Contents
AI is useful in a CRM to summarize an account’s history, prepare a meeting, structure a dictated call report and suggest follow-ups that the sales rep approves. A first use case on an existing CRM starts at $2,300 (€2,000)excl. VAT; in a custom CRM, the AI building block counts for $2,800 (€2,500)excl. VAT.
A CRM holds a wealth of information that sales reps never have time to reread: emails, quotes, complaints, visit notes. AI can surface what matters at the right moment and cut down the data entry that so many teams resist. It can also create a lot of noise if you add it without a clear goal. Here are the use cases I recommend, what they cost, and the ones I steer clear of.
Use cases that actually help
I sort AI use cases in a CRM with one simple question: does the rep save time, and can they easily check the result?
- Summarizing an account before a meeting. One paragraph: recent orders, open quotes, complaints, points raised in recent conversations. The rep shows up prepared without rereading six months of history.
- Turning a dictated note into a structured call report. After a visit, the rep dictates two minutes on their phone; the AI produces the report, the next steps and the follow-up date. This is often the use case that most improves CRM data quality.
- Suggesting a follow-up draft tailored to the history, which the rep edits and sends.
- Flagging accounts to watch: customers whose orders are slowing down, quotes with no reply, repeated complaints. The AI flags, the manager decides.
- Cleaning up and completing records: detecting likely duplicates, standardizing job titles or industries.
Use cases, data and level of oversight
| Use case | Data needed | Risk of error | Human review |
|---|---|---|---|
| Account summary | Conversation history, orders, tickets | Low, easy to check | Read by the rep |
| Dictated call report | Voice note or free text | Low to medium | Reviewed before saving |
| Follow-up draft | History, latest quote | Medium (tone, commitments) | Always before sending |
| Account alerts | Orders, quotes, complaints | Medium (false signals) | Manager’s call |
| Deduplication | Contact and company records | Medium (wrong merges) | Merges approved |
| Predictive scoring | Rich history of won and lost deals | High if history is thin | Essential |
What it costs
These amounts exclude VAT and come from my pricing.
| Situation | What is delivered | Indicative budget | Timeline |
|---|---|---|---|
| AI connected to your current CRM (through its API) | A first use case: account summary or dictated call report | From $2,300 (€2,000)excl. VAT | 2 to 8 weeks |
| Custom CRM with built-in AI | Focused CRM + AI building block | About $9,600 to $16,300 (€8,500–€14,500)excl. VAT | 1 to 3 months |
| CRM connected to ERP and invoicing, with AI | Complete CRM + integrations + AI | $22,400 (€20,000)excl. VAT and up | 3 to 6 months |
The middle row is a focused CRM ($6,800 to $13,500 (€6,000–€12,000)excl. VAT, see how much a custom CRM costs) plus the built-in AI block from the estimator, at $2,800 (€2,500)excl. VAT. AI API usage comes on top, billed by the provider. I do not put a number on productivity gains upfront: they are measured during the prototype.
What it takes to work
- A CRM that is actually filled in. AI summarizes what exists. If important conversations happen outside the CRM, the summary will be incomplete.
- One use case at a time. Start with one, measure adoption, then add another.
- Controlled access: the AI only sees what the logged-in rep is allowed to see.
- GDPR compliance: customer data sent out is limited to what is necessary, and the processing is documented. The page on what data to share with AI covers the precautions.
When it is not the right solution
- Your team barely enters anything in the CRM: the real project is getting the tool adopted, not AI. A simple mobile app, such as a mobile CRM for field sales, may need to fix that first.
- Your current CRM’s AI features meet your needs after a real trial.
- You want predictive scoring with little history: explicit rules, discussed with the sales team, will do better.
- Volume is low: three reps who know all their customers do not need automatic summaries.
Where to start
- Ask your reps which CRM task they dislike most: it is often the visit report.
- Check your CRM’s API, or consider a custom CRM if the current tool is already blocking other needs.
- Run a one-week audit and a prototype on one use case, tested by two or three reps.
The full approach, with human review kept in place, is described on the AI integration page.
Frequently asked questions
Our off-the-shelf CRM already has AI features. Should we build something else?
Test what exists first, on your real accounts, for a few weeks. Custom development makes sense when the AI has to combine data your CRM cannot see (ERP, invoicing, support tickets) or follow your own sales rules.
Can the AI send follow-ups to customers on its own?
Technically yes, but I advise against it at first. The AI suggests a draft tailored to the history, and the rep reviews and sends it. Only simple, low-stakes messages get fully automated, once quality has been measured.
Is AI lead scoring reliable?
It depends on how much sales history you have and how good it is. With few well-documented won and lost deals, an AI-generated score gives a false sense of precision. Simple, explicit rules are often more useful at first.
Is customer data sent to the AI protected?
What is sent is limited to the necessary fields, through the business API of a provider whose terms exclude training on your data. Under GDPR, this processing of personal data needs to be documented, for example in your record of processing activities; your DPO or lawyer can tell you what applies in your case.