Marketing automation is a seemingly universal ambition of mobile advertisers — in some cases, bordering on obsession. For companies spending millions of dollars in advertising media per month, automation provides for the ability to launch and scale new products at will, as well as the ability to increase or decrease spend flexibly without needing to change the marketing team’s headcount. For advertisers with smaller budgets, user acquisition automation unlocks access to growth by sidestepping the “chicken and egg” problem in mobile user acquisition wherein small companies with unproven products can’t attract accomplished user acquisition talent. For both groups of advertisers, automation is seen as a sort of panacea: a common cure to the specific challenges faced at both low and high scale of monthly spend on direct response digital marketing.
Like almost everything related to automation, artificial intelligence, and machine leaning, user acquisition automation as a concept drafts some considerable distance behind the enthusiasm (hype) from investors and executives about its applicability as a commercial tool. Firstly, user acquisition automation already largely exists on mobile — in the form of Facebook and UAC campaigns. Facebook and Google automate budget distribution, general audience targeting, and ad creative exposure. The ascent of algorithmic campaign management tools from Facebook and Google essentially handed fully-automated marketing machinery to all mobile advertisers. So any discussion of automation as a proprietary tool necessarily implies a tool that interfaces with Google’s and Facebook’s — given that Facebook and Google represent 60+% of ad spend for most major mobile advertisers — which somewhat deflates the scope of any such product.
That aside, the core functional user acquisition work that a team might undertake to automate falls into the following buckets:
- Ad creative production: ad creative is one of the primary inputs to user acquisition, and the volumes of creative needed for rigorous experimentation are so vast that automating creative production can serve as a real competitive advantage for advertisers;
- Audience testing and definition: audiences (eg. custom audiences, lookalike audiences, etc.) are incredibly important factors in efficient advertising, and testing them is time consuming and burdensome, especially as pairings with other campaign settings, such as audiences and bid / budget levels;
- Campaign settings experimentation: bids, budgets, campaign strategy, etc.;
- ROAS projection and reporting: ROAS estimation is built into most advertisers’ existing analytics suites, but automating it in such a way as to calculate and publish relevant values on a cadence that is actionable can reduce decision-making time and prevent advertising loss. Additionally, ROAS timeline targets can change over time with spend levels and budget distribution, and automating that can reduce a substantial amount of analysis work.
There’s a fifth bucket, which is the connective tissue which unifies each of the above four buckets and replaces the human “If this, then that” decision making and execution workflow. For instance, if a performant audience is found, the connective tissue might deploy that against a new campaign versus just signaling the audience’s viability to the team.
Like the “levels of driving automation” classifier that exists for categorizing autonomous driving systems, I think it’s valuable to define a “levels of marketing automation” rubric for marketing systems. Having seen marketing automation platforms that fall across a broad spectrum of functionality, I think the general progression towards a completely automated marketing AI happens in five different stages:
Level 0: No automation
At this level, all creative production, audience testing and definition, campaign experimentation and optimization, and ROAS projection and reporting is done manually. Humans manage every aspect of marketing and serve as the connective tissue between decisions. No automation exists and everything is done by hand.
Level 1: Tools replace tedious manual work
At this level, the marketing team deploys a set of scripts and tools to automate the more tedious manual work. Ad creatives are templated so that variants of concepts can be produced quickly; audiences are pulled via a script from the advertiser’s data warehouse and deployed via API to various advertising platforms (eg. a script pulls a list of the top 1000 most engaged users every week and submits that via Facebook API as a custom audience); campaign settings are circulated automatically (eg. the marketing team receives a weekly email with suggested, manually-calculated bid levels for different Facebook campaigns based on geography, platform, campaign strategy, etc.); ROAS models and timelines are defined and assessed manually but they are made available in a central system to be used in forecasting in an automated way.
At this level, most of the substantive work undertaken by the team is still done manually, but much of the tedious button-pushing is replaced by scripts that move datasets around or do simple calculations and make those available to the team. This level requires the same general level of marketing headcount as the previous level, but the tools and scripts make each of those people more productive.
Level 2: Marketing workflow automation
It is at this level that impactful, high-value work begins to be automated, replacing the work of humans. At level 2, much of the analysis that was formerly undertaken by marketing team members is automated, leaving only very specific tasks to be done manually.
Creative concepts are informed in an automated way by past performance; audience definitions are likewise created automatically and deployed without any assistance from humans. Bid and budget decisions are made automatically, meaning that marketing team members merely receive recommendations from the system. ROAS timelines and the underlying monetization model is calculated in an automated way.
Level 3: Marketing strategy automation
At this level, the marketing team’s focus begins to shift from managing campaigns to managing the system that manages campaigns; the marketing team’s composition changes from media buyers to analysts and engineers that maintain and continuously improve the automation system.
At level 3, ad creative concepts are generated and produced automatically. Audiences are defined, sourced, and deployed automatically, with audience pairings expanded to every possible campaign setting (eg. audiences are tested against creatives, campaign strategies, geographies, etc.). And ROAS timelines and models are not only derived automatically, but they trigger changes to campaign settings automatically. Combined with bid and budget decisions being made and deployed automatically, at level 3, a team member no longer needs to make direct adjustments to campaigns.
Level 4: Full marketing automation
Full marketing automation includes all of the features of level 3, except that the connective tissue between each of the four functional buckets is complete in such a way that all decisions and reactions are implemented automatically: campaigns are automatically created and tuned, creative concepts are automatically defined, produced, and deployed to campaigns; ROAS models and timelines are automatically calculated and used to adjust campaign settings, etc. At this level, theoretically, no marketing team is needed beyond the engineers and analysts that maintain the automation framework.
Where are most advertisers?
Having seen many permutations of marketing automation, I believe that most advertisers investing the time and resources into building out an automated user acquisition platform are at Level 1, with a small minority at Level 2.
I have seen a number of advertisers building towards Level 3, but there are some meaningful hurdles to clear in reaching Level 3:
- Total creative automation is very difficult, especially for video creatives. Contextual image classification analysis is possible at scale, but without a human-made template, the videos created using just imagine recognition and performance pairings tend to not perform well themselves. Identifying pieces of a story that correlate to strong performance is different from constructing a performant story;
- ROAS timeline optimization and monetization modeling is very hard to do in an unsupervised way. Most LTV and ROAS modeling is done by iteratively, manually tinkering with curve parameters and milestones based on deep knowledge of the product. Building a reactive ROAS timeline and model based on spend levels and budget distribution is one step ahead of media mix modeling.
But the value of even Level 1 marketing automation shouldn’t be discounted: when the tedious, manual work required of most marketing managers is removed, those people are given the opportunity to be much more analytical and strategic. Level 1 marketing automation is a laudable goal, and the robustness of Facebook’s and Google’s APIs makes it fairly straightforward to implement.
In general, automation is such a powerfully productive tool that all advertisers should strive to at least some level of it — what that level should be depends on the advertiser’s ability to scale a marketing engineering organization and the absolute value of that automation efficiency.