculations
  are necessary to arrive at a forecast.

25
Conjoint analysis design considerations (1)

• You should only include those attributes that
  differentiate products.
• Attributes are assumed to be independent (avoid
  including attributes that correlate with one
  another).
• Each attribute level must be mutually exclusive
  of the others (a product has one and only one
  level of that attribute).
• Attribute levels should have concrete/unambiguous
  meaning.
• Attribute levels should be formulated using
  terminology used by the respondent (physician vs.
  patient).
• You should not include too many levels for any
  one attribute (the usual number is about 3 to 5
  levels per attribute).
• Attribute levels have to span the relevant range
  and beyond (we can only simulate choice behavior
  within the boundaries of our design).
• Make sure attribute levels can combine freely
  with one another without resulting in utterly
  impossible combinations (very unlikely
  combinations are OK).
• Resist temptation to make attribute prohibitions
  (prohibiting levels from one attribute from
  occurring with levels from other attributes).

26
Conjoint analysis design considerations (2)

• The conjoint design must reflect the indication /
  disease properties in terms of
• Treatment lines
• Number and type of (current and future) treatment
  modalities
• Type of patient
• Etc.
• Use of patient profiles in market potential /
  forecasting studies
• A patient profile should be representative of a
  broad group of target patients
• The patient profile should provide enough detail
  for the physician to be able to make a
  prescribing choice and feel that prescribing is
  justified
• Refrain from using very detailed patient profiles
  unless you are trying to estimate the market
  potential for a very specific segment
• Make sure that you obtain information (from
  external sources or through the survey) on the
  size of the target patient population(s)

27
Sample make-up considerations challenges

• Ideally a representative sample
• A large random sample will be representative.
  However, our samples are typically fairly small
  and thus cannot be expected to be representative.
• So in order to recruit a representative sample
  you need to have a full understanding of the
  structure of the physician population in terms of
  how many different patients they see per
  month/have in their portfolios, how many
  prescriptions they write in a given time
  interval, etc.
• In addition, we typically screen the respondents
  to ensure that
• they are personally responsible for initiating a
  given therapy
• they are knowledgeable about the subject matter
  (a sufficient number of target patients or
  prescriptions, awareness of current and future
  treatment options, etc.)
• etc.
• Solution?

NEVER AVAILABLE
28
Conditions under which conjoint results may
provide accurate forecasts

• Respondents are the decision makers for the
  product category under study
• Respondents are representative of marketplace
  decision makers (the conjoint results can be
  projected to the target market)
• The conjoint exercise induces respondents to
  process information as they would in the
  marketplace (the estimated preference model is a
  valid representation of how consumers make
  trade-offs among product attributes in the
  marketplace)
• The alternatives can be meaningfully defined in
  terms of a modest number of attributes (or
  relevant but omitted attributes can remain
  constant as respondents evaluate partial
  profiles)
• Respondents find the conjoint task meaningful and
  have the motivation to provide valid judgments
• Respondents make their marketplace choices
  independently (word-of-mouth, formulary
  committees, preferred drug lists)
• The set of attributes is complete in the sense
  that relevant variations between existing and
  future products can be accommodated in market
  simulations
• The set of alternatives for which the analyst
  makes predictions is the set the respondent
  considers
• Respondents predicted choices are weighted by
  their purchase / prescribing intensities

29
Other considerations

• Experts argue that conjoint analysis works best
  (provides most accurate predictions) for
  continuous innovation / mature products
  (relatively minor changes in existing products or
  services). Respondents know the product category
  and can easily imagine their liking for possible
  product configurations.
• For discontinuous innovations, it is still
  possible to use conjoint analysis provided that
  respondents are educated about the alternatives
  from which they are asked to choose.
• Market conditions change
• Advertising and sales efforts may modify
  respondents awareness of products and the
  relevance of individual attributes in their
  choice decision over time. Respondents
  preference may also be influenced by opinion
  leaders and word of mouth.
• Such dynamic elements increase the difficulty of
  validating conjoint-based predictions against
  marketplace behavior.
• Essentially, the conjoint study is static in that
  it provides forecasts of marketplace behavior for
  various conditions but does not indicate how the
  forecasts change over time. However, one can
  allow, for example, awareness and availability of
  products to vary over time as a function of
  planned marketing activities.

30
Outline of the presentation

• Concepts of market potential and forecasting
• Importance of forecasting
• Overview of sales forecasting methods
• Market research and sales forecasting
• Intentions
• Conjoint analysis
• SKIMs approach to forecasting
• Case study

31
Difference between modeling product uptake and
predicting peak sales
Conjoint market simulator assumptions Equal
availability (distribution)Respondents are aware
of all productsLong-range equilibrium (equal
time on market)No out-of-stock conditions

No other limitations (reimbursement,
formulary, preferred drug lists)
PEAK SALES
32
SKIMs approach to forecasting three building
blocks
33
Preparing for a market potential / forecasting
study

• 8 step guide to forecasting
• Step 1 Find out in what terms you have to report
  your findings (number of tablets, number of
  patients, revenues, etc.)
• Step 2 Determine the relevant time interval
  (month, quarter, year, etc.)
• Step 3 Identify target patients devise /
  obtain a disease model
• Step 4 Decide how to determine the size of the
  target patient population
• Step 5 Decide how you are going to collect
  predictions of Rx
• Step 6 Determine disease-related assumptions
  about treatment (duration per type of patient,
  compliance rate, etc.)
• Step 7 Gather internal marketing assumptions
  (prices, distribution, size of the sales force,
  promotions, etc.)
• Step 8 Determine relevant environmental
  (external) factors

34
Step 1 Reporting the findings

• Knowing exactly what you are going to be
  forecasting
• Find out (ideally already in the proposal phase)
  how you have to report your findings
• What do the clients mean when they say estimate
  market potential? Number of patients, number of
  tablets, sales or revenues?
• Do not delay this decision till the analysis
  phase it might be too late

35
Step 2 Relevant time interval

• Knowing exactly what time interval you are going
  to be covering
• Find out (ideally already in the proposal phase)
  what time period your client is interested in
  (month, quarter, year)
• Do not delay this decision till the analysis
  phase it might be too late

36
Step 3 Identification of target patients

• Determining the characteristics of a target
  patient
• A target patient is the patient who
• is suffering from the condition of interest
• is eligible for treatment (is not excluded from
  treatment)
• possesses certain clinical or socioeconomic
  characteristics
• Devise / obtain a disease or patient flow model
• Decide whether patients suffering from conditions
  other than the one the product will be indicated
  for should be included in your model (off-label
  prescribing)

37
Step 3 Identification of target patients
Disease or patient flow modeling

• What factors should you take into account when
  modeling the disease?
• Total patient population
• Drug treated versus not treated (prevalence vs.
  treatment seeking)
• Symptomatic versus non-symptomatic
• Contraindications (patients suffering from
  concomitant diseases)
• Exclusion factors (patients rejecting treatment,
  patients who cannot tolerate a certain
  administration form, etc.)
• Treatment line
• Death rates
• Etc.

38
Step 4 Size of the target patient population

• Sizing the target patient population
• Do reliable secondary (epidemiological) sources
  of information exist?
• Current or projected patient numbers?
• In case there are no reliable external sources
  and you need to collect the necessary information
  through the survey, make sure that you follow the
  disease model you have devised to identify the
  target patient population

39
Step 5 Predictions of prescribing

• Deciding how you are going to collect predictions
  of prescribing
• Product profile evaluation and collecting
  intentions data or conjoint analysis?

40
Step 6 Disease-related assumptions

• Determining disease-related assumptions about
  treatment with the product under investigation
• Duration of treatment (per type of patient)
• Minimum duration of treatment
• Response / failure rate
• Compliance rate
• Etc.

41
Step 7 Internal marketing assumptions

• Gathering internal marketing assumptions
• Price(s)
• Distribution
• Size of the sales force(s)
• Promotions
• Reimbursement
• Formulary
• Preferred drug lists
• Etc.

42
Step 8 Relevant environmental factors

• Determining relevant environmental factors
• Generic substitution
• Reimbursement legislation
• Parallel trade
• Reference pricing

43
SKIMs forecasting principles

• KEY to pharmaceutical forecasting understanding
  the disease
• Have the conceptual framework in place before you
  design the study and write the questionnaire
• Review patient epidemiology data available from
  the client prior to questionnaire writing and
  decide what else you need to find out
• Make sure all the questions are asked in a
  conceptually consistent manner (use the same
  denominator)
• For conjoint-type studies, base case in place
  before fielding
• Aggregate analysis at the country level (often
  reliable patient numbers are available at this
  level only) with weighted shares of choice
• In the presentation be as transparent as
  possible about the assumptions and variables used
• Forecasting studies 90 of the work is done in
  the set-up phase

44
Outline of the presentation

• Concepts of market potential and forecasting
• Importance of forecasting
• Overview of sales forecasting methods
• Market research and sales forecasting
• Intentions
• Conjoint analysis
• SKIMs approach to forecasting
• Case study

45
CASE STUDY
Handouts