Having worked across multiple teams in data and having exchanged experiences, I have come across some recurring themes that lead to companies being unhappy with their ROI on data. This is not about the problems of data teams but about the problems companies have regarding data usage.
1. Pain Point: No ROI reporting
Due to the diversity of marketing channels, as well as attribution tracking, it is difficult to always have a complete view of the spend and revenue of each marketing channel while maintaining high granularity in the data.
Maintaining granularity means to have your data in a still non-aggregated way, such as keyword level for SEA campaigns, billboard address level for out of home, etc.
Why do we need ROI reporting?
In an ideal world, deciding what we are willing to spend on advertising is determined by our return. You look at what money you make per sale, your margin. And that’s what you can afford to spend to make a sale. This means that anything up to your margin is an acceptable cost, as anything that you bring in at this value will make you some ( >0 ) money. This also means that if you spend more than what your margin is, you lose money. This part of auction theory.
So the key to understanding what we can afford to spend on marketing is understanding our margin.
For example, if our margin per sale is 20 euro, it would be silly to only spend 3 euro per sale in acquisition cost. Sure, get those customers that you can for 3 euro, but also get the ones that cost you 19.99, because some money is better than no money.
Enable your data team to do complete ROI reporting
In order to have proper attribution, we must design our data collection correctly – how you collect and attribute sales for online campaigns is different to how you collect data for emailing, post campaigns, flyers, or billboards.
Additionally, accurate records must be kept of cost data, to be able to understand what each sale is costing us in terms of cost of acquisition.
The solution to this is straightforward: Involve the data team from the start to have accurate tracking and attribution. Also support them in their data collection with good tracking solutions. Once the attribution is clear, cost is only a matter of importing data (integrations) you already have next to your other data.
Where possible, it’s important to automate data centralization through API integrations. This will keep your latency, time consumption, and error rates low, and your marketing employee satisfaction high.
2. Pain Point: No direction in the data team
Many data teams do not have initiative or direction. As a result, they are merely reacting to requests but never producing anything of their own initiative. There can be a couple of reasons for this.
Lack of empowerment
The first reason causing this data pain point is a lack of empowerment. The data team is treated as a service that supports other teams and is not given the chance to participate in key project planning.
This causes the data team to always be outside of the loop and late to the party. “Can you check the performance of this campaign you did?” – “Uhm, no, you did not track anything, you should have said something BEFORE starting!”
Also due to lack of empowerment and being kept outside of the loop, the data team becomes disconnected from the business. This makes them unable to spot opportunities for improvement.
A data team can only act on information they have, so bringing data in early on key projects is very relevant to the team’s potential to add value.
Lack of experience
The second reason causing this data pain point is a lack of experience. Let’s not put it on creativity, logical intelligence, etc. Most of these things are learned skills. The naked truth is that if you do not know of a technology that solves a problem, it is unlikely that you will manage to invent this technology from scratch. An experienced team has worked in many environments, on a variety of problems, and has an easy time re-using past ideas or learnings.
Giving your data team direction
If the lack of direction is due to lack of empowerment, start bringing your data team (representative) into your ‘heads meeting’, into key project planning at the conception phase, and into your budget planning.
Chances are that the team will have valuable input on all these things, and it will keep the team focused on optimizing the business, not on optimizing loading times into the DWH. That billboard campaign you wanna run? Data team probably has some ability to extrapolate location based ROI, design tracking, and offer follow up improvement hypotheses for future campaigns.
Hire expertise & invest in learning
If the lack of direction is due to the lack of experience, this is easy to solve. In the short term, bring in someone who has the experience to assess the situation and offer some actionables. You can find these people in the form of team managers in companies that attract talent, or in the form of freelancers.
In the long term, encourage your team to allocate time to learning. Encourage them to meet other BI people at meetups by counting it towards their work time. Create a learning/development plan for them and track their progress against it. Always pay for books and courses, and conferences if you can afford it.
3. Pain Point: Inconsistent metrics
Between home brewed department-only metrics and missing definitions, different numbers will tell different stories about the same events. This leads to a lot of back and forth and frustration between teams as they try to align, and can even lead to incorrect decisions being made. Measuring different things (customers vs orders) or at different times (ex: order date vs delivery date) is a typical example.
A good way to go about defining KPIs is to map the team’s work process and identify the measurable steps across the process. Decide what you want to optimize, and create the right metrics for it. Make sure they are applied consistently throughout the whole team.
If a KPI cannot be acted upon to change the performance of a process, then we should remove it, to avoid any confusion or dilution of information.
4. Pain Point: Show data in the right place
A typical pain is that many employees end up having to do a lot of manual work to be able to use the data available to them. Log in here, copy this, open that, paste here, sum up this…
In the spirit of automation and encouraging data consumption, we need to make this data available in the right place.
For data-driven decision making, the information needed has to be easily available to the decision maker.
What does this mean?
It means that if a salesman wants to call his least spending clients, this spend data should be available in his CRM, integrated into his workflow.
If a manager wants a quick peek at top level performance, it should be in an easily accessible place such as an email or dashboard.
If a customer support agent wants to know how much time he should allocate to this client, he needs to easily see his call/email queue size, to be able to prioritize. A convenient way to do this is a screen on the wall.
If a marketer needs to do budgeting, the performance and ROI information should be available to them in the same place where they calculate their budgets.
If we need to tell our operations to restock a product, the data should be accessible within their workflow of ordering.
You get the idea. Dashboards are not everything. The real value of data is when it becomes directly usable for your team.
5. Pain Point: The right time
A lot of companies focus on having real-time data available. But does this really make sense for you?
Can you change your actions in a matter of hours from observing an event? Typically, once an anomaly is noticed, it takes quite a while to analyse the cause, come up with a solution, and put it into practice. Unless you are delivering real-time relevant services like recommendations, bidding, perishable merchandise, stock-keeping, etc, it is unlikely anything you can do in a few hours would have any significant impact.
Averting disasters like “What if checkout doesn’t work and nobody notices for a day” is what monitoring and alerting is for. This is not the task of large data analytical pipelines.
The complexity of building data streaming pipelines and their associated maintenance cost are significantly higher than in a batch-jobs scenario. The latter are the ones where data is refreshed at regular intervals, like hourly, daily etc. In fact, in the wild, many near-real-time pipelines are just batch jobs that start up as soon as they have finished, to give the lowest possible latency.
Such, the question is reduced to how quickly you can react to data, which is company specific. If you cannot react in real time, don’t focus on having real-time data.
What typical pain points do you experience, and what works best for you to resolve them?