CRM with AI: which features help with sales and customer follow-up?

What does AI add over traditional automation?

Traditional workflow rules in a CRM system work on the basis of preset conditions: as soon as a deal gets the status 'Quotation sent', the system schedules a follow-up task for seven days later. This is deterministic process automation that works reliably for fixed steps.

Features based on generative artificial intelligence focus on unstructured data. Instead of fixed rules, a language model processes free text in emails, meeting notes and customer notes. In practice, we distinguish four main uses within sales and service teams:

• Text and email summaries: Generating a short summary of a lengthy email exchange or ticket history, so that a colleague can quickly understand the file.
• Drafting messages: Writing a draft email based on meeting notes and earlier contact moments.
• Field extraction from documents: Recognising phone numbers, contacts and action points in free text and automatically assigning them to CRM fields.
• Searching by content in natural language: Querying the CRM database without manual filters or SQL queries.

Documented AI capabilities per supplier

Different suppliers choose their own approach to integrating AI into their software. We discuss the documented specifications of three vendors.

SuperOffice: SuperNotes and productivity features

In SuperOffice, AI focuses primarily on speeding up employees' administrative tasks. According to the information on SuperOffice AI and CRM, the software helps with summarising communication, rephrasing texts and routing requests.

In the official documentation on SuperOffice prices and subscriptions, the 'SuperNotes' functionality is listed as an add-on and usage runs through a credit model. The price overview also states that advanced agent orchestration ('agent orchestration') is marked as 'coming soon'. That mention is not an available feature on which you can base a purchase. Ask the supplier to confirm the status for your subscription.

Efficy CRM: Ask Efficy and platform vision

The European Efficy CRM positions its AI features around the 'Ask Efficy' module. According to the AI product information from Efficy, this module allows users to ask questions about data within the CRM and to summarise files more quickly.

In the demonstration, ask which of the listed features are available for your edition, which are a pilot and which are only planned. Have the same fictitious customer question run in the version you can actually buy.

HubSpot: Breeze AI and sales support

HubSpot describes AI within its CRM. Use that as a starting point for the comparison and check, per feature, the required subscription, usage and language support. A general AI product page does not prove that every feature is included in every package.

An important point of attention with HubSpot is the link to subscription tiers: many advanced predictive analyses and automated suggestions are tied to the higher Professional and Enterprise tiers. It is also advisable to test the quality of the generated Dutch texts against the business communication style your team wants.

Comparative overview of AI specifications

The table below summarises the points of attention you can go through during a supplier demonstration:

Practical examples: prompt templates with fictitious data

When you use generative AI in a sales environment, it is essential to handle data carefully. Never upload sensitive personal data, confidential contracts or credit card details to public models. Use anonymised data where necessary.

The examples below illustrate how sales staff can structure prompts using fictitious company data:

Fictitious prompt for an email summary:
"Summarise the correspondence below with the fictitious company Machinefabriek De Vreese in three concrete bullet points. Specifically name the questions raised about the warranty period and the date on which the amended price proposal is expected."

Fictitious prompt for a follow-up message:
"Write a short, businesslike draft message in Dutch following a software session with the fictitious customer Transportbedrijf Helder. Thank them for the conversation, confirm that we are checking the integration with their planning package and suggest a phone call next Thursday at 14:00."

Data security, privacy and human oversight

Applying AI within a CRM database directly touches on privacy legislation (GDPR, or AVG in Dutch). After all, a CRM contains customer contacts, order amounts and meeting notes. When selecting software, pay attention to the following conditions:

• No public model training: Check contractually whether the supplier excludes customer data from training public language models.
• Location and data processing agreements: Ask where processing and storage take place and have the agreements assessed against your organisation's policy. A European supplier does not automatically mean that all processing takes place in Europe.
• Human oversight (human-in-the-loop): Make sure generated texts or proposed status changes are never applied automatically without an employee having reviewed the content. Do not use AI for automated rejection of applicants or automatic termination of agreements.

Cost structure and a vendor-neutral worked example

AI features regularly come with extra costs in the form of add-on licences, token bundles or mandatory upgrades to more expensive subscription packages. The fictitious worked example below compares the situation for a sales team of five people over a period of one month:

The example uses twenty working days per month and assumed time savings. AI can therefore pay off, provided the time saved is actually spent on valuable customer contact.

First assess one recurring task

Choose a task your team often performs, such as summarising a meeting report. Compare the result with the original notes and count the correction work too. Then have a colleague with more limited permissions ask the same question. That way you assess both usefulness and how access is handled.

Look at SuperOffice and Efficy as CRM candidates and include their concrete AI features in the same trial. The contact data, follow-up tasks and collaboration must also fit well without AI. Otherwise you are adding an assistant to a process that does not yet work well.