This article was written in collaboration with Datomni, our partner company, which provides a Dockerized, real-time customer data infrastructure that you can deploy in your cloud to capture data from multiple sources, including Pipedrive.
To say that the AI market is hot in mid-2024 is banal. There are tons of new AI applications coming to the market every day. In this flood of new applications, you may be wondering whether there’s any point in actually using your legacy tools any longer. After all, sometimes they have a legacy setup and are not so quick to embrace the latest in AI because of technological debt. We’re happy to put everything that’s going on into a new context.
We believe that the surge of AI solutions is an even stronger argument for you to stay with your legacy tools and simply use them in a better way. We believe that if you take the time to properly restructure the legacy platforms and take care of the data contained in them, then over time you’ll turn them into extremely valuable resources that AI will be able to elevate to the next level and help you automate your operations. Conversely, if you start skipping through the newest tools without ever taking the time to properly configure any single one and accumulate significant amounts of data there, then no AI will be able to help you effectively and will only provide you with rather generic outputs.
The ideal situation, if you have legacy tools in your tech stack, is to select a tool that has AI as a strategic development direction (SaaS platforms are open about this) and then make sure that you perfect your end of the equation, such as data models, data richness, and all things data. Accumulate enough data for this platform’s AI to boost your experience.
In this article, we’ll give you very specific recommendations and tips regarding the preparation of your Pipedrive account to utilize Pipedrive AI solutions to the maximum advantage. We believe that if you adopt these recommendations and stick with them long enough for specific patterns and data points to accumulate in significant amounts, you will be able to derive disproportionately high value from the AI upgrades that the Pipedrive team is working on.
Create rich data schema
The first thing you should do to future-proof and prepare your Pipedrive for the AI boost is to ensure a rich data schema.
When we audit Pipedrive accounts, either by ourselves or by our partner company, Datomni, we see companies ignoring the greatest tool they’ve been given by the creators of Pipedrive—the ability to make the Pipedrive data schema their own. This schema can be completely adapted to who their customers are, what they do, what characterizes them, what impacts deal valuation, their ability to close, or their general sentiment towards the sales process they’re in. Pipedrive custom fields, formula fields, and the recently added required and important fields give you the chance to capture everything that is unique about your process in the contact, deal/lead, or product data model available in the account.
And capture them you should. The ultimate purpose of getting your Pipedrive account structure ready is to support the backend AI models that Pipedrive’s team is working on, as well as the completely custom ML models included in our own Samurai’s package, Pipedrive AI Booster. This preparation enables the attribution of sales and engagement signals to the specific characteristics of your deals, leads, contacts, organizations, and other key objects Pipedrive offers. All things being equal, the deeper, more detailed, and more intricate your data model, and the higher the data quality it stores, the more associations ML and AI models will be able to capture. By precisely and accurately capturing the lifecycle and evolutions, as well as changes to the specific objects, and by describing them thoroughly in terms of their characteristics, you maximize the chance of the ML capturing associations and relationships you never even thought existed and then using them to optimize and fuel your sales process. Ensuring your custom schema is spot-on and robust, and keeping it so over time, will create multiple layers of added value from the data you already have in your account.
Now that we’ve established the need for using custom fields and a rich, deep data structure, let’s consider how to create a robust custom schema that maximizes the effectiveness of Pipedrive AI. The key rule is to design your custom schema to maximize the differentiation potential of the data captured in your account. To put it simply, custom fields should cover a wide range of values, encompassing and correlating with both high and low sales engagement. The specific fields you choose will depend on the entity or object you are documenting and your business needs. For organizational contacts, it’s crucial to capture company size, number of employees, and location. For individual contacts, include whether they are decision-makers, their seniority and job title, and verify details like email or phone legitimacy and testing status. For deals/leads, capture the source, the sales representative handling the transaction, and all relevant deal characteristics such as pricing plan, subscription method, delivery method, and ideally, a composite score indicating deal quality. If you use products, extend your standard product data model with custom properties like product variants, types, and categories. These additions will prove valuable downstream for modeling the efficiency of new deals or leads through your sales systems.
Another important aspect is to always capture data properly in the designated fields you’ve created, and more importantly, to use predefined value lists whenever possible. Properly capturing data in designated fields, such as numerical data in numerical fields instead of text fields, ensures that the underlying AI or backend systems can apply appropriate data transformations to derive maximum value. For numerical fields, these transformations typically include logarithmic transformations, which are also utilized in our Samurai Predictive Data Infra. For categorical fields, AI and ML models can employ techniques such as one-hot encoding. However, this doesn’t mean that you cannot use unstructured text data at all. For instance, in our Samurai AI Predictive Customer Infra backend, we apply tokenization to unstructured text data fields to derive valuable signals. In summary, ensure that the data type accurately represents the actual data in your account’s specific fields, and avoid contaminating this data with irrelevant information. This approach will enable the Pipedrive AI system to associate activities with specific elements of the deal evolution effectively.
Ensuring a robust and AI-ready account structure goes beyond just capturing a rich data schema in the appropriate data types for deals, organizations, contacts, and other entities used in Pipedrive. It encompasses the entire sales process. Starting from the initial contact, you should qualify each interaction to assess if there’s any potential for the contact to convert into a lead in the future. It’s crucial to label and tag contacts not only with significant potential but also those with no relevance to your company’s offerings. Over time, such a diverse dataset, though initially overwhelming, ensures the highest quality data model in your account. Continuously feed Pipedrive with new data. When evaluating a contact and preparing to assign a lead label, utilize Pipedrive’s built-in lead labeling function. This allows labeling contacts as hot, cold, promising, etc., including custom labels. Design the lead label schema to cover the full spectrum of values your representatives can assign, including negative labels like “fake” or “no-go.” After labeling, proceed with a qualification process where you define the lead’s needs and budget. Implement a well-defined qualification regime such as BANT (Budget, Authority, Need, Timeline). Ensure the qualification status is accurately recorded in the lead’s detail view. Once a lead is well-qualified and there is confirmed interest or collaboration potential, convert the lead into a deal. Structure the deal pipeline as a unidirectional flow from cold deal to won deal, without cycles. If multiple pipelines exist, ensure each is uniform in deal characteristics while maximizing differentiation from other pipelines to enhance information gain. Name stages in the deal pipeline clearly. Each deal’s lifecycle should conclude in either a won or lost stage. Use Pipedrive’s feature for managing open deals, including setting up deal rotting at the structural level. Define probabilities of closure based on current stage. Eliminate catch-all stages like “all won deals” to maintain clarity in the deal pipeline structure. Instead, use Pipedrive’s filtering function for specific visibility needs without affecting model clarity. Remember, the deal pipeline represents a unidirectional sales process, guiding deals towards closure. In summary, ensure that everything you record in your Pipedrive fields accurately represents real-life progression of deals through the pipeline.
Finally, regarding account structure, it’s essential to create a dedicated account for each sales representative and utilize the teams feature if multiple teams sell your products, such as location-based teams. Additionally, one of the most crucial custom structured elements to implement is the custom activity structure. It’s evident that every company will have a slightly different sales process and will use various sales activities like demos, presentations, or even offline events such as lunches and golf trips. While leveraging standard fields is beneficial since the machine learning models operating behind Pipedrive include a benchmarking component to extrapolate general patterns, it’s also important to incorporate a custom activities schema in your account.
If you remember all these structural aspects of your Pipedrive account, including a rich data schema with proper data types representing the full range of values (not only positive signals but also negative signals), you will ensure that the backend AI, whether running externally or on top of your account, provides maximum value and even identifies patterns that you never thought correlated with deal success.
Fill rich data schema with high quality data
Even if you create a perfectly rich data schema, it’s still not enough to ensure that you can derive maximum value from Pipedrive AI or custom ML models built on top of your Pipedrive data model. This is because, of course, a data schema will not populate itself with data—this responsibility falls to your team and yourself. Automated processes for data enrichment will likely handle populating these custom fields, which we will cover in later sections.
You need to ensure you have a few high-quality data pieces in place. Data quality holds a significant place in analytical and data-oriented literature, so let’s explore what certain authors say about it. According to Svolba (2012), data quality can be measured across several dimensions: data accessibility, data completeness, data correctness, and data quantity. Data accessibility refers to the ability to retrieve data for analysis purposes, which is guaranteed by default as long as all data is entered into predefined slots in Pipedrive. Data completeness means ensuring your contacts, organizations, leads, deals, and other records contain complete, accurate customer information whenever possible, minimizing information gaps. This is crucial because sometimes customers may not provide certain data points. Data correctness pertains to whether the collected data accurately reflects reality, while data quantity concerns having a sufficient number of records in your database. As mentioned in the FAQ section, to start utilizing Pipedrive AI features, you need at least 100 successfully closed deals, assuming the data related to these leads meets all the criteria for high-quality data (accessible, accurate, etc.). In your Pipedrive account, your goal shouldn’t just be to create a rich data schema but also to populate it with high-quality records. Translating these general guidelines into actionable steps means accurately reflecting in your Pipedrive CRM system what happens with deals, leads, contacts, or organizations in the real world, as we’ve covered in the previous section. However, this relationship is reciprocal. Use your CRM to record future data, such as planned actions or activities, and ensure these are executed and documented to provide comprehensive activity and engagement signals to the backend.
There are specific steps you can take to achieve efficiency in any data entry job, particularly within CRM systems like Pipedrive. Always resort to manual data entry as a last resort. In Pipedrive, there are several methods to minimize manual data entry: Firstly, utilize built-in Pipedrive automations from the automation suite. These automations can generate various data points automatically. For example, using LeadBooster’s chatbot or Pipedrive’s web forms can automate lead qualification by assigning labels based on entered data. Similarly, for deals, you can assign qualification scores or specific labels like “hot” or “cold” based on desired characteristics. Once you’ve established a rich data schema and initial data input, leverage custom fields to further automate labeling and categorization for leads and deals. However, this is just the beginning. Secondly, automate scheduled tasks, task reviews, and deal stage assignments based on specific criteria using Pipedrive automations. These automations streamline deal categorization, activity assignment (including custom activities), and pipeline progression. The second method to minimize manual intervention is the smart data search feature. This feature automatically assigns valuable data points to contacts, organizations, leads, and details with a single click. Details such as organization size or address can be automatically populated by Pipedrive. To use smart data search, enable this option in Pipedrive settings and request data enrichment from third-party sources. Finally, ensure high-quality data by integrating Pipedrive with third-party flows to enrich contact, lead, and organization profiles with additional data such as email validation. These integrations can categorize and label raw data to reduce complexity and extract essential information. Treat Pipedrive as part of your entire data ecosystem, integrating it with third-party event or data flows for additional enrichment. For instance, integrating a calling feature can add call details to contact or lead profiles, enhancing their view in Pipedrive. Additionally, explore Pipedrive’s marketplace for third-party apps that automate data enrichment, ensuring structured addition of new data points without requiring additional maintenance time from your team. Another way to enrich Pipedrive with aggregated data, beyond event-based data, is through reverse ETL (Extract, Transform, Load). Reverse ETL facilitates moving data from a warehouse to a unified customer view, providing metrics and categories not available from Pipedrive alone, such as historical deal metrics, total spending, website visits, and more. Several platforms offer reliable reverse ETL solutions. Lastly, periodically run Pipedrive’s duplicate removal tool and merge contacts to maintain data cleanliness. Consider implementing Pipedrive’s web visitor extension to provide valuable pre-sales activity information for Pipedrive AI/ML models.
Another way to automatically provide valuable data points to Pipedrive’s AI system is to connect your email inboxes and manage all communication related to deals, leads, and contacts from there. To enhance efficiency further, it would be beneficial to create response templates for different types of deals and inquiries. This allows Pipedrive AI to assess the effectiveness of emails based on specific templates. All emails sent to your Pipedrive contacts are integrated into the Pipedrive interface. Connecting your inbox to Pipedrive and enabling Pipedrive AI offers additional advantages, such as generating response emails using Pipedrive’s built-in email generator. This tool enables you to create emails of varying lengths and tones. Moreover, having all email communications directly accessible within your Pipedrive account provides summaries of interactions, which are invaluable when managing numerous deals.
For aspects that automated data processes cannot address — and there will always be data requiring manual entry, such as subjective assessments of a deal’s value or personal notes — a well-defined sales process is crucial. This process outlines how deals should be handled and when. Another critical aspect is controlling data input. When creating new custom fields, consider whether you can limit input to predefined values to maintain data quality for manually entered fields and data points where automation isn’t feasible. From Pipedrive AI’s perspective, it’s vital to ensure that manual processes accurately reflect reality at all times. For instance, strive to categorize each deal as won or lost within a reasonable timeframe. If it takes years to decide a deal’s outcome, there might be a process issue. Record every detail accurately without embellishment. Failing to accurately represent deal or lead evolution in Pipedrive, such as for the sake of presenting favorable reports, could hinder Pipedrive AI’s ability to automate your sales processes effectively. Cultural aspects are also important. Foster a culture of transparency where teams openly discuss their progress. A strong culture of openness ensures salespeople are diligent about maintaining accurate deal records without embellishment. Avoid idealizing processes; train AI based on realistic deal progressions rather than ideal scenarios. Implement automations to standardize pipeline management and ensure timely actions. Regularly update pipeline statuses, convert leads promptly, and move deals through pipelines while promptly setting final statuses as won or lost. Maximize the quality of data signals fed into Pipedrive to accurately reflect real-life deal activities and performance.
Another important element, situated between fully automated and fully manual data entry, is standardization and imputation facilitated through Pipedrive’s workflow automations. Pipedrive’s automation suite allows for automatic editing of deal, lead, or contact names upon entry to conform to predefined templates or values. For instance, you can generate the final deal name using a combination of custom fields and labels, enhancing deal clarity and promoting standardized data within your Pipedrive account. This standardization positively impacts AI model training, facilitating predictive recommendations. Having addressed the importance of configuring your account and maintaining data integrity, don’t hesitate to experiment.
Master change
The final piece of the puzzle for preparing to leverage Pipedrive AI is understanding and accepting that your approach to sales will evolve—and this evolution is perfectly normal. The key is to manage this change effectively to minimize any potentially disruptive impact on your backend AI analytical system. After all, change is the only constant, so let’s explore how to position ourselves to adapt to it. There are two types of changes to consider: structural or strategic changes and operational, low-level changes. Each type requires a different approach.
Let’s begin with operational changes. These changes encompass typical business activities as they scale, such as onboarding new users, entering new markets, expanding product offerings, and adding new activities. My well-considered recommendation in these cases is always to add new elements instead of editing existing ones whenever possible. For example, if a certain label or categorization for deals or leads is no longer used, instead of editing the label, create a new one and start using it. After some time, when the new label has accumulated significant data and provides valuable statistical information, you can delete the old label, especially if it’s no longer applied to new deals. We suggest applying this recommendation to all micro-changes in the custom field models, activities, labels, and similar elements. The primary rationale behind this approach is that editing introduces external information to specific data points that the underlying data model cannot explain, as it lacks access to your reasoning behind the change. While a well-trained predictive model may not be significantly affected by such externalities, they could potentially introduce unnecessary errors. The same principle applies to minor changes in the pipeline, such as adding new stages or adjusting probabilities. It’s preferable to implement a new option, begin using it, allow sufficient data to accumulate, and then remove the outdated or legacy configuration once it has not actively participated in deal processing for a significant period.
The second type of change, structural changes, occurs when market strategy or market conditions force a complete overhaul in how you sell, what you sell, or the overall operation of your business. Changing the business model or completely revamping the sales process often necessitates rebuilding the sales pipelines and stages. In such cases, we recommend one of two actions: either recreate your entire Pipedrive account (cleanse the legacy account and re-import relevant old deals into the new schema using the data import feature), or simply create a completely fresh account and start collecting data from scratch. The impact on your AI backend will be the same in either scenario—you will essentially reset it, requiring it to relearn the sales patterns exhibited by the deals processed in your account. If faced with the choice between resetting your account or creating a completely new one, we strongly recommend setting up a new account and starting afresh. Even if the data model has completely changed and you’re re-implementing an entirely new schema, importing old deals labeled with critical sales statuses (won/lost) will generate more statistical insights compared to mixing pre-change and post-change data.
Another important component of managing change is onboarding new sales reps. Ideally, when onboarding them, you should edit the deal distribution criteria to include the newly onboarded sales reps in your account so that they can specialize in specific types of deals. Additionally, it’s advisable to connect the new sales rep’s email inbox to Pipedrive.
Will you have to retrain your AI models whenever you change anything in your account? The models are regularly retrained anyway, so generally, no, you won’t need to do it separately. The only instance where changes might be required is in the data schemas within the Pipedrive warehousing pipeline. It’s important to remember to update these schemas as needed. Additionally, if you are also running a warehousing system alongside your Pipedrive AI, any changes implemented should prompt a review and potential adjustment of your warehousing schema to stay current. Typically, adding new custom fields may necessitate dashboard updates. Regardless of the changes made, allow some time for the new elements in the account to take effect.
The final consideration regarding changes is that account modifications should not be made impulsively but rather thoughtfully planned. We often advise clients to maintain a structured CRM maintenance process during which they gather data on necessary changes and implement them collectively. However, if feasible, it’s preferable to implement changes incrementally to isolate their statistical impact on your account and models.
This is easier than it sounds
The number of tips and recommendations may make achieving AI-readiness sound daunting. No worries, Pipedrive’s infrastructure, on which much of it relies, is rather forgiving of errors (unlike classical ML).
So let’s assume you’re ready. You’ve created a rich data schema populated by automated enrichment processes. You’ve established data quality standards and processes within your company, and your deals flow through unidirectional pipelines with clearly established stages, becoming increasingly qualified and ready for sale. You follow through on details promptly, track all activities, prepare your inbox for AI, and in short, you’ve done a great job setting up your account for AI in Pipedrive.
If this describes you, that’s fantastic. But it’s important to know that the work on your account doesn’t stop once you’ve done all these things—it actually begins. Once your smart AI assistant is in operation and Pipedrive occasionally handles minor operational details, you need to follow through and ensure you utilize the AI recommendations effectively. From taking recommended future actions on your account and observing how your deals respond, to using pre-made generated emails by Pipedrive, and verifying if the predictive activities and properties of the custom AI model accurately represent reality—these are all crucial steps. Be critical about the AI’s value added to your account and actively participate in making it better, not just by continuing what you were doing before AI implementation, but also by acting on AI-driven insights.
Acting on AI recommendations will require a mindset shift, which is normal and might be uncomfortable at first, but you will get used to it over time. Treat it as an exercise in empirical science—observe where AI-derived recommendations lead you. If something goes wrong, remember that AI is only as good as the historical representation of deal lifecycles and evolutions, based on the data accumulated in your account. Therefore, if things go awry or the quality of insights is poor, remember you can temporarily turn off AI, enrich and refine your data model, and relaunch when ready.
This article was written in collaboration with Datomni.
Leave a Reply