Really enjoyed this post about how HubSpot manage their SaaS Sales process:
It’s all interesting, but the bit that really caught my attention was the part about the interaction between sales and marketing (found under “How do you work with Marketing to ensure a consistently high quality of leads?”). Here, Mark Roberge makes the point that you have to assess the quality of leads that marketing pass through to sales as much as the volume, indicating a scoring system (e.g. “Requests a Demo” might be 10 points – a good lead, “Downloads whitepaper” might only be 1 point). I really like this and it reminds me of one of the sections in the book:
One of these rules is to “Generate Demand, not Leads”. Here, Christine Heckart argues a similar point – that generating low quality leads (which all of us can do, with a few basic tricks 😉 ), is not a fruitful use of your time, and that you need to look at generating real base-line demand for your product (i.e. so that customers have a genuine desire for your offering).
Both of these beg the question though of – how do you measure what is a high quality lead? Here are a couple of options to start with:
- Good lead information is provided – full company details, name, job title, phone numbers, work email addresses. This is surely a better lead than “Mickey Mouse, firstname.lastname@example.org”?
- As Mark suggests, use their actions on the site – a demo request is surely a better lead than just a whitepaper download, isn’t it?
Both of these are likely to be true. However, the thing to watch is – what are you trying to target here? Certainly, if contact details are provided, a customer is easier to contact. However, do you know that customers with poorer contact details are actually worse leads? Equally, do you know that a demo request is actually a better lead that a whitepaper download?
There is analogy here which might be helpful. In machine learning, there are two sorts of learning algorithms – unsupervised and supervised. Unsupervised learning algorithms including many clustering techniques that allow you to find patterns in a set of data. For example, you might find, from looking at lead data, that people who enter correct email addresses, also tend to enter correct phone numbers and their own names. And, those that enter no phone number, also tend to be cagey about their personal details. These clusters are interesting, but, because there’s not a target variable, you can’t really say anything about whether cluster A is “better” than cluster B.
Supervised learning algorithms, in contrast, provide ways of predicting an output (e.g. “This is a good lead, this one isn’t”) based on input variables (such as whether they entered a phone number, or downloaded a demo).
To really test whether a statement about lead quality is true, you need to define a target (or output) for your lead classification system. By this I mean – “What, at the end of the day, would mean a lead was good, and what would mean it was poor?”. For me, there’s two answers to this – money and “ability to influence the purchase decision”.
The purpose of a lead classification system is to figure how your sales people should best spend their time. If you only get 1 lead per week, then I don’t think you need an automated lead classification system – you can assign an individual to that lead to work out a fully personalised marketing and sales strategy for that individual. However, hopefully you have a slightly richer sales funnel than that! If you have a lot of traffic and and are wasting a lot of time on poor leads, then the two outputs you should be targeting are – which leads are likely to end up generating more revenue, but also, less obviously, which ones can I actually affect?
There are certain types of leads which, to be honest, there really isn’t a whole lot you can do to impact (either up or down). Customers who already know what they want, know your full product range and have bought from you 10 times before – for these people, you might just be best getting out of their way. Sure, we’d all like to think we can upsell them to the bonanza bonus pack for twice the price, but if they know what they want – this is rare.
Other customers however – who don’t really know you, who are just dipping their toe in (perhaps with a whitepaper, or a “Can you tell me something about you guys?” email) – these might be the people where your expertly trained sales people can really influence those people – tell them the story of your products, about your other clients, demo the product to them and so on, and convert a relatively cold lead in to strong demand. And perhaps that might only end up with a small deal to start, but by looking at the Lifetime Value (LTV) for that type of lead, you might realise that they go on to be your most loyal and interested customers.
This does require some analysis – you need to be able, firstly, to understand the input variables that are of interest. Is it “Amount of lead details completed”? Is it what they did on your site? Is it their vertical, their lead source on Google Analytics (assuming you can trace back as far as you need..)? What?
A precursor to figuring this out, is knowing what you’re trying to target. As mentioned above, you can look at revenue or “ability to influence”. To keep it simple, you can start by just looking at the first of these though, even there you have to assess – “Is it value of first purchase, or LTV?”. I suggest you need to look at LTV to properly assess which are the leads which really end up generating long term revenue for your company.
As mentioned, once you know your target, you need to work out your input variables – which values indicate whether a lead is interesting or not. Methods like CHAID and similar ideas are useful ways of assessing things like “Does the fact that my lead came from Google Organic rather than Google PPC make a blind bit of difference to how much they spend with me?”. It’s very important to do this sort of proper analysis – using gut instinct (“Surely a Google Organic lead is more valuable than a PPC lead?”) can be dangerously misleading as you mis-use resource to focus in the wrong areas.
Once you have your model, you need to test it out – has all that extra effort actually ended up with a process where your sales people have made a genuine difference to the right leads? It’s actually very hard to get a model like this right, so you should assume you’ll be going through several iterations..
Finally, on the specific, early subject of “Lead details completion quality” – I’ve got a spreadsheet that I use to assess things like phone number and email quality (weeding out “123123123123” and “email@example.com”), which I’ll post up next time, once I’ve cleaned it up a bit!