The Complete Guide to Product Qualified Leads

Get the in-depth framework — plus stories from leaders at companies like DigitalOcean, Moz, AdEspresso, Typeform, and more — to help you convert more of your product qualified leads into happy, paying customers.

Start Reading

Brought to You By

What's Inside

Want to know the minute a new chapter is published? Sign up now to get every chapter delivered straight to your inbox!

Sign Up Now

Chapter 5: Identify Your "Aha!" Moments

Now you’ve established where useful data might be, and where it needs to be sent, you need to work out what data might be most interesting for triggering personalized messages and cueing in your sales team.

What is the “aha!” moment?

What is the giveaway sign of a big spender?

When is the *perfect* moment to send an email or tag in your sales team?

With simple segmentation, like dividing by leads and customers, or dividing leads sourced through different campaigns, gut instinct and guessing is good enough to get started. Lifecycle stages are common across many companies, too, so it makes sense to generalize.

But with product usage data, you need to dig into YOUR data, and identify in YOUR data what the triggers are, which you can use to cue and clue in your sales team.

Gut instinct is NOT good enough for triggering personalized messages.

Neither is copying what worked for someone else. You need to find the specific triggers to your business - the Product-Market-Aha! fit!

There are three common techniques to this:

1. Raw product usage (quick start, easy. Less than 15 minutes). Guess at a trigger.

2. Quick regression analysis (identify more and better opportunities. An hour or less). Remember, specificity allows you to hone in your content on what matters and convert more. Find a handful of triggers.

3. Data science and artificial intelligence. Find many triggers, and deliver precise content (ongoing, five-figure spend)

Since these increase in complexity and potential return, start with the simplest. The basis of the more advanced techniques comes from the simpler techniques anyway. Start with a guess.

As your product qualified leads project continues, you can move onto progressively more complex analysis (and also earn the buy-in for the resources you need to develop your campaign further).

First, seek raw product usage

By finding the patterns in users’ product usage data, you can identify a simple, basic number to draw a line in the sand and count people as those who are getting value from your product, or aren’t yet.

This number is based on your value metric: the key, measurable thing people do to get value from your product.

“In its basic form, your value metric is essentially what and how you’re charging. If you’re Help Scout help desk software, you’re charging for each seat per month. If you’re selling MacBook Airs, it’s each MacBook Air one time up front. If you’re Wistia, you’re charging for number of videos hosted and the amount of bandwidth those videos take up each month (a dual value metric).” -- Patrick Campbell, CEO at Price Intelligently 

For Slack, the chat tool for work, value is determined by how many messages a team has sent. Slack found their magic number at 2,000 messages sent within a team.

“Based on experience of which companies stuck with us and which didn't, we decided that any team that has exchanged 2,000 messages in its history has tried Slack — really tried it. For a team around 50 people that means about 10 hours’ worth of messages. For a typical team of 10 people, that’s maybe a week’s worth of messages. But it hit us that, regardless of any other factor, after 2,000 messages, 93% of those customers are still using Slack today.” -- Stewart Butterfield, CEO & Co-Founder at Slack

Contentful employs a similar model. Contentful is a content management system that manages your content away from the design and development. As an API product, they look for accounts with more than 500 API calls.

“If somebody is doing 500 delivery API calls, that's a good indicator that they're going to use us for real. If you cross that 500 threshold, chances are you set something up. That POC's up and running, and you delivered some things of worth, and then we decide we really want to try and build it. Before a user has crossed that threshold, it’s more likely that they’re just playing around, testing things out.’ After 500 API calls, there’s a good chance that you become a legitimate regular user.” -- Chris Schagen, CMO at Contentful

This doesn't need to be tremendously sophisticated. What is something that's easily measurable and delivers value for your customer? For example:

- Accounts above their paid plan (not throttled)

- Accounts over their trial limits

- Accounts over a primary value metric in their trial

It may help to view this visually. Pull all your accounts (trial and paid accounts) and plot this out graphically, like the value metric against account spend. Observe visually where there's an interesting change, if any).

This is not supposed to be science. It's supposed to be the starting point of a project.

In summary, if you’re starting out, a basic number is enough to make your first move and “test the waters” with product qualified leads.

If you’re already using a basic “line-in-the-sand” number, and want to progress onwards, it’s time to do some quick and dirty regression analysis.

Find correlations with quick regression analysis

If you can get hold of your product usage data, you can do some quick and simple regression analysis into what are the key actions for driving revenue.

In plain English, linear regression is about trying to draw a line of best fit to show the relationship between two variables (like product usage and account spend). If you can draw a line with a good correlation, where most of the values are on or near to the line of best fit, that shows a strong relationship (though there’s still more analysis to find if one really causes the other).

This can give you a lot of bang with very little buck, simple spreadsheet tools, and a little bit of statistics.

The goal here is to find actions which predict levels of spending that we would have missed otherwise.

This involves:

1. Finding possible triggers in your data

2. Testing your data  with linear regression (to create a model)

3. Testing your assumption (in your model)

Finding Possible Triggers In Your Data

You need to create a shortlist of actions that could indicate high value spend. Whether you’re running analysis yourself or handing it over to someone else, it’s better to focus your attention on fewer actions or a specific area than to look anywhere and everywhere.

There also needs to be a large enough number of those events to be useful. There’s less benefit in finding a valuable indicator that only a tiny proportion of people actually do.

Fetch that data you have already recorded. There are three sources of this:

1. Export your analytics events

If not, dive into your analytics and find a table of all your events. Export this, or copy-paste a summary table, or all your events over a period.

e.g. Google Analytics Event Actions.

2. Database structure

Another alternative is to go into your production database (or ask someone who has access) and extract your database schema. This outlines how and what data is stored.

3. Export your tracking plan

If you’ve documented your tracking plan previously, you’re already good to go!

You might find it to be just one main action that people do or don’t take. For Tim Chard at AdEspresso, connecting their Facebook Ad account was the gateway to getting value from their product.

“In our software, they have to connect their Facebook Ad account to access AdEspresso. A good proportion of them don't, so the marketing effort there is, 'All right. How do we get them to connect the account?'. If they don't connect their account, then they don't create a campaign, those are like the big plummet points. If they don't connect their account, they're definitely not going to become a customer.” --Tim Chard, Marketing Director at AdEspresso

You might want to look at product usage which is going to deliver ongoing value, like payments, connecting integrations, installing code. Look out too for collaborative features like inviting team members.

“People who use Typeform for collecting payments have amazing retention rates.

Integrations, of course, also highly correlated with retention. And then on the other side, people that use logic jump. They're not really sticky because these are features that survey people use.” --Pedro Magriço, Head of Product Growth at Typeform

For more advanced companies with significant numbers of actions and conversions already, you might also want to look at how a series of actions can predict higher spend. For example, Digital Ocean identified the following patterns in customers’ usage data:

- If they're spiking, they may be trying to test a higher workload.

- If they turned on team feature, because if there's a team, it's more likely that they have more workload.

- If they're using a load balancer, then there's more likelihood that the application is more sophisticated.

- If they're using block storage, it's more sophisticated.

- If they are requesting the number of droplets that they have and if they're requesting more, that is an indication of more sophistication

“Essentially, the PQL process is looking for indicators of a propensity to spend more sophisticated-type applications.” -- Emmanuelle Skala, previously VP Sales at DigitalOcean, VP of Customer Success at Toast

Your list may look very different to the above examples, because your product is different. Your goal is to create a shortlist of common actions which you suspect might correlate with higher spend.

Now, to test that shortlist!

Test Your Shortlist With Linear Regression

Next, you need to test whether each action has a significant impact on conversion or not.

To do this, you can use regression analysis. This Harvard Business Review article gives a short refresher on how regression analysis works.

In visual terms, linear regression plots your data on a graph and draws a line of best fit through the middle. How “strong” the relationship is can be measured by the line of best of fit - the “R squared”.

The idea is to find regressions which have a high R2  (so a strong relationship) between an action been taken (the dependent variable) and account spend (the independent variable).

To do this, you need to pull your data. For each action, add the number of occurrences for each account. “Integrations connected = 2”. You also need to include the account spend (or whatever leading indicator matters to you that drives revenue).

For example, here’s a table of 5 accounts with different numbers of connectors and account spend each month.

Account ID Number of Connectors Account Spend (MRR)
1 7 $450
2 13 $1350
3 8 $1100
4 4 $250
5 5 $300

Once you have all your data, you can do basic linear regression analysis with a spreadsheet quite simply.

Here’s an example of what this may look like with a small sample (in our example, just five accounts).

In our example, let’s consider the number of integrations connected. This is our independent variable, and sits on horizontal x-axis on our chart. 

Account spend is the dependent variable on the vertical y-axis.

In plain English, we ask “is there a relationship between {action you’re testing} and account spend?”

To do this, we’re plotting a line of best fit with the following equation (remember the formula, y = ax+ b)

Account spend = “a number” x “number of integrations” + an error term*.

*the error term captures all the other variation. A strong relationship would have a small error.

You can do this graphically!

1. See It: Create it graphically.

Create a chart with account spend and the key thing you’re measuring on a chart. In Google Sheets, go to Chart Editor → “Customize” → Trendline: Linear. Make sure you check R2 to get the measure of fit.

2. Measure It: Create the equations.

Then, test your assumptions in real life.

Once you’ve found correlations between actions and purchasing behaviour, you need to test it.

“Correlation does not imply causation”. Just because two things are related does not mean one caused the other.

To see this illustrated, take a look at this excellent website outlining various spurious correlations, such as the age of Miss America and the number of murders by steam, hot vapor, and hot objects. They’re ridiculous reminders that correlation does not imply causation (like the age of Miss America impacting a specific type of murder).

There are many reasons for correlation to not mean causation in your data. 

There may be other variables you haven’t included in your trendline linear regression analysis.

To test this fully, you must build your hypothesis - your proposed explanation - of the relationship and test it with new data. If you see the same results as you predicted, you can be more confident that one causes the other.

For instance, if we suspect the number of integrations drives higher account spend, then work on a campaign to encourage people to connect more integrations and see if those accounts have higher spend.

But, hold back some people (don’t include relevant people in the campaign) to see if they would have spent more without your campaign encouraging them to connect more integrations.

This analysis is key to moving forward and proving you have the right triggers.

For DigitalOcean, they wanted to prove that their product usage sales outreach was driving revenue. To do this, they setup a sales A/B test.

“We basically did an AB test. It wasn't a marketing AB test, it was more of a sales AB test. This is the very first time we ever used product data for any kind of proactive outreach, was with PQL. 

We did a control group where we held back 20 percent of the PQLs and no one would follow up with them. We looked at the performance of three different groups: the performance of the control group that had no outreach, the performance of the group that had outreach but never responded to that outreach, and the performance of the group that had outreach and did respond, so therefore, had some kind of interaction. 

What we proved was that the third group, the one that had interaction, was two times more likely to cross a spend threshold that we deemed significant and once they did cross that spend threshold, they spend, on average, 30 percent more than the first, the control group, which is the one that had no outreach. 

Interestingly, even the group that got outreach but did not reply, it didn't perform as well as the one that obviously had interaction but it still performed 50 percent better than the group that got no outreach. We were able to prove it was significant.” -- Emmanuelle Skala, previously VP Sales at DigitalOcean

There’s a far more advanced statistical analysis you can do - but quick and simple linear regressions are a great way to get started.

The statistics for in-depth analysis of series of events, testing the relationship of multiple events together, and so on is the subject of a dedicated statistics guide.

The question still starts from the same roots - does X cause higher spend?

If you’re struggling to do this yourself, or with your existing team, but recognize you’ve a lot of complex data to use, then 

Scale up? Bring in the Data Scientists

If you’re not convinced in your quick regression analysis, struggle to do it, or have such volume and complexity to your product data that you can’t confidently understand what’s going on and why, then look to hiring professional data scientists.

Data scientists can do the more complex analysis and modelling you to find multiple high-value triggers you can use to qualify and close more business.

But this takes a lot of specialist skill to get it right.

Unless you’re lucky to have a team on-hand who can help you, take a look at these dedicated Data Scientist Marketplaces. They allow you to define your project (to find leading actions which correlate to revenue) and collaborate with data scientists from around the world.

https://www.experfy.com/

https://www.pivigo.com/

http://datastars.co/

Beware of lead scoring. It can be a mixed blessing for sales enablement.

You can take this a step further - artificial intelligence can power predictive lead scoring. 

Usually, our products aren't so simple that one value metric totally sums up the value a customer is getting.

Instead of managing a bewildering number of metrics, scoring is used to sum up all that activity and action into one number. This is very common with marketing qualified leads to aggregate lots of individual signals.

This takes some volume of trial users and successful sales to train the models. At lower volumes, the “learning” element may take too long to train onto what matters most for driving revenue.

This makes it easy to classify and group - "we only qualify leads over 70" - but these can have vastly different profiles and means of achieving the same score. 70 what?

Remember, the goal is to change the message. Grouping people by score doesn't do that. The Appcues sales leadership team initially tried a lead score, but they found it didn’t work.

“They don’t want this intelligent thing that say “this lead is 66% more likely to buy”, because they can’t use that to communicate with the prospect. They can’t say “hey, I’m reaching out to you because you’re 66% more likely to buy” but they can reach out to a prospect and say ‘Hey, you just installed. What are you looking to do next?’” -- John Sherer, Director of Sales at Appcues

Scoring is pretty meaningless on its own without context, so scoring alone doesn’t change sales behavior. Generic scores blur all the context needed to deliver meaningful, relevant content. This is not good enough when you have product usage data - actions people have taken - at your fingertips.

“We thought it was going to be the usage score, but ultimately that’s not as valuable as we thought. I think there’s an interesting concept of doing all the thinking and aggregating - almost some kind of algorithm telling them. That’s kind of the extreme of helping a salesperson workflow. 

Sales people are emotional. They have the impression - there’s a lot of context that is not a part of the score. And then the score just becomes a part of the whole story. And it’s about as valuable as the raw inputs.” --John Sherer, Director of Sales at Appcues

But scoring is still useful for managing a complex set of activities.

One avenue to this is to use multiple, topical scores. Score engagement and activity based on something which can start a conversation.

The insights for this may lie deeply within your product data, but you can also infer more signals from data outside of their product usage. If they’re inviting their team to your product AND you know they’re a manager from your CRM, you can reach out to them with specific advice around that.

But without the “reason” from a lead score, it doesn’t enable your sales team. They need to be cued in with something to say.

The key takeaway - lead scoring alone isn’t helpful. You need to have the context as to why a lead is a product qualified lead, so your sales team has all the clues, cues, and conversation starters to be helpful.

Start Small. Then go from there.

Some of these techniques may seem daunting if you’re unfamiliar with data analysis.

But.

Don’t be afraid to try and get started with some basic techniques.

Appcues started with anyone who saw their “trial limits reached” modal.

This was shown by their product when an account reached the limits of their free trial. This served as a proxy for usage in general.

“We want to alert sales to people who are using the product effectively within the trial so they can prioritise their time to be spent with those people.” --John Sherer, Director of Sales at Appcues

They could measure accounts who reached this by triggering an event when this specific modal was viewed. This was then sent to Hull, a customer data platform, to update their marketing tools, sales CRM and Slack to close product qualified leads.

This was enough for them to unqualify 30% of their leads from sales and see a jump in their sales efficiency.

If Appcues could get started with PQLs by creating and syncing one event, then you can get started too.

Here’s how to get started.

Takeaway Points: How to find your “aha!” moments

1. Pull your raw product usage and graphically determine what features appear to result in a jump in spend. This isn’t science, but it’s a start. You can do this yourself quickly.

2. Next, find correlations with quick regression analysis. This may take an hour of your time looking over several combinations of product usage data. Again, you can do this yourself without significant budget.

3. Finally, when you’re sure you have the volume and complexity, look into building a predictive model with data scientists. This will take time to train and refine.

Authored By

Claire Suellentrop

Founder, LoveYourCustomers

Claire Suellentrop helps high-growth SaaS companies get inside their customers’ heads. Previously the Director of Marketing and #2 employee at Calendly, she’s seen firsthand that truly effective marketing stems from a deep understanding of existing users’ needs.   

Now, she works with companies like Wistia, FullStory, and MeetEdgar to uncover their best customers’ needs and desires, then uses those juicy details to create more relevant, high-converting marketing and onboarding campaigns. 

Ed Fry

Head of Growth, Hull.io

Ed Fry is passionate about helping marketers grow their organizations and directly contribute to revenue. He was the first employee at Inbound.org and worked with thought-leaders like Rand Fishkin, Co-Founder of Moz, and Dharmesh Shah, Co-Founder of HubSpot. During his tenure, membership grew from 5,000 to 165,000+ members between 2012 and 2016.

Ed currently oversees Growth at Hull - a customer data management platform that eliminates data problems for marketing and sales teams alike. 

Get Each New Chapter In Your Inbox

Don't want to miss out on the upcoming chapters? Sign up now and we'll deliver each chapter as soon as we publish! 

No spam, no gimmicks.

Just great content.

By submitting this form, you are subscribing to future content from Hull (including the PQLs guide) and acknowledge that the information will be processed according to our Privacy Policy.

First Name

Last Name

Company Email