Data and AI can predict success
Nearly half of promotions lose money. Angus McLean and Randy Kohl look at some of the most effective promotional tactics for defining and measuring promotion effectiveness using deeper promotion analysis
Year after year, companies go all-in on organising and marketing promotions, spending billions to increase customer affinity, and ultimately drive revenue. But when it comes to proving their worth, it’s an unfortunate reality that many of these promotions fail to do so. Some statistics show that as little as 16% of promotions are profitable.
While many brands invest heavily in these programmes, the reality is that nearly half of all promotions lose money for a variety of reasons – poor planning, poor product selection, poor marketing, or inadvertently cannibalising other products, to name a handful.
Still, about 93% of shoppers use a coupon code or discount throughout the year. That means that, when done well, promotions are still among the most potent tools in a brand’s arsenal. But without careful analysis, it is easy to under or overestimate the benefits and return on investment (ROI) of a promotion.
Currently, most businesses fail to optimise their promotions. This is often because, rather than being methodical, they regularly run various discounts and campaigns and, by doing so, devalue their promotions right out of the gate.
Many businesses believe that they lack the tools to thoroughly analyse the data or plan an insights-based promotion. In reality, most are already sitting on plenty of data that could inform methodical and successful promotion campaigns, thanks to apps that are part of their existing infrastructure.
From basic Google Analytics insights to more in-depth insights that can be leveraged via apps built into commerce platforms such as Adobe Commerce, the simplest of starting points is to mine existing data for insights. And brands should not underestimate the value of learning from past promotions to inform strategies for improving future efforts.
DEFINING PROMOTION OBJECTIVES
Of course, there’s more to planning a promotion than extracting numbers, which without context can mean very little. Instead, when looking at all types of data, it’s important to first define an objective for any given promotion. Some of the fundamental questions and considerations in formulating that objective are:
What is the ROI of this promotion?
This is the fundamental question companies must answer to ensure they’re offsetting the associated costs.
What type of promotions will maximise traffic, engagement and profit?
It’s important to understand whether your customers will best respond to buy-one-get-one offers, total purchase percentage discounts, free shipping, or product giveaways, for example.
What is the sweet spot in terms of the amount of the discount?
This is a critical question to answer when determining what a discount amount does to profit margins. Dissecting the data can help uncover the biggest opportunities and trade-offs.
What is the proper timing to maximise the promotion’s impact?
Consider whether you want to run a short-term 'blowout' type sale or a longer-term offer. Ultimately, this ties back to defining how much of a discount you can afford to offer and still remain profitable.
How can you predict how much demand and revenue a promo will generate?
Forecasting is another area where it’s critical to mine the data you have at your disposal. Machine learning can capture the impact of recurring patterns, internal business decisions, and external factors to generate accurate, granular, and automated short- and long-term demand forecasts.
How can promotions be best tailored to the most appropriate audiences?
Some customers won’t purchase, regardless of promotion, while others will engage only with certain types of promotions. That’s why it’s important to determine what type of offers spark action from which customer segments.
While it can be tempting to look only at total revenue when analysing your promotion, it’s not the most accurate indicator of success. High revenue on its own can be exhilarating for a business owner, but a deeper look often reveals that the numbers don’t tell the full story.
FRAMEWORK FOR AI-ENHANCED PRMOTION STRATEGY
Promotion effectiveness modelling
If you can build a system in which promotion planners can input different attributes to predict and estimate how well a promotion will perform, a true promotion effectiveness model can be created. This is also known as forecasting. By employing AI into your system, you can use your inputs, such as parameters, historical data and observations, to forecast the success of a given promotion.
Better yet, we would like to get to a point where we input what KPIs or objectives the system should maximise, and it will model many different scenarios to suggest promotion strategies that best meet those objectives (like maximising traffic or maximising profit).
Targeting (or granular forecasting)
Recent advances in big data empower us to take promotion forecasting to an entirely new level of accuracy by targeting promotions to specific customers that need them most rather than working in broad strokes. Why give away savings to your whole customer base when only a select few would greatly benefit from a slight incentivisation?
Ideally, we could identify which customers are most likely to contribute the most profit to our organisation after a proportionally small discount. To do this effectively, we need to be able to predict how likely a customer is to purchase a product or engage with a particular promotion. What is the propensity for customer X to engage with promotion Y or promotion Z? This question can be answered (statistically) with ‘propensity modelling’, which will route specific promotions to specific customers, also known as targeting.
The goal of propensity modelling is to find consumers who have a relatively high probability of behaving in a certain way or committing a certain action after being delivered a call to action, usually in the form of an exclusive offer or customised banner. Obviously, there are many customer actions that can be forecasted using a propensity modelling approach:
- propensity to try a new product
- propensity for category expansion
- propensity to buy more
- propensity to churn
- propensity to engage
Typically, we want to use propensity modelling to predict the probability that a customer will engage with a particular promotion delivered to them. Propensity modelling can also be useful to identify customers who are likely to leave but may be enticed to stay when offered an incentive (retention/churn-mitigation).
Customers respond differently to messaging and promotions. Some are likely to engage and purchase again, while others are going to throw it in the spam folder.
Knowing who those customers are often spells the difference between retention and churn. Anyone familiar with targeting strategy is likely already well acquainted with the four quadrants of customer responsiveness:
- persuadables – likely to respond positively to a promotional offer.
- sure thing – likely to purchase with or without promotional offer.
- lost causes – won’t accept either way
- do-not-disturbs – likely to respond negatively if targeted.
As most businesses know, there is no silver bullet for promotion success. But by beginning to ask the right questions and conduct the proper analysis, brands can better position themselves to plan future promotions that will yield revenue, drive customer engagement and retention, and ultimately land them on the side of profitability and progress.
It all comes down to getting at the right data, analysing it properly, and uncovering trends, anomalies and opportunities to drive future success. From there you can develop superior strategies that are repeatable and provide the uplift to deliver true ROI.
THE DO’S AND DON’T’S OF PROMOTION SUCCESS
As much as we can employ technology, the human element is still key in guiding effective processes along. Ultimately, you are still responsible for defining the inputs that will deliver the information you need. Here are some important things to keep in mind and to ensure you’re getting the most out of the information you’re analysing and using as future guidance:
Work from the top down and follow the money. Prioritise ruthlessly
Data science projects drive ROI through superior decision making. Don’t spend time on insignificant parts of the business; prioritise looking for small wins in highly impactful areas.
Value process, structure and simplicity
Clarity in approach breeds clear and consistent analysis results. Aimlessly investigating massive datasets is likely to create more questions than answers. Start with the end in mind. What questions
does the business have? What data speaks to those questions? What do hypotheses say about expected results? What does the data say and how does that compare to hypotheses?
Do your data due diligence
Don’t assume data cleanliness. Check thoroughly for outliers, missing data, duplications and unusual correlations. At the end of the day, if you don’t trust the dataset, you can’t trust the results. Starting with a clean dataset will ensure a strong foundation for accurate analysis and results.
Don’t undervalue customer retention
It is significantly easier to convert an existing customer to a repeat purchaser than it is to entice completely new customers. Retention is worth its weight in gold; know your customers and keep them coming back.
Don’t get stuck in spreadsheets
Initially, it might be tempting to run analysis and deliver results in spreadsheets, but this doesn’t scale well and will limit your sophistication going forward.
Don’t blindly use AI
Shortcuts and skipping steps are a sure-fi re way to run into major issues and oversights down the road making you wish you had paid more attention earlier. Data science is a highly precise process, and the devil is in the details.
06 December 2022
More in Technology & data
AI is serving up a holiday feast
AI can tell us what to eat and how to prepare it this holiday season. Jeff Malmad of WPP’s Mindshare explains
AI regulation: it’s about finding the right balance
AI regulation is a government, industry, regulator, company and private individual issue – and it must be addressed collaboratively
Making reach more effective in building brands
Sudarshan Saha and Dibendu Moulick of WPP's EssenceMediacom in Vietnam explain how a simple framework – for use in data-fragmented markets – can help with investment decisions by combining views and reach