A loyalty program: a connection between personalized marketing and the points pressure mechanism

Thanks to the data obtained, it is clear that none of the respondents chose the answer "Daily". The answer option "Less than once a month" was chosen by 191 respondents (80% of 240 respondents), the answer option "More than once a month, but less than once a week" was chosen by 38 respondents (16% of 240 respondents), and only 11 respondents (4% of 240 respondents) chose the answer option "More than once a week, but not daily".

Figure 23. Frequency of receiving notifications.

Based on the data obtained, the most common channel is SMS messages, through which 212 respondents receive notifications (88% of 240 respondents), the second place is the e-mail channel, through which 132 respondents receive notifications (55% of 240 respondents). The least common is the distribution of alerts via social networks, this channel was chosen by only 11 respondents (5% of 240 respondents), and 56 respondents (18% of 240 respondents) receive notifications via mobile apps.

Also, among the answer options were Apple Wallet, getting information in person in the store, and receiving a call.

Figure 24. Channels of receiving notifications.

According to the figure 25, the majority of respondents receive notifications only by 1 channel, such as 111 respondents (46%) out of 240. In second place - the experience of receiving notifications through 2 channels (38%). 32 people get through 3 channels and only 5 people get through 4. the Average value is 1.33. the mode and median are the same and equal to 1.

Figure 25. Distribution of number of communication channels used by respondents.

The next step was to evaluate the degree of agreement with the statements on a scale, where 1-completely disagree, and 5 - completely agree. The developed statements are aimed at identifying under what circumstances the respondent spent their bonus points when reaching the threshold.

Statement 1: when a respondent receives a message that their points are about to expire, the Respondent always makes purchases (1 strongly disagrees and 5 strongly agrees).

Figure 26. Distribution of respondents' ratings of agreement with statement 1.

The results show that the majority of respondents do not agree with this statement (about 45% of the 240 respondents) and they do not spend their points after receiving an alert. And only 7 respondents (3% of 240 respondents) agree with the statement that they always make purchases when they receive notifications that points will soon burn out. 53 respondents (22% of 240 respondents) expressed a neutral attitude and 16 respondents (7% of 240 respondents) are more likely to make a purchase.

Statement 2: “after receiving a message about points burning, I would like to make purchases only if there is money for it” - here the factor of influence of the availability of free money on the probability of making a purchase is evaluated (1 strongly disagrees and 5 strongly agrees).

Thanks to the data obtained, it is clear that 50 respondents (21% of 240 respondents) answered with the maximum confidence that this factor is very significant and had an impact on their intention to make a purchase, another 54 respondents (23% of 240 respondents) rated the degree of this statement at "4", we can conclude that they rather agree with this statement, 60 respondents (25% of 240 respondents) rated the degree of this statement at "3", which implies an average response between "agree" and completely "disagree", we can conclude that, that the respondents did not fully understand the idea of this question, or that it is difficult for them to imagine such a situation. Another 29 respondents (12% of 240 respondents) rather disagree with this statement, and 47 respondents (20% of 240 respondents) completely disagree with this statement.

Figure 27. Distribution of respondents' ratings of agreement with statement 2.

Statement 3: when a respondent receives a message that their points are about to expire, "they only make a purchase if they need to buy a certain item in this store", where 1-completely disagree, and 5-completely agree.

Figure 28. Distribution of respondents' ratings of agreement with statement 3.

Only 12 respondents (5% of 240 respondents) completely disagree with the statement 3, another 12 respondents (5% of 240 respondents) rather disagree with this statement, 30 respondents (13% of 240 respondents) rated the degree of this statement at "3", which implies a response of average significance between "agree" and completely "disagree". Another 74 respondents (31% of 240 respondents) rather agree with this statement and 112 respondents (47% of 240 respondents) completely agree with this statement.

Statement 4: when the Respondent receives a message that the points will soon be burned, "he ignores it and does not make a purchase", the answer options here are 1-completely disagree, and 5-completely agree.

Figure 29. Distribution of respondents' ratings of agreement with statement 4.

Thanks to the data obtained, it is clear that 12 respondents (5% of 240 respondents) are members of loyalty programs, but the presence of accumulated points and the received notification of the burning of these points does not affect their intention to make a purchase. 36 respondents (15% of 240 respondents) rather disagree with this statement. 82 respondents (34% of 240 respondents) rated the degree of this statement at "3", which implies a response of average significance between “agree” and “completely disagree”. Another 42 respondents (18% of 240 respondents) rather agree with this statement and 68 respondents (28% of 240 respondents) completely agree with this statement. Based on the results obtained, it can be concluded that the received notifications affect the intention to make a purchase and spend the accumulated funds only for 12 respondents (5%) out of 240 respondents who are members of loyalty programs, as well as 36 respondents (15%) out of 240 respondents may be affected by the received notifications. And 228 respondents out of 240 are more likely not to be affected by the received notifications and make a purchase. Based on this, we can conclude that the system of notification of loyalty program participants does not have a very effective effect on the customers ' intention to spend the accumulated points and make purchases. Further the results of this question will be compared with answers devoted to simulated situations to analyze how they differ.

Preference of get notifications about threshold.

Based on figure 30, it can be seen that the most convenient channel for receiving notifications, according to respondents, is SMS notifications. This channel was chosen by 205 respondents (67% of all respondents). Next, for convenience, respondents chose such a channel as notifications in apps, 93 respondents voted for it (30% of 308 respondents). The next convenient channel was chosen for notifications via E-mail, which was chosen by 85 respondents (28% of all respondents). Then there is a notification channel such as social networks, which was chosen by 23 respondents (7% of all respondents). It is worth noting that 33 respondents (14% of all respondents) expressed a desire not to receive such notifications. Also, 1 respondent did not choose any option and answered "I don't know". loyalty rewarding customer behavior

Figure 30. Preferred notification channels.

The average number of used channels is 1.5. the Mode and median are the same and equal. The graph below (figure 31) shows that most often 184 respondents (60% of all respondents) chose only 1 channel. In second place there are 2 communication channels, with this frequency 97 respondents (32%) answered. 3 channels were selected by 24 respondents and only 2 people answered that they are comfortable receiving notifications through all 4 channels.

Figure 31. Distribution of the number of respondents' responses about Preferred notification channels.

Next, a histogram (figure 32) with the most convenient time intervals identified, for which respondents would like to receive notifications about the accumulated points and their usage periods. Thanks to the data obtained, it is clear that the most convenient time period for receiving notifications about accumulated points and their usage periods is to send notifications one week before the points are expired.

Figure 32. Preferred time intervals for receiving notifications.

This answer option was chosen by 135 respondents (44% of all respondents). The next, most convenient time period for receiving the notification was the option - 2-3 weeks before the date of points expiration, it was chosen by 105 respondents (34% of all respondents). The next option was the answer option - a month before the points are expired, it was chosen by 83 respondents (27% of all respondents). The least common response option was the option - 2-3 days before the points are burned, it was chosen by 32 respondents (10% of all respondents). We also received 11 responses from respondents (4% of all respondents) with "other" options: "I don't know", "it doesn't matter", "I don't want to receive".

Figure 33. Distribution of the number of responses from respondents about the preferred time intervals for sending notifications.

It can be seen from figure 33, respondents most often chose 1-time period, this option was chosen by 243 people (79%). There are 2 time periods in the second place, they were chosen by 50 respondents (16%). Only 4 people voted in 3 time periods - this is 1% of the entire sample. For all 4 periods, no one voted. The average was time intervals 1,2. The mode and median are the same and equal to 1.

Situations.

In this section, situations were modeled, the purpose of which was to put the respondent in the place of a fictional character and find out how the respondent would behave in a particular situation.

In situation 1, the respondent needs to put themselves in the place of the fictional Elena, who is a member of the loyalty program in the coffee shop. Elena has an app that displays her accumulated points in the amount of 100 points equal to 100 rubles, which she can spend within 5 days, otherwise the points will be burned. the coffee shop did not send notifications about the accumulated points and the period of their use to Elena. However, Elena can only spend 50% of the accumulated points (50 rubles), and goes to this coffee shop once every two weeks. Will Elena spend the accumulated points? The purpose of putting the respondent in this situation is to determine whether the value of remuneration is important for the respondent, when the respondent did not receive notification about the terms of use of accumulated points and when he or she received such notification.

Figure 34. Situation 1 (using accumulated points without receiving a notification).

As a result, when asked whether Elena will spend the accumulated points without receiving a notification about the terms of use of points, 148 (48% of all respondents) said that they would use the accumulated points even without receiving an alert about the terms of use, and 160 (52% of all respondents) said that they would not spend the accumulated points without receiving an alert.

While, when asked whether Elena would spend the accumulated points if she received an alert about the availability and terms of use of points, 251 (81% of all respondents) answered that they would spend the accumulated points if they received an alert. And only 57 respondents (19%) said that even if they received an alert about the availability of accumulated points and the timing of their use, they would not use the accumulated points.



Figure 35. Situation 1 (using accumulated points without receiving a notification).

Next, the question was raised about the most convenient way to receive notifications by Elena in a coffee shop. The figure 36 shows that the most convenient way to receive notifications is via SMS, as 155 respondents expressed their preference (50% of all respondents), the next channel in the notification rating is - notifications in the app, and 127 respondents expressed their preference for this channel (41% of all respondents). The least common channel choices are E-mail alerts, preferred by 15 respondents (5% of all respondents) and receiving notifications via social networks, preferred by 11 respondents (4% of all respondents).

Figure 36. Situation 1. The most convenient channel of receiving notifications.



The next question also relates to the first situation and is aimed at identifying the most convenient period of time for which the coffee shop should send notifications to Elena, based on the condition that Elena goes to this coffee shop once every two weeks.

Based on the data obtained, it can be said that the most convenient time period for receiving notifications about points burning is one week before points burning, this is how 143 respondents (46% of all respondents) voted. In second place, 97 (31%) respondents answered “2 weeks before the threshold”, 64 (21%) respondents answered in 2-3 days, and only 4 people (1%) preferred the answer option in 1 day.

Figure 37. Situation 1 (the most convenient time interval for sending an alert).

Then a new fictional situation 2 was presented, in which the respondent was supposed to put himself in the place of the fictional character Maria, who received an alert about the burning of loyalty points at a grocery store that she goes to no more than once a week. Respondents were given the opportunity to choose the material significance of remuneration in order to identify the dependence of making a purchase, and therefore spending points on the material significance of remuneration.

Based on the bar chart (figure 38) below, you can see that 100 rubles is a significant amount for 10% of respondents (29 people). For 67 (22%) respondents, 200 rubles is enough to spend in a grocery store. 102 people voted for the category of 300 rubles, and 110 respondents (36% of all respondents) voted for the answer "it will not go, since it has already been purchased".

Figure 38. Situations 2 (the effect of deducting part of the purchase amount from bonus points on the intention to make a purchase).

In the situation 3, respondents are given the opportunity to imagine themselves in the place of the fictional character Mikhail, who has accumulated 800 points (1 point is equal to 1 ruble) and these points will burn out within 10 days. You can use points to pay for part of your purchase at a pharmacy. This situation was developed in order to identify the material significance of the accumulated points.

Based on the results obtained, 155 respondents (50% of all respondents) agree to make a purchase with the ability to pay with accumulated points 50% of the purchase price. Another 118 respondents (38% of all respondents) are ready to make a purchase if they can pay 100% of the purchase price with their accumulated points. Further, 19 respondents (6% of all respondents) are ready to make a purchase, having the ability to pay 30% of the purchase price, and 16 respondents (5% of all respondents) are not ready to make purchases for any of the proposed options.



Figure 39. Situation 3 (the effect of deducting a percentage of the purchase amount due to bonus points on the intention to make a purchase).

3.2 Frequency statistics

Table 2 Tabulation of getting notification about reaching threshold and intention to make a purchase with notification in situation 1

Get notifications about threshold	Intention to buy
	No	Yes	Total
No (number of responses)	13	55	68
(%)	4.22%	17.86%	22.08%
Yes (number of responses)	44	196	240
(%)	14.29%	63.64%	77.92%
Total (number of responses)	57	251	308
(%)	18.51%	81.49%	100%

Table 2 compares the response rate of respondents regarding their experience of receiving notifications about reaching a threshold and their intention to make a purchase after being notified of reaching a threshold. Based on the data obtained, this table shows that 13 respondents (4% of all respondents) do not receive an alert and said that they will not make a purchase after receiving the alert. 196 people are already receiving alerts and are ready to make a purchase after receiving the alert.

Table 3 Tabulation of getting notification about reaching threshold and preference not to get notifications

Get notifications about threshold	Prefer not get notifications
	No	Yes	Total
No (number of responses)	60	8	68
(%)	19.48%	2.60%	22.08%
Yes (number of responses)	215	25	240
(%)	69.81%	8.12%	77.92%
Total (number of responses)	275	33	308
(%)	89.29%	10.71%	100%

Table 3 compares the response rate of respondents regarding their experience of receiving alerts about reaching a threshold and choosing not to receive alerts about reaching a threshold. Based on the data obtained, this table shows that 215 respondents (70% of all respondents) do not receive alerts about points burning, but would like to receive such alerts, and 60 respondents (19% of all respondents) receive and want to continue receiving such alerts, and there are only 8 respondents (3% of all respondents) who receive such alerts, but would not like to receive them.

Table 4 Tabulation of getting notification through app and preference of getting notifications through app

Get notifications by app channel	Prefer notifications by app channel
	No	Yes	Total
No (number of responses) (%)	151 62.92%	33 13.75%	184 75.67%
Yes (number of responses) (%)	22 9.17%	34 14.17%	56 23.33%
Total (number of responses) (%)	173 72.08%	67 27.92%	240 100%

In the table 4 the response rates of respondents regarding their experience of receiving alerts about approaching a threshold via the app channel and their preference for receiving alerts via the app are compared. Based on the data obtained, this table shows that 151 respondents (63% of 240 respondents) do not use the app notification channel and do not want to use it, while only 34 respondents (14% of 240 respondents) use this channel and would like to continue receiving notifications through it.

Approximately the same number (33 respondents) do not receive notifications via apps, but would like to receive them.

Table 5 Tabulation of getting email notifications and preference of getting email notifications

Get notifications by e-mail channel	Prefer notifications by e-mail channel
	No	Yes	Total
No (number of responses)	94	14	108
(%)	39.17%	5.83%	45.00%
Yes (number of responses)	72	60	132
(%)	30.00%	25.00%	55.00%
Total (number of responses)	166	74	240
(%)	69.17%	30.83%	100%

The table 5 compares the response rates of respondents regarding the experience of receiving notifications about approaching the threshold via e-mail and the preference to receive notifications via e-mail. Based on the data obtained, this table shows that 94 respondents (39% of 240 respondents) do not use this channel, and do not want to receive notifications through this channel. 72 respondents (30% of 240 respondents) receive notifications via this channel, but would not like to receive notifications via e-mail, 14 respondents (6% of 240 respondents) do not receive notifications via e-mail, but would like to, and 60 respondents (25% of 240 respondents) receive and would like to continue receiving notifications via this channel.



Table 6 Tabulation of getting sms notifications and preference of getting sms notifications

Get notifications by sms channel	Prefer notifications by sms channel
	No	Yes	Total
No (number of responses)	22	6	28
(%)	9.17%	2.50%	11.67%
Yes (number of responses)	61	151	212
(%)	25.42%	62.29%	88.33%
Total (number of responses)	83	157	240
(%)	34.58%	65.42%	100%

Table 6 compares the response rates of respondents regarding their experience of receiving alerts about reaching a threshold via sms and their preference for receiving alerts via sms. Comparing the responses of respondents, it is clear that 151 respondents (63% of 240 respondents) receive notifications via sms and would like to receive them through this channel, 61 respondents (25% of 240 respondents) receive notifications via sms, but would not like to receive them through this channel. Only 22 respondents (9% of 240 respondents) do not receive sms notifications and would not like to receive them through this channel, and 6 respondents (3% of 240 respondents) do not receive sms notifications at the moment, but would like to receive them through this channel. Based on the data obtained, it is clear that the majority of respondents do not want to receive notifications via sms, but they already receive notifications via this channel.

Table 7 Tabulation of getting notifications via social networks and preference of getting notifications via social networks

Get notifications by social network channel	Prefer notifications by social network channel
	No	Yes	Total
No (number of responses)	214	15	229
(%)	89.17%	6.25%	95.42%
Yes (number of responses)	8	3	11
(%)	3.33%	1.25%	4.58%
Total (number of responses)	222	18	240
(%)	92.50%	7.50%	100%

Table 7 compares the response rates of respondents regarding their experience of receiving alerts about expiration of accumulated points via the social network channel and their preference for receiving alerts via the social network. Based on the data obtained, this table shows that 214 respondents (89% of all respondents) do not receive notifications and do not want to receive them through social networks, while those who receive notifications through this channel and would like to continue receiving only 3 respondents (1% of 240 respondents).

Table 8 Tabulation of intention to buy with notification in situation 1 and channel of communication in situation 1

Situation 1. Intention to buy with notification	Situation 1. Preferred channel for getting notification
	E-mail	Sms	Social network	App	Total
No (number of responses)	6	16	1	34	57
(%)	1.95%	5.19%	0.32%	11.04%	18.51%
Yes (number of responses)	9	139	10	93	251
(%)	2.92%	45.13%	3.25%	30.19%	81.49%
Total (number of responses)	15	155	11	127	308
(%)	4.87%	50.32%	3.57%	41.23%	100%

Table 8 compares the frequency of responses from respondents regarding the intention to make a purchase after notification of reaching the threshold and the most convenient notification channel. It is observed that 139 (45% of all respondents) chose the SMS channel as the most convenient. In second place, receiving notifications via the app was chosen by 93 respondents (30% of all respondents). Social media and E-mail selected approximately the same number of respondents, 10 and 9, respectively (3% of all respondents).



Table 9 Tabulation of intention to buy with notification in situation 1 and choice of timeframe in situation 1

Situation 1. Intention to buy with notification	Situation1.Preferred timeframe for getting notification
	1 day	2-3 days	1 week	2 weeks	Total
No (number of responses)	1	8	22	26	57
(%)	0.31%	2.60%	7.14%	8.44%	18.51%
Yes (number of responses)	3	56	121	71	251
(%)	0.97%	18.18%	39.29%	23.05%	81.49%
Total (number of responses)	4	64	143	97	308
(%)	1.30%	20.78%	46.43%	31.49%	100%

Table 9 compares the frequency of responses from respondents regarding the intention to make a purchase after notification of reaching the threshold and the most convenient time period for sending the notification. Comparing the responses of respondents regarding situation 1 and the evaluation of the proposed time intervals, it is clear that the respondents who will make a purchase after the notification gave the highest number of preferences for receiving notifications in the week before the expiration of points - 121 people (39% of the respondents). In 2nd place, the answer is "two weeks before the expiration of points", 71 people voted for it (23% of 308 respondents). It can be concluded that receiving notifications about points expiration in a week of actual date of points expiration is the most convenient option.

Table 10 Tabulation of intention to buy with notification in situation 1 and sum in situation 2

Situation 1. Intention to buy with notification	Situation 3. Discount in percentages sufficient for Intention to buy
	30%	50%	100%	Not spend	Total
No (number of responses)	4	20	24	9	57
(%)	1.30%	6.49%	7.79%	2.92%	18.51%
Yes (number of responses)	15	134	94	8	251
(%)	4.87%	43.51%	30.52%	2.60%	81.49%
Total (number of responses)	19	154	118	17	308
(%)	6.17%	50.00%	38.31%	5.52%	100%

Table 10 compares the frequency of respondents ' responses about their intention to make a purchase after notification of reaching the threshold and the percentage discount that will be sufficient to make a purchase in the described situation 2. Based on the data obtained, this table shows that without receiving an alert, the most valuable reward for respondents was the opportunity to pay with accumulated points 100% of the purchase price, this option was chosen by 24 respondents (8% of all respondents), while when receiving an alert about the burning points, the most frequent response was to pay 50% of the purchase, 134 respondents voted for it (44% of all respondents)

Table 11 Tabulation of intention to buy with notification and sum in situation 3

Situation 1. Intention to buy with notification	Situation 2. Discount in rubles sufficient for Intention to buy
	100	200	300	Not spend	Total
No (number of responses)	2	7	23	25	57
(%)	0.65%	2.27%	7.47%	8.12%	18.51%
Yes (number of responses)	27	60	79	85	251
(%)	8.77%	19.48%	25.65%	27.60%	81.49%
Total (number of responses)	29	67	102	110	308
(%)	9.42%	21.75%	33.12%	35.71%	100%

In the table 11 the frequency responses of the respondents about the intention to make a purchase after receiving the alert about threshold and discounts in rubles, which will be enough to make a purchase in the situation described 3 are compared. Based on the data, this table shows that without the receiving of notification, the respondents prefer not to make a purchase, this option answers chose 25 respondents (8% of all respondents), 23 respondents (7% of all respondents) will make a purchase without a notification in case if it is possible to redeem 3000 points or 300 rubles . It is seen that in case of receiving notification the majority (85 respondents, 28% of all respondents) will not make a purchase, and 79 respondents are ready to make a purchase if they have the opportunity to deduct 300 rubles from the purchase price.

3.3 Regressions

To confirm the first hypothesis, Conditional (fixed-effects) logistic regression was used, since the dependent variable is binary and with a fixed effect relative to the variable “X1. Notification”, which showed whether an alert was sent or not (alternating values of 0 and 1). Data was structured using the “Respondents” variable, which describes the respondent's number.

Table 12 Conditional (fixed-effects) logistic regression

Y Intention to buy	Odds Ratio	St.Err.	t-value	p-value	[95% Conf	Interval]	Sig
X1. Notification	4.962	1.067	7.45	0.000	3.256	7.561	***
Mean dependent var	0.500	SD dependent var	0.501
Pseudo r-squared	0.347	Number of obs	308.000
Chi-square	74.666	Prob > chi2	0.000
Akaike crit. (AIC)	142.209	Bayesian crit. (BIC)	145.946
*** p<0.01, ** p<0.05, * p<0.1

In the “X1. Notification” variable, the value " 0 "corresponds to the absence of an alert, and value “1” corresponds to the sent notification about reaching the threshold. In the “Y Intention to buy variable”, the value “0” corresponds to the intention not to make a purchase, and “1” corresponds to the intention to make a purchase.

Following conclusions can be drawn from the table 12 above:

1) the model is adequate to the studied data: Prob > chi2 = 0.000;

2)the b-coefficient of the model is statistically significant: sig < 0.001;

3) changing the level of the variable "X1. Notification" from 0 to 1 increases the chances of a positive value of the variable "Y Intention to buy" by 4.962 times (Sending an alert when reaching the threshold of points expiration increases the probability that the buyer will have the intention to make a purchase);

4)Log likelihood = -70.104583,

5)Pseudo r-squared = 0.347

Based on data obtained from the conditional logistics model, H 1.0: Receiving notifications about points expiration does not affect customer intention to make a purchase - was not confirmed, H 1.1: Receiving notifications about points expiration affect customer intent to make a purchase - was confirmed.

In order to test the second and third models, it was decided to expand the already built model by adding independent variables “Channel” and “Timeframes”. Data was structured using the “Respondents” variable, which describes the respondent's number (1 1 2 2 3 3, etc.)

Table 13 Conditional (fixed-effects) logistic regression

Y Intention to buy	Coef.	St.Err.	t-value	p-value	[95% Conf	Interval]	Sig
X1. Notification	1.602	0.215	7.45	0.000	1.180	2.023	***
X2. Channels	0.000	(Omitted)	.	.	.	.
X3. Timeframes	0.000	(Omitted)	.	.	.	.
Mean dependent var	0.500	SD dependent var	0.501
Pseudo r-squared	0.347	Number of obs	310.000
Chi-square	74.666	Prob > chi2	0.000
Akaike crit. (AIC)	142.209	Bayesian crit. (BIC)	145.946
*** p<0.01, ** p<0.05, * p<0.1

In the variable “X1. Notification” value “0” corresponds to absence of a notification, value “1” corresponds to presence of notification about reaching threshold. In the variable “Y Intention to buy” value “0” corresponds to the intention not to make a purchase, and value “1” corresponds to the purchase intention. The variable “X2. Channels” corresponds to the most preferred selected channel (1.E - mail; 2. Sms; 3. Social media; 4. App), and "X3. Timeframes" corresponds to the most preferred time period for which respondents would like to receive an alert (1. 1 day; 2. 2-3 days; 3. 1 week; 4. 2 weeks).

According to the table 13 following conclusion can be:

1) the model is adequate to the studied data: Prob > chi2 = 0.000;

2) not all b-coefficients of the model are statistically significant: sig < 0.001;

The only p-value of the "X1. Notification" variable is statistically significant and affects the intention to make a purchase. The remaining coefficients are greater than 0.05 and therefore statistically insignificant. With a high probability, the coefficients can be equal to 0.

3) Changing the level of the variable "X1. Notification" from 0 to 1 increases the chances of a positive value of the variable "Y Intention to buy" by 4.445 times (Sending an alert when reaching the threshold increases the probability that the buyer will have the intention to make a purchase);

4) Log likelihood of this model is equal to -66.24825, this is less than in the table 12 (-70.104583). Pseudo r-squared = 0.383, which is higher than the model in the table 12 (Pseudo r-squared = 0.347), which indicates that the quality of the model has improved.

5) However, “X2. Channels” and “X3. Timeframes” were omitted because of collinearity. When variables are related to each other, they are essentially carriers of the same information, and their standard errors overlap, which distorts the results.

Time and channel were omitted because of collinearity. Hence, this model can't be used. Due to this reason correlation matrix was examined.

Table 14 Matrix of correlations

Variables	(1)	(2)	(3)	(4)
(1) Y Intention to buy	1.000
(2) X1. Notification	0.350	1.000
(3) X2. Channels	0.275	0.887	1.000
(4) X3. Timeframes	0.307	0.945	0.862	1.000

In the variable “X1. Notification” value “0” corresponds to no alerts, value “1” means that notification about reaching threshold was sent. In the variable “Y Intention to buy”

a value of “0” corresponds to the intention not to make a purchase, and “1” correspond to the purchase intention. The variable “X2. Channels” corresponds to the most preferred selected channel (1.E - mail; 2. Sms; 3. Social media; 4. App), and "X3. Timeframes" corresponds to the most preferred time period for which respondents would like to receive an alert (1. 1 day; 2. 2-3 days; 3. 1 week; 4. 2 weeks).

Based on the table 14, the following conclusions can be drawn: there is a strong correlation between the variables “X2. Channels” and “X1. Notification” (0.887), the variables correlate positively, and there is a very strong correlation between the variables “X3. Timeframes” and “X1. Notification” (0.945), the variables correlate positively.

Hence, it is confirmed that we cannot build a conditional logit model for these variables due to their collinearity.

Therefore, it was decided to build a logistic regression separately for the dependent variable "Situation 1. Intention to buy with notification" and the independent variables "Situation 1. Preferred timeframe for getting notification", "Situation 1. Preferred channel for getting notification". This model was chosen to remove grouping by factors and see how the above independent variables will affect those who have an intention to make a purchase after the notification.

Table 15 Logistic regression

Situation 1. Intention to buy with notification	Odds Ratio	St.Err.	t-value	p-value	[95% Conf	Interval]	Sig
Situation 1. Preferred timeframe for getting notification	0.673	0.142	-1.88	0.060	0.446	1.017	*
Situation 1. Preferred channel for getting notification	0.744	0.108	-2.03	0.042	0.559	0.989	**
Constant	36.581	28.450	4.63	0.000	7.966	167.981	***
Mean dependent var	0.815	SD dependent var	0.389
Pseudo r-squared	0.031	Number of obs	308.000
Chi-square	9.272	Prob > chi2	0.010
Akaike crit. (AIC)	291.784	Bayesian crit. (BIC)	302.975
*** p<0.01, ** p<0.05, * p<0.1

In the "Situation 1. Intention to buy with notification" variable, the value "0"corresponds to no intention to make a purchase after the notification, and "1" means that there is an intention to make a purchase after the notification of approaching the threshold. The "Situation 1. Preferred timeframe for getting notification" and "Situation 1. Preferred channel for getting notification" variables have multi-level encoding for selecting the most convenient notification channel for respondents (1. E-mail; 2. Sms; 3. Social media; 4. App) and the most convenient time interval (1. 1 day; 2. 2-3 days; 3.1 week; 4. 2 weeks), for which it is better to send an alert.

Conclusions from the table 15:

1) the model is adequate to the studied data: Prob > chi2 = 0.010;

2) Log likelihood = -142.8921, higher than the Conditional (fixed-effects) logistic regression in the table 13. Pseudo r-squared = 0.031, lower than that of the model in table 13. This indicates that the quality of the model has deteriorated;

3)Not all b-coefficients in the model are statistically significant: sig < 0.05; only "Constant" has a p-value= 0.000, and "Situation 3. Discount in percentages sufficient for intention to buy" and "Situation 1. Preferred channel for getting notification" have a p-value equal to 0.699 and 0.122, respectively. Therefore, we can assume that most likely there are other factors that can influence the choice.

4) Only reference group 1 level of variable (App) "Situation 1. Preferred channel for getting notification" and 1 level of a variable (Everyday), "Situation 1. Preferred timeframe for getting notification" increases the chances of positive values of the variable "Situation 1. Intention to buy with notification" in 36.581 times (sending notifications about points expiration through the app and sending notification in a day before points expiration increase the likelihood that the buyer will have the purchase intention).

Based on the table 15, H 2.0: Channel through which notifications are received by a customer does not affect customer intention to make a purchase - was not confirmed; H 2.1: Channel through which notifications are received by a customer affect customer intention to make a purchase - was confirmed.

H 3.0: The time frame of sending notifications does not affect the customer intention to make a purchase - was confirmed, H 3.1: The time frame of sending notifications affect the customer intention to make a purchase - was nor confirmed.

However, due to the fact that the factors above do not have an impact on customer intention, it was decided to build the following logistic regression, which covers more factors that may affect the choice of respondents.

Table 16 Logistic regression

Situation 1. Intention to buy with notification	Odds Ratio	St.Err.	t-value	p-value	[95% Conf	Interval]	Sig
Situation 1. Preferred timeframe for getting notification	0.758	0.165	-1.27	0.203	0.495	1.162
Situation 1. Preferred channel for getting notification	0.678	0.107	-2.47	0.014	0.498	0.923	**
Situation 3. Discount in percentages sufficient for Intention to buy	0.595	0.137	-2.25	0.024	0.379	0.935	**
Situation 2. Discount in rubles sufficient for Intention to buy	0.798	0.141	-1.27	0.203	0.564	1.130
Age	0.851	0.094	-1.46	0.145	0.685	1.057
Attitude towards discounts	0.651	0.150	-1.87	0.062	0.415	1.022	*
Education	0.920	0.145	-0.53	0.595	0.676	1.252
Family status	0.814	0.124	-1.36	0.175	0.604	1.096
Financial situation	1.191	0.224	0.93	0.353	0.824	1.720
City size	1.171	0.187	0.99	0.324	0.856	1.603
Constant	942.254	1602.9	4.03	0.000	33.581	26438.809	***
Mean dependent var	0.815	SD dependent var	0.389
Pseudo r-squared	0.096	Number of obs	308.000
Chi-square	28.179	Prob > chi2	0.002
Akaike crit. (AIC)	288.877	Bayesian crit. (BIC)	329.908
*** p<0.01, ** p<0.05, * p<0.1

In the "Situation 1. Intention to buy with notification" variable, the value "0"corresponds to no intention to make a purchase after the notification, and value "1" means that there is an intention to make a purchase after the notification of reaching the threshold.

The variables "Situation 1. Preferred timeframe for getting notification" and "Situation 1. Preferred channel for getting notification" have multi-level encoding for selecting the most convenient notification channel for respondents (1. E-mail; 2. Sms; 3. Social media; 4. App) and the most convenient time interval (1. 1 day; 2. 2-3 days; 3. 1 week; 4.2 weeks) for which it is better to send an alert.

Multilevel variables "Situation 3. Discount in percentages sufficient for Intention to buy" (you can pay for: 30% of a purchase, 50% of a purchase, 100% of a purchase, "do Not spend points") and "Situation 2. Discount in rubles sufficient" (sufficient is: 1000 points (100 rubles), 2000 points (200 rubles), 3000 points (300 rubles), "do not go, as already purchased") show which part of the purchase in percentage and in rubles, respectively, will be a sufficient discount to make a purchase.

The following variable "Age" defines the age groups of respondents: 18-25, 26-30, 31-35, 36-40, 41-45, 46-50, 51-55, 56-60, 61-65, older than 65.

The following levels are encoded in the "education" variable: Incomplete secondary education, Secondary General education, Secondary special education, Incomplete higher education, Higher education, Two or more higher education, and the presence of a candidate/doctor of science degree.

The "Financial situation" variable includes: "There is not enough money even to buy food", "there is only enough Money to buy food", "there is enough Money to buy necessary food and clothing, larger purchases have to be postponed", "Buying most durable goods (refrigerator, TV) does not cause difficulties, but we can not buy a car", "We can afford to buy a car, but we can not buy an apartment", "there is enough Money to not deny ourselves anything at all".

The "Attitude toward discounts" variable shows how often respondents buy discounted products. Answer choice: "Very often, I try as much as possible to buy goods with a discount", "Often, if I come across products that I would like to buy, with a discount, then I will definitely buy it," "Sometimes, if I come across products, that I would like, at a discount I might buy it", "Never, I never buy discounted items".

The next variable "Family status" determines the respondents' family status. Answer options: "Married", "civil marriage", "Divorced", "Single", "Widower".

The last variable "City size" estimates the size of the population of the locality where the respondent lives. Answer choice: More1mln; 500.000 - 1 mln; 100.000 - 500.000; 50.000-100.000; less 50.000; Not sure.

The following conclusions can be drawn from table 16:

1) the model is adequate to the studied data: Prob > chi2 = 0.002;

2) not all b-coefficients of the model are statistically significant: sig < 0.001; only "Constant" has a p-value = 0.000, and "Situation 1. Preferred channel for getting notification" and "Situation 3. Discount in percentages sufficient for intention to buy" have a p-value equal to 0.014 and 0.024, respectively.

3) Log likelihood = -133.43852 lower than the model in table 15 (-142.8921). Pseudo r-squared = 0.096 is small enough, but already higher than the model presented in the table 15.

4) Changing the level of the variable "Situation 1. Preferred channel for getting notification" from 0 to 1 increases the chances of a positive value of the variable "Situation 1. Intention to buy with notification" by 610.589 times and changing the level of the variable "Situation 1. Preferred channel for getting notification" from 0 to 1 increases the chances of a positive value of the variable "Situation 3. Discount in percentages sufficient for intention to buy" in 560.641 (the Channel for sending an alert about reaching the threshold and receiving a discount of % of the purchase amount increases the probability that the buyer will have the intention to make a purchase).



Conclusion

In this section the discussion of the results obtained and their comparison with previous researches are presented. Also, there are developed recommendations on how companies which use loyalty programs with points expiration policy can manage their loyalty programs from the perspective of personalized marketing. In the end of the section limitations and future research directions are discussed.

Based on the results of the research, there can be concluded that a successful bonus loyalty program is aimed at a deep and personal understanding of customers. Namely, personalized marketing is the mechanism by which company can ensure control over the system of relationships with the company's customers. In the context of this study, personalized marketing involves notifications about points expiration. According to the results, receiving notifications about accumulated points and their terms of use increases the probability of LP customers' intention to make a purchase and spend bonus points by almost 5 times. Consequently, as indicated in Lemon & Verhoeff (2016) research, personalized marketing has a positive impact on companies using LP.

Diving deeper, only 57 (19%) people out of all the respondents will not make a purchase, regardless of whether the notification was sent or not. based on this, it can be concluded that for 57 respondents there may exist other factors that affect their intention to make a purchase (e. g. the value of remuneration, financial situation, etc.).

For those who would make a purchase after receiving a notification, additional factors that influenced their decision were studied. Thus, it was found that the amount of remuneration and the communication channel affect the buyer's intention to make a purchase.

It was found that the most convenient way to receive notifications for customers is receiving notifications using only one channel. Thus, companies should use only 1 channel to avoid possible negative reactions. This may also indicate that it is more convenient for people to use a single channel to organize notifications, so that they do not need to search for them through different channels when necessary. A similar conclusion was made in the article Bombaij & Dekimpe (2020), the authors found that customers, who use only one communication channel, have higher level of satisfaction from the loyalty program.

According to the conducted analysis, the most common and convenient channel of communication for buyers is sms notifications. This conclusion is also supported by previous research. Bruneau et al. (2018) revealed that sms are the most preferable way of communication among grocery stores consumers. Perhaps it can be caused by the fact that in today's world almost everyone has a mobile phone, but not always Internet enabled, and the option to not receive or ignore the alert about the availability of points and the available time of redemption is high (e. g. via email, social network or app) while SMS message will be received even without Internet.

Next important aspect, which was examined, is time frames in which consumer receive notifications about points expiration. Thanks to the regression analysis conducted, it was revealed that time frame is statistically insignificant in terms of customers intention to make a purchase. However, according to findings provided by Breugelmans et al., (2015), time frame can influence consumers intention to make a purchase. The findings of this paper may contradict to other research papers due to sample limitations. Thus, this aspect needs to be examined more detailed. It can be useful to examine the influence of different time frames of sending notifications on the basis of a certain company or a group of companies to analyze actual customer behavior in terms of different time frames of receiving notifications about points expiration. For example, an experiment may be conducted where different groups of customers receive notifications about reaching threshold in different time frames.

Additionally, the value of remuneration was studied as one of the factors that can affect the intention to make a purchase. According to the data obtained, for most respondents, 50% deduction from the purchase amount with bonus points is sufficient reward to influence the intention to make a purchase. Since the value of the reward makes sense for most respondents, this can be explained by the fact that most respondents are probably concerned about the fact that they may miss the chance to use accumulated points and get even a small benefit.

The obtained data is useful information for both company managers and marketers when developing and implementing loyalty programs with thresholds. First of all, the company need to determine the most convenient channel of communication with the participants of the loyalty program. Despite the fact that this research has shown that such a channel is an SMS notification, due to the specifics of a particular company, this channel may be different. Also, companies should not put pressure on customers using multiple communication channels, but use only one channel, so the buyer will feel only minor pressure and as a result, satisfaction from participating in the loyalty program may be higher. In order for the loyalty program to work effectively with the points expiration policy, companies must remind their customers that their points will be burned. Despite the fact that this study did not reveal the impact of different time periods between sending notifications about points burning and actual points burning on customers intention to make a purchase, in order to establish effective deadlines for sending notifications, companies are recommended to: a) keep records of the frequency of purchases made by loyalty program participants; b) create personalized questionnaires for loyalty program participants in order to collect customer information.

In this study, the results obtained have sampling errors, since the non-probability sampling method is used. Since there is a limited ability to access the geographical coverage of participants, there is a bias according to this factor.

Since the survey was aimed at the entire territory of the Russian Federation, however, the majority of respondents is from St. Petersburg are 154 respondents, which is 50% of all respondents. The remaining 50% is distributed among respondents from Moscow, 34 regions, and missing values. This bias may be due to the fact that the survey was sent out primarily among friends of the authors of the work who also live in St. Petersburg.

Sample bias was also found in the age category, which is most likely due to the fact that the survey was distributed online, in particular among peers, which led to a shift to the age category from 18 to 25 years (228 respondents, which is 74% of all respondents).

Another limitation identified is sample bias based on gender, namely the number of respondents belonging to the female gender (228 respondents, representing 74% of all respondents), which is also due to the fact that the survey was distributed online, in particular among friends and according to the Federal state statistics service - the number of women in Russia by 1 January 2019 is 78684293 man, and men 68096427 person. Women outnumber men by 4% (approximately 11 million people) (Population of the Russian Federation by gender and age, 2019).

Due to the fact that the survey was distributed online, there is a sample bias of respondents, since the opinions of people who are offline are not taken into account.

For future research, in order to increase the reliability of the results obtained, it is recommended to expand the sample of respondents by expanding the geographical coverage and conduct an experiment rather than a survey. In this study, the sample is biased by gender towards the female gender. To eliminate this bias, other data collection tools can be used (for example, not a survey, but an experiment), and if a survey is used, it is possible to use other channels for distributing the survey in order to keep balance between female and male respondents. Also, in the future, researchers can focus on the use of a frequency rewards loyalty program in a certain area of its operation (e. g. grocery stores, transport, etc.), and identify patterns of buyers' behavior in terms of points redemption relative to this area.

Comparing the results of this work with subsequent researches will also help to analyze in more depth the impact of the main aspects on the positive dynamics of companies using loyalty programs.

Reference list

1. Banik, S., & Gao, Y. (2020). Status demotion in loyalty programmes: The role of perceived unfairness. The Service Industries Journal, 40(3-4), 315-336.

2. Bijmolt, T. H. A. (2010). Loyalty Programs: Generalizations on Their Adoption, Effectiveness and Design. Foundations and Trends® in Marketing, 5(4), 197-258.

3. Bijmolt, T. H. A., & Verhoef, P. C. (2017). Loyalty Programs: Current Insights, Research Challenges, and Emerging Trends. In B. Wierenga & R. van der Lans (Eds.), Handbook of Marketing Decision Models (Vol. 254, pp. 143-165). Springer International Publishing.

4. Bruneau, V., Swaen, V., & Zidda, P. (2018). Are loyalty program members really engaged? Measuring customer engagement with loyalty programs. Journal of Business Research, 91, 144-158.