Prioritization is the apex artifact bridging strategy and execution.
I’ve used prioritization models throughout my career to identify the most propitious experiments, customer acquisition channels, and product / marketing investments to make.
Here’s an actual backlog I prioritized using PXL for a specific page experience.
In theory, prioritization is a function, or even an algorithm, that takes in a variety of inputs (including constraints, historical data, qualitative insights, stakeholder wishlists, and subjective experience) and delivers a clear and simple output: an ordered list of actions that maximizes return and minimizes investment.
In practice, however, prioritization models are often used as cursory tools to get your boss or client off your back.
>“How are you prioritizing initiatives?”
>“We have this scientific formula that stack ranks initiatives based on an expected value computation of impact, ease, and business relevance.”
>“Oh very nice – carry on, then!”
But do you really live and die by three dimensions, summed up and then divided by three, on your cloud-hosted collaborative Google Sheet? Do you really equally weigh keyword difficulty, keyword search volume, and business relevance, and then write content based on their composite score?
If you do, congratulations – you’re Lawful Neutral. But this rigidity will probably result in lower than optimal returns. Why?
- Because not all companies should weigh the same dimensions equally.
- Because different types of initiatives have different goals. Some need to be sequenced so other things can be accomplished or amplified. Templated models don’t allow us to compare apples to oranges.
- Because prioritization includes both filtering (omitting certain themes and segmenting them) and sorting (stack ranking similar initiatives based on quantitative scoring).
- Because sometimes you uncover a bit of data, a black swan if we’re being hyperbolic, and you should put it at the top of your backlog even if it doesn’t fit your model.
- Because the Chaotic Good win the organic growth marketing game. I hate to break it to you, but it’s true.
I’ll be honest with you: there’s no plug-and-play template at the end of this essay that will solve all of your prioritization problems.
I’ve been working on this problem for years, and I’ve always run into the same roadblock: templated models are underfit. They lose utility almost immediately with real world data and decision making.
So I’ve surrendered to this, humbled myself to the complex organic marketing universe that refuses to be reduced to three static dimensions on a spreadsheet.
And if you, too, can accept that no such model exists, then we can move forward and elevate the concept of prioritization above a static artifact, and into a dynamic ritual that actually improves decision making and program outcomes.
You Need to Understand Expected Value
Every single model you’ve ever used or heard of is trying to approximate what is known as Expected Value.
At an extremely high level, expected value is a calculation that predicts the average value you can expect to get if you repeat an action many times.
Basically, it takes into account the upside, the probability of the upside, and the cost of taking an action (as well as, sometimes, the downside potential and the probability of the downside).
It’s a simple concept (in a casino) and a very complex one in the real world.
Basically, if you decide to pull a slot machine arm, you can calculate the expected value if you know:
- How much it costs to play
- What the range of rewards is
- What the probability of getting a reward is
Technically, you could complicate the formula if there is also a range of negative outcomes (i.e., if you short a stock, you can win X, but you can also lose Y).
Where this gets harder is, well, the real world, where cost includes the freelancer you pay, but also the subjective difficulty of publishing a given article (is it 1k words or 5k, technical or breezy, images or no?), as well as the unpredictable difficulty of ranking and achieving a reward.
Then there’s the opaque range of rewards. Sure, you can model out MSV, but do you know that you can also rank for 100s of other variants of that keyword, effectively 10Xing your estimate?
Okay, so impact and ease are both pretty tricky to nail down, but we can be pragmatic – we don’t need 100% accurate inputs to get directional utility.
But then we get into interaction effects – in SEO, it’s not 1 page that makes a program, but rather the cumulative benefit of 100s or 1000s of pages. Due to many factors outside the focus of this essay (topical authority, internal linking, backlinks, etc.), choosing 50 pages with maximum impact and ease across 50 different topic verticals may have a much lower composite outcome than choosing 50 pages with lower volume and higher keyword difficulty, but in the same topic cluster that is related to your core product.
We could spend all day discussing the limitations of expected value calculations in SEO. For now, let’s agree that there is some utility, but expected value sorting alone doesn’t give you much information.
I’ll return later in this essay to show you where you can use these calculations, but for now, let’s move onto some more useful frameworks.
Prioritization Should be Hierarchical
First order of business: there are multiple, hierarchical layers of prioritization.
Every business is unique, its own country. Some are complex multinational companies with hundreds of products, features, and personas. Some are startups with 1-2 personas and a very narrow search TAM.
So you probably shouldn’t just dump ALL POSSIBLE keywords into a single spreadsheet and sort them by the same formula.
Let’s use a concept from machine learning here – hierarchical clustering.
Before we sort keywords and stack rank their order, we first need to prioritize higher order dimensions. This depends on this business but could include:
- Business Unit or product line
- Personas or audience targeting
- Product or feature
- Topic cluster or theme
- Geographic targeting
As a simple example, we can imagine a MarTech company with 3 core product lines, one for marketers, one for sales, and one for customer support professionals.
Within each of those product lines, there are several distinct products and features to prioritize. For instance, their product suite for marketers includes landing pages, email marketing, marketing automation, and lead capture.
And within each of those product features, there are distinct keyword or topic clusters. For example, we could break down email marketing into several smaller groups:
So our hierarchical prioritization may look something like this:
So every business is unique. We can’t bypass the strategic layers of prioritization, which is why it’s so frustrating when some SEO professionals just sort a big ol’ list of keywords.
Some of these dimensions have…unscientific constraints. Team A has no budget, team B is flush with money. The CEO adamantly wants to grow customer segment C, even though LTV and remaining TAM in customer segment B is most opportune. Deal with it. This is where SEO strategists need to become, or at least communicate with, business strategists.
(Side note: we offer an “Organic Growth Blueprint” service we’ve used to help Fortune 500 companies identify their growth opportunities, index their current status on various dimensions and products, and build a roadmap towards leadership status).
Filtering Versus Sorting
There are two core types of prioritization: filtering and sorting.
Imagine I have an open afternoon and want to decide how to spend my time. It’s not until I filter out activities I don’t want to do, leaving what I would be interested in remaining, that I can sort among similar options.
So I don’t want to play sports, drink alcohol, or work, and it’s not evening yet, so there are no concerts I could attend. But I would be interested in reading books or playing board games. That’s a filtering exercise. Then, I can sort which books I’d most be interested in or which board games I’d most like to play.
Most of the higher order priorities are chosen by way of filtering. And most lower order priorities, say keyword selections or technical SEO fixes, are chosen by way of sorting.
Before you sort, you need to filter. Filtering and segmenting lets you compare apples to apples. It wouldn’t make much sense to compare sports (pickleball, skiing, tennis) to books (Catcher in the Rye, Infinite Jest, Good Strategy Bad Strategy) to board games (Cards Against Humanity, Settlers of Catan). So first we filter.
SEO research can help with filtering, but you have to understand much of it goes beyond and above SEO metrics. Where it can help is opportunity sizing and cost estimates.
You can help business stakeholders understand that, yes, you may want to drive adoption of product A, but product B represents a 20X total addressable search market, and you’ve already got strong topical authority in this area. So, perhaps consider allocating a significant percentage of your investment in product B, and we can stake maybe 10-20% of our portfolio in product A to start building out that growth horizon.
This is the juncture where SEO strategists become business strategists, by the way. Time to dust off that old growth model.
Summary: map out the hierarchy of categories that matter to your business and filter through top priorities. Use SEO and market data to assist in opportunity sizing, and agree upon top business priorities, product and persona priorities, geographic priorities, before sorting keywords based on formula.
Use Dynamic Variables to Customize Sorting Models
Alright, let’s assume we’ve gotten to the point where we agreed upon which persona, product, and business unit we’re investing in.
We’ve also done our cursory market and competitive research, and we have a handy little SWOT analysis that tells us our comparative advantages, weaknesses, and opportunities. This helps us allocate our content portfolio appropriately among emerging topics, product-led content, buzzworthy link bait, and programmatic template pages.
Now it’s time to break out the spreadsheet, with one small change: add custom weights to the variables.
The variables we typically use to stack rank content topics are impact (search volume), ease (composite score – topic complexity, SERP competition by domain rating, SERP competition by quality), product relevance, and conversion intent / customer journey stage.
We can then apply a weight to each of these from 1-10 that gives it a lower or higher effect in the overall composite score. These are chosen based on our strategic direction and the goals of the client.
For example, a leading company with a complex sales cycle may want to saturate their topic surface area without worrying too much about direct conversions. So we’ll heavily weigh impact and product relevance while deprioritizing ease and conversion intent.
For an early stage startup, we may index heavily on ease and product relevance with medium weight on conversion intent and low weight on search volume. Then we’ll complement this with linkable assets to drive authority and hit “escape velocity.”
This framework requires a lot more critical thinking than your typical prioritization model, which is a good thing. We’ve spent too many years as an industry using the same keyword research tools, content brief tools, and simplistic sorting models, which has resulted in a sea of sameness and impenetrable competition.
Applying basic logic gives you a fighting chance to drive unique value and real ROI through SEO.
Beyond Apples and Oranges: Portfolio Thinking
Models are only as good as their inputs, and sometimes you don’t have access to predictive historical data.
For instance, linkable assets / passive link magnets.
Sure, you can reverse engineer content that has already been linked to a lot, and you can even assume that higher search volume may equate to higher link building potential (because more potential searchers will see it). This is how most people do “passive link building.”
But that doesn’t tell the full story, because most historical links don’t always equate to future links, and reverse engineering a bunch of “[industry] statistics” blog posts and “[keyword] calculator” interactive tools (which are always the most linked) isn’t the bedrock of a forward thinking strategy.
So what do you do?
Change the rules.
Only apply your model to pages and actions you can actually predict expected value with and be intellectually humble (and courageous) enough to invest in things with unpredictable upside.
If you, for example, find through customer research that a TON of people want a certain free interactive quiz or calculator, and there’s no search volume; or if you have amazing 1st party product data that you think would get picked up by real journalists at top publications….
Just f*cking create it, regardless of what your model says.
Because your model would and should never account for the potential that a piece of buzzworthy content could garner 1000 referring domains. This is why the “barbell strategy’ concept is so important. Some percentage of your portfolio should, in fact, be incompatible with standardized models based on normal distribution outcomes.
Live a little, eh?
Okay fine, here’s a spreadsheet template
I made you a template.
It’s based on a scoring rubric Sam Lund created for hiring in our editorial team.
You’ll still have to customize it. For example, maybe you want different variables. We typically use Impact, Ease, Product / Business Relevance, Conversion Intent.
You’ll also have to come up with a scaling system to fit your scores into a 1-10 scale. I use percentiles and deviations from the mean for impact (e.g. take all keywords and their volumes, run some summary statistics, and then build a 1-10 scoring system based on which percentile they fall into).
Then choose how to weigh each variable.
As in, maybe ease is your most important value. You can publish like 100 articles per month with AI, so you’re doing a volume play. Very smart. We’ll be talking to you in a few months to a year when you need to fix that mess.
Snark aside, you’d just choose “10” for ease and lower the weight of the other variables:
Everybody has a plan until they get punched in the mouth
While I do think the above model is more useful than the industry standard, the main point of this essay is this: prioritization can’t actually be boiled down to a formula.
You and I can both easily think of cases that fall outside prioritization models.
Imagine you’ve got a piece at the top of your model. It’s delectable – 10,000 monthly searches, no difficulty or competition, pretty strong conversion intent. Very likely to have a good payback.
But then an opportunity comes up. Your coworker is friends with Tim Ferriss, and if your team can spend a little extra time (and effort) mining some first party data your product collects on, I don’t know, cortisol levels during extended fasting protocols, then there’s a good chance that your company will be featured on the Tim Ferriss podcast as well as syndicated among his collection of friends and fellow podcasters.
Well, you don’t have the search volume to attach to that opportunity. And it’s a higher effort content piece. But it seems like you should just throw the model aside for a second and take the opportunity, no?
Maybe, analogously, you had an itemized list of traits you want in an ideal romantic partner, but then you meet someone who doesn’t match many of them, particularly the superficial preferences (you like The Beatles, they like Blink 182, you like reading, they like adventure sports). What are you going to do, not fall in love because they didn’t match your little model?
So I’m advocating for critical thinking over templates, rituals over artifacts, and a bit of common sense to adjudicate the numbers your very simple spreadsheet model spits out.
Want more insights like this? Subscribe to our Field Notes.
