Lattice Engines has taken the pole position in the emerging Predictive Analytics space. In yesterday’s blog, I covered its pricing, value proposition, content, and integrations. Part two covers model building.
When first launched, Lattice Engines and its peers had long deployments and black-boxed models that required data science expertise. The firm now offers 24-hour deployments, simplified model building, and greater transparency around models and recommendations. Furthermore, the system allows marketers to either build their own models or import industry standard PMML files constructed by their data science teams.
Predictive models are built by importing training files which are matched against the Lattice Data Cloud using D&B DUNSMatch logic and Lattice proprietary techniques. Training models contain examples of both positive and negative outcomes (e.g. win / lose, renew / drop). A model is typically available within thirty minutes of the training file upload.
Ideal Buyer Profile scores (Lattice’s term which is similar to Ideal Customer Profile scores) are available to sales and marketing and include both scores and recommendations. Marketing can view the model via a graphical Data Cloud Explorer which highlights the key signals and variables in the model and makes the data available for export to other platforms.
To make the data more actionable for sales reps, Lattice provides Salesforce Talking Points which display recommendations and explanations that include Lattice data, transactional history, and buyer behavior. A Lattice Buyer Insights CRM I-frame contains Lattice recommendations, talking points, company profiles, company fit, engaged contacts, engagement activity, intent analysis (surging topics), web activity, and purchase history tabs.
Future plans include a user interface for segmentation analysis and simplifying intent scoring to high/medium/low.
Back when I was a product manager, I used to conduct sales training classes. I often opened up the session by asking the question, “Who is your biggest competitor?” The reps invariably listed a company or two they had heard over the prior day and a half of training. Even seasoned reps would answer the question incorrectly.
Unless you are in a duopoly or there is a competitor that controls half the market, your biggest competitor is probably NO DECISION. Either the purchasing decision is kicked down the road or no funding is found. It may also be that the opportunity was poorly qualified to begin with.
Sales reps no longer control the conversation due to the informed buyer who leverages the Internet and social media in order to research vendors prior to contacting them. This is one of the reasons that marketing is looking at digitally influencing anonymous individual on the web via Visitor ID, SEO, SEM, and Programmatic. Sales reps are also confounded in their sales efforts by a second change in purchasing patterns. B2B budgetary decision making processes have become more complex.
Budgetary centralization and committee-based buying decisions have increased the number of decision makers in the purchasing process, resulting in a greater likelihood of no decision. According to a Forrester survey of IT sales reps, 43% of lost deals weren’t to competitors but to a category titled “lost funding or lost to no decision: customer stopped the procurement process.”
Furthermore, the rise of cloud computing has shifted budgetary decision making authority away from the CIO to the heads of various functional departments. Purchasing decisions are being compared to a broader set of non-related purchases from across the organization. It is therefore critical that sales reps “understand and navigate complex agreement networks and processes within the buying organization that span different altitudes and functional roles,” blogged Forrester Sales Enablement Analyst Mark Lindwall. “Because decisions are more cross-functional, every dollar is compared against how it could add value in potentially completely non-related areas of investment.”
Thus, sales reps need better tools for identifying who to engage and when best to engage. They also need to be better informed about companies, individuals, and the industries into which they sell. In short, they need to know who to call, when to call, and what to say. They need to quickly navigate what Forrester calls agreement networks to establish relationships across multiple levels and job functions at the organization.
Fortunately, Sales 2.1 tools provide rich biographies and full family trees for navigating these networks. Users can target specific job functions and levels across the corporate hierarchy, research the appropriate individuals, and reach out to them via social media, email, or phone.
Newer ABM tools help identify the Ideal Customer Profile (ICP), score leads based on the ICP, and call out similar accounts and contacts that are not on the company’s radar. Thus, it’s not just about selling more intelligently based on insights, but targeting and prioritizing one’s sales efforts more effectively.
Sales triggers assist with identifying executive changes, M&A events, product launches, and other reasons for reaching out to individuals. Triggers can also indicate an expanding opportunity or that a proposal is potentially at risk due to company or market dynamics.
And yes, sales reps should research both the company and the executive. They need to understand the key trends in the prospect’s industry, why their last quarter was soft, and what does the executive muse about on social media. While such facts may not be immediate hooks, they provide context and potential talking points down the road. It also shows that the rep is willing to invest time in understanding the exec, her company, and the environment in which she is making decisions.
There is an opportunity cost to poor targeting, prioritization, and account planning. It shows up as No Decision in your CRM, slow deal velocity in your pipeline metrics, and disappointing sales growth.
French predictive analytics firm Sparklane unveiled their version 2.0 Predict platform which employs artificial intelligence (AI) and active learning to score millions of companies and determine which prospects are most likely to become net-new customers. The Predict platform is available for the UK and French markets with localized language and datasets. A German edition is in development.
Sparklane ingests and enriches company data, matching it against firmographics and trigger events to score millions of companies. The system then models the Ideal Customer Profile (ICP) and Total Addressable Market (TAM). Sparklane also identifies “sparks” (hot prospects) based upon sales triggers and delivers real-time alerts, messaging, and contacts.
Models can be deployed for both new and existing business. New business models can be constructed from historical data (e.g. CRM win / loss flags) or estimated and refined for new market entry. Existing business data can also be deployed for churn models to help identify companies that are more likely to drop as well as upsell and cross-sell models.
CEO Frédéric Pichard said that employing artificial intelligence to identify your next best customers “is probably the most amazing promise B2B marketing and sales tools can fulfill” as it provides “a new way of working to help our customers be more efficient and successful.”
Sparklane users begin by importing datasets from CRMs or CSV files. Logic is employed to determine both positive and negative sample records. For example, a CRM Win / Loss flag could serve as such an indicator. The file is then enriched and an ICP model is constructed. The ICP contains three types of variables: Fit (firmographic), Need (Triggers), and Behavior (Marketing Automation prospect activity). Marketers or Sales Operations are able to view the model and adjust weights. This model is then employed for constructing a TAM with net-new accounts which can be saved as a fixed account list or dynamic model.
Sparklane onboarded file mapping.
An accuracy score helps define how well the model distinguishes between good and bad prospects. Thus, an 80% accuracy score indicates that 8 out of 10 companies in the seed file are properly predicted by the model.
An accelerated learning option is available for new market entry. Thus, if a seed list of good and bad prospects is not available for a new product line or market, an initial set can be manually selected from Sparklane company lists and deployed as a first generation seed list.
An active learning option allows users to perform a qualification pass on a list to help expedite model construction. While engaged in active learning, the user is shown company profiles which include account overviews, triggers, and family trees. The marketer can then give a thumbs up or down to each proposed account.
As output, the platform provides a set of “sparks” which are high probability accounts or contacts. The user sets the number of sparks displayed in a spark list. Qualified prospects can be sent to a CRM as accounts or leads.
The French dataset covers three million firms and two million contacts. The UK universe provides 200,000 companies and 300,000 contacts. The UK dataset focuses on large companies with sales triggers.
The French file includes 600,000 emails while the UK file supports 100,000 emails.
The firm claims that Predict increases the opportunity conversion rate by 70% and shortens the sales cycle by 30%.
Sparklane employs sixty headcount in Paris, London, and Nantes. It invests over 20% of its turnover in R&D and has nearly 200 customers in Europe.
As with many other technologies and business processes, sales is subject to its set of TLAs (three letter acronyms) such as ICP, TAM, and ABM. As I regularly reference these terms in my blog, I obtained permission from InsideView to republish their slide on these acronyms.
The Ideal Customer Profile (ICP) is your best customer definition. It is a hybrid of both company and contact variables. While it can be as simple as “the Fortune 500,” a true ICP looks at firmographic, biographic, technical, and signal variables. By technical, I mean industry specific variables such as which platforms are used, how many beds are in the hospital, or whether the company is a direct seller or employs channel sales. By behavioral, I’m talking about business signals such as funding events, partnerships, and M&A activity (what InsideView calls agents and other vendors call triggers).
Defining your ICP is key to strategic targeting. Without an agreed upon ICP, sales and marketing will take an ad hoc approach to customer targeting and prioritization. At best, the lack of an ICP is sub-optimal. At worst, it results in sales ignoring marketing leads and taking a “we’ll do it ourselves” approach.
The Total Addressable Market (TAM) is the full set of customers, prospects, and net-new accounts that match your ICP. Of course, some of your customers and prospects will fall outside of your ICP, but it is the net-new accounts that are the most interesting. Some call these the white-space accounts, but they are basically the companies you should begin nurturing as they represent your best hope of growing revenue. Likewise, prospects within your TAM should be a high priority while those outside should be triaged. Finally, the accounts that fall within your TAM should have high retention rates. They also represent an easy path for cross-selling, upselling, and expanding to other departments, functions, and locations. You want to go from beachheads (land and expand) to strategic partnerships with these firms so deep company intelligence is required (family trees, org charts, additional contacts, sales triggers, SWOTs, industry research, etc.)
Of course, Account Based Marketing (ABM) is the broader strategy that is supported by a focus on your TAM and ICP. ABM is the set of programs, campaigns, and activities by which B2B companies target their best prospects. ABM encompasses sales, marketing, customer support, operations, etc. Once the firm agrees on which accounts are strategic, it can direct its energy towards landing these accounts and ensuring they receive the white glove treatment. While traditional demand generation and content marketing have focused on lead volume, ABM directs sales and marketing resources towards targeting and expanding business within your TAM.
Implementing ABM encompasses a set of tools and services for identifying the ideal customer profile, sizing the total addressable market, identifying white space target accounts and contacts (i.e. net-new leads), supporting web forms, automating batch and ongoing enrichment of MAPs and CRMs, prioritizing leads, embedding sales intelligence within workflows, event alerting, prioritizing leads, and assisting with lead-to-account mapping, segmentation analysis, and campaign targeting. Other ABM technologies include programmatic marketing, dynamic website display based upon real-time firmographics (visitor id), predictive analytics, and proactive sales recommendations. No vendor provides all of these tools today, much less has them integrated into an ABM suite.
Dun & Bradstreet, which has had a series of major product announcements over the past few weeks (the Avention acquisition, rebranding of its OneSource platform as D&B Hoovers, a Beneficial Ownership product), has quietly added powerful new functionality to their Workbench Data Optimizer platform. The new Profile capability features an automated profile builder, Total Addressable Market (TAM) analysis, and look-a-like prospecting based upon the Workbench profiles.
The new functionality helps marketers evaluate the size of targetable sub-markets, identify audiences with a high propensity to purchase, discover overlooked whitespace opportunities, and target new accounts and contacts. According to Alex Schwarm, Sr. Director of Marketing Analytics Products, “Profile enables our Workbench customers to begin to use data-driven, ABM-oriented Profiles based on their successful sales. These automated analytics allow you to quickly and easily identify the best whitespace opportunities and characteristics of your target audiences including those with the highest propensity to buy – no data scientist needed.”
Profile is a black-box analytics engine which clusters customer files without biases. Marketers upload a file of their customers’ data for a specific product or product family. Workbench standardizes, de-duplicates, and verifies the input file; matches and enriches it with Dun & Bradstreet’s WorldBase firmographics; and then provides segmentation and file health analysis. The Profile module identifies between two and eight distinct segments containing similar companies across multiple dimensions. The user can define the number of profiles or the system can automatically identify the optimal number of profiles based on the variation of the customer file. The marketer is not required to define the key segmentation variables. Instead, the system automatically performs affinity clustering (my term) to build the segments. Execution time is typically 5 to 10 minutes.
The results are displayed on a downloadable dashboard that provides a side-by-side firmographic analysis of the clusters. Results include company size, ownership (e.g. parent, branch), primary industries, cluster size, and average deal size (if revenue figures are also shared with Dun & Bradstreet). Thus, the system may identify segments with a lower average deal size but a larger number of prospects alongside clusters containing top customers with high average deal size but a small number of targetable opportunities.
While Dun & Bradstreet does not use the term “Ideal Customer Profile” (ICP) the system is basically identifying the attributes of a customer’s ICP, determining the average deal size, and sizing the overall market opportunity.
Dun & Bradstreet has two major assets in performing TAM analysis: The WorldBase file of global companies and trust built up over 170 years of credit research. WorldBase provides them with a consistent, global file of 260 million active and inactive companies for credit and supplier risk research, sales intelligence, and B2B marketing. The file includes broad global company linkages, corporate and location sizing, industry coding, Tradestyles, and D-U-N-S Numbers (the de facto global company numbering system). This intelligence provides the core reference file against which market sizing can be performed. But TAM analysis requires customer level revenue information against which company counts can be converted to market sizes. And here is where a strong credit analysis brand helps build confidence amongst marketers to share company revenue data. While they will be reluctant to share revenue details with most vendors, firms have been sharing private financial details with Dun & Bradstreet over the better part of two centuries.
Marketers can then take any of the profiles and immediately identify net-new similar companies as well as net-new contacts. The system also sizes potential target market audiences that can be reached programmatically through their Audience Solutions group.
While prospect scoring based upon these definitions is not yet supported, that is a likely future offering for the platform. Profile, along with a set of predictive scores and paired with D&B Hoovers’ business signals, represents a toe in the water of the predictive analytics space.
Intelligence vendor DiscoverOrg announced a new Account Based Marketing (ABM) tool called AccountView which helps marketers identify the attributes of their Ideal Customer Profile (ICP). The new feature analyzes an account file which it calls a portfolio, enriches it with firmographics and technographics, and then provides a portfolio visualization dashboard of the accounts. The service also identifies similar companies to the top accounts, prioritizes them, and identifies best fit decision-makers at the net-new accounts.
The portfolio segmentation dashboard tiles include
Size: Revenue and Employee Bar Charts
Industry: Primary Industry Pie Chart; SIC and NAICS top frequency lists
Technology: Technology lists
Ownership: Ownership Structure Pie Chart
Companies: Portfolio companies with employee and revenue data. Company names are hyperlinked to their DiscoverOrg profiles.
Although geographic segmentation is not yet available, it is on the product roadmap.
Within the list tiles, users can search for specific elements (i.e. SIC, NAICS, technology, or company name).
Proposed contacts are shown within org charts with direct dial phones and emails to assist with organizational context and reach out. DiscoverOrg also provides detailed platform information and a set of sales triggers.
Marketing and sales teams can drill into specific bars or wedges to further research segments. To quickly refine models, customers can remove outliers to focus the ICP around high frequency variables.
Portfolios may be uploaded as CSV files, bulk matched within DiscoverOrg, or generated via DiscoverOrg prospecting. Result lists may be saved as lists, viewed as searches, or exported to CSV files. Models may also be loaded into DealPredict where company lists are displayed with Deal Predict scores of zero to five stars. Next to DealPredict scores, DiscoverOrg displays a lightning bolt icon if the company has a Sales Trigger or OppAlerts in the past sixty days. OppAlerts are intent based triggers which have been researched by DiscoverOrg editors or gathered through B2B publishers’ online content consumption data. By clicking on the lightning bolt, reps are shown the related events.
Within DealPredict, company lists are dynamically maintained to reflect the current firmographic and technographic lists of companies. If there is a change in company size or implemented technology, the DealPredict scores are automatically updated every time a search is conducted. Likewise, companies which are added to the DiscoverOrg database are automatically scored.
The very foundation of successful sales and marketing is figuring out who your best customers are, understanding why they are the best, and finding more prospects just like them. What could be a painful analytical exercise is made simple and straightforward with DiscoverOrg’s account-based marketing features, and the result is faster growth for customers who can more effectively identify, understand, and engage with their ideal buyer.
DiscoverOrg CEO Henry Schuck
DiscoverOrg suggests a number of account list categories that can be analyzed including the full customer list, high or low spend customers, renewing or non-renewing customers, high or low profitability customers, competitor customer lists, and prospect accounts. For example, running a competitor’s customer profile through AccountView helps you “determine ways to improve your product, messaging, or positioning. Likewise, running the non-renewed customer list through AccountView will help identify high-churn candidates for special programs.
Although DiscoverOrg recommends sets of strong and weak account lists, AccountView does not have the ability to discriminate between the two categories. Thus, marketers would need to separately run the paired lists, compare the portfolio results, and adjust the models for overlapping variables. For example, knowing that Microsoft Office is heavily used by both strong and weak accounts would indicate that MS Office is a frequently occurring, but non-predictive variable.
Future features include support for multiple models, grouping tech functions by category, sharing models across all users, geographic segmentation reports, and uploading contact information to assist with defining job functions and levels.
AccountView is the latest capability within DiscoverOrg’s ABM Toolkit. Other features include DiscoverOrg’s DealPredict predictive rankings for companies and contacts, OppAlerts intent-based opportunities, and sales triggers.
DealPredict provides predictive scores similar to those provided by predictive analytics companies. DiscoverOrg CMO Katie Bullard noted that unlike some black-box predictive platforms, AccountView analysis and DealPredict models are fully visible to sales and marketing users.
The AccountView analytics and net-new account service is included as part of the DiscoverOrg service. Firms license access to specific DiscoverOrg datasets and a set number of seats. Licensed users then have unlimited access to the licensed content for viewing, uploading, or downloading.
Other sales intelligence companies that have developed AccountView-like functionality include Dun & Bradstreet (Workbench), Avention (DataVision), and Zoominfo (Growth Acceleration Platform).
DiscoverOrg, which hit $71 million in Annual Recurring Revenue (ARR) at the end of 2016, has expanded its customer base beyond technology companies. Over 15% of revenues now come from marketing agencies, staffing firms, and consultancies.