Skip to main content
AI strategy

Automate Salesforce with AI: optimise your sales pipeline

Automate Salesforce with AI: from conversation to filled fields, pipeline checks, competitor insight, GDPR, and Einstein, Agentforce or your own solution.

Author: Agent360 editorial team
Published: 18 December 2025
Last updated: 7 November 2026
Reading time: 14 min read

How is AI used to save time on Salesforce upkeep?

If you use Salesforce, you know the problem. The system can do almost anything, but it is useless without data, and the data comes from your most expensive employees. Salespeople are hired to talk to customers, but a large part of the week goes on updating stages, writing activities and correcting fields. After a long day, it comes down to two lines. And then the pipeline does not look like reality.

AI can take a large part of that work: write the conversation out, find the information Salesforce needs, and put it in the right fields. The salesperson sees a short draft, approves it, and moves on. The result is more time for customers, and data that is better because it is created as part of the work rather than after it.

The key points in 30 seconds:

  • Salesforce has its own AI features (including Einstein and Agentforce). They are good, but they do not suit every task, and the price depends on the licence. Always check the current terms with Salesforce.
  • The most useful tasks are often the most down-to-earth: conversation to fields, pipeline checks and competitor insight.
  • A tailored solution is built on top of your Salesforce and your other systems, without you changing anything in them.
  • Start with a small, delimited task. Build on it once it works.

What is Salesforce AI integration, and why does it matter?

Salesforce AI integration is a collective term for technology that connects AI with Salesforce, so that sales data is captured, analysed and updated automatically. Unlike manual typing, this happens through Salesforce's interfaces (APIs), transcription of conversations and language models.

There is a good reason it is often the first place people begin: pipeline analysis, forecasting and coaching are only as good as the data they are built on. If the data is incomplete, everything else is too.

Salesforce itself describes, in its annual survey State of Sales, that salespeople spend a smaller part of the working week on selling. The exact figures change from year to year, but the trend is clear: administration eats capacity.

What is the difference between Einstein, Agentforce and a tailored solution?

Salesforce has several AI tracks. They overlap, and the names change. Here is the broad difference, which helps with sorting:

Three routes to AI in Salesforce

Salesforce Einstein Salesforce Agentforce Tailored solution on top of your Salesforce
Best for Forecasts, lead and deal scoring, recommendations Agents that carry out tasks themselves in specific flows Your own processes and conversations in Danish
Strength Built into the system Can work independently within Salesforce Flexible, can connect several systems
Trade-off Closely tied to Salesforce data Requires set-up and rules Requires someone to build and maintain it
Price Depends on licence and terms (check with Salesforce) Depends on usage and terms (check with Salesforce) Depends on the task

None of the routes is "the right one". Many end up with a mix: Salesforce's own features for what they are good at, and a tailored solution for what is specific to you. The question is not which tool is best, but which bottleneck you want to solve.

From conversation to Salesforce fields: how it works

After a meeting or a telephone call, four things can happen automatically.

1. The conversation is written out. The recording is fetched from your telephone or meeting system, and speech becomes text. The quality in Danish depends on sound conditions and technical terms, so test on your own conversations.

2. AI finds the relevant information. A language model reads the transcript and extracts what you have decided should be included: a short summary, the agreed next step with a deadline, objections, competitors, budget signals and timeline.

3. The information is structured. It is placed in the right fields and objects in Salesforce (typically opportunity, account and task).

4. Salesforce is updated, or the salesperson approves. If the AI is sure, it writes directly. If it is uncertain, the field is marked for approval. That is the difference between a system that helps and one that creates a mess.

The most important gain is not only time. It is that activities which would otherwise never have been recorded now sit in the system, so everything else becomes more accurate.

Pipeline check: the lie detector that keeps you honest

One of the most concrete uses is a pipeline check. Every sales manager knows the situation: a deal sits at "negotiation" with high probability, and the salesperson is convinced. But the activity log shows there has been no contact for several weeks, the only contact is an employee without decision-making authority, and a competitor was mentioned in the last conversation without anyone noting it.

A pipeline check compares what is recorded in the stage field with what has actually happened:

  • When was the last activity?
  • Is the decision-maker involved?
  • Have promises and deadlines been kept?
  • Have competitors or objections been mentioned that were not recorded?

It is not a verdict on the salesperson, but a starting point for a conversation. If the pipeline is more optimistic than reality, you make decisions about hiring, capacity and targets on a wrong basis. Salesforce has its own deal-scoring features, but a tailored check can cross-reference data from several sources (CRM, mail, meetings, calls) and use your own rules.

Competitor insight that does not disappear

Salespeople hear about competitors all the time: "We use X today, but we are unhappy with the price." "Y has given us an offer." That is gold for product development, marketing and management, but it disappears because nobody records it.

The approach is simple:

  1. Make a list of the competitors and alternatives you meet.
  2. Let the system find mentions in the transcripts.
  3. Save them in fixed fields on the account: who, when, and what was said.
  4. Produce a weekly overview: whom are we meeting, and what are customers complaining about?

After a few months you have a picture you could never have got by asking the salespeople.

Which tasks are worth starting with?

Ordered by what typically gives the most for the least trouble

Task What it saves Complexity Start here?
Conversation to Salesforce fields Typing after each meeting and call Low to medium Yes, often the first
Pipeline check The sales manager's time on pipeline reviews Low to medium Yes, once the activities are in place
Competitor insight Manual research and lost knowledge Medium After the first two
Lead and deal scoring Prioritisation Medium After the data is in order
Automatic follow-up Forgotten follow-ups Medium to high When the rules are clear
Connection to ERP and other systems Double entry across systems High When the rest works

The order matters. If you start with the complex and skip the fundamental, you end up with an overrun budget and a disappointing result.

What does it cost, and what are the alternatives?

Prices for Salesforce's AI features change, and they depend on licence, users and usage. We therefore do not state figures here. Check the current terms with Salesforce, and make your own calculation:

  1. What does the licence cost per user if we add AI features?
  2. What does usage cost, if it is charged per use?
  3. What do implementation and operation cost internally and externally?
  4. What is the value of the time that is freed up?

For businesses with specific processes, who want to connect Salesforce to systems outside Salesforce (for example a Danish accounting system or an in-house database), or who need a solution tailored to Danish conversations, a tailored solution is often more flexible. For businesses with standard processes, Salesforce's own features may be enough.

How do you ensure GDPR compliance?

It is not optional. Recording and analysing conversations is processing of personal data, and it requires:

  • A lawful basis. For example consent or legitimate interest, assessed concretely.
  • Information. The customer must know that the call is recorded and analysed (GDPR Article 13), and it is good manners to say if AI is involved.
  • Data minimisation. Only what is necessary is stored, and irrelevant information is filtered out before it lands in Salesforce.
  • Erasure. Data about a person must be findable and deletable (Article 17), including transcripts.
  • Processors and location. Where does the processing happen, and by whom? Are there agreements in place (Article 28), and does data leave the EU/EEA? Salesforce offers its own solutions for regional data, but that does not automatically apply to the tools you connect.

Get a legal assessment of your specific setup before anything goes into operation.

How do AI and salespeople complement each other?

The sales organisation of the future is not AI instead of salespeople, but AI and salespeople in one system, where each does what they are best at.

AI is good at:

  • Never forgetting to record an activity
  • Finding patterns across many agreements
  • Filling in fields consistently
  • Reminding people of deadlines
  • Pointing to agreements that seem to be stalling

People are better at:

  • Creating trust and relationships
  • Understanding how a customer's business fits together
  • Navigating politics in large organisations
  • Negotiating creative solutions
  • Reading the mood

The point is that AI does not make salespeople better by doing what they already do. It makes them better by taking the tasks they never should have had.

Technical overview: Salesforce and the APIs

If you want to understand or build an integration yourselves, these are the most important building blocks. (Always check Salesforce's own documentation for the current API version.)

  • Authentication. Use a "Connected App" with OAuth 2.0 (for server-to-server typically the JWT Bearer Flow). Avoid usernames and passwords in operation, and grant only the permissions the integration needs.
  • Reading and writing. Salesforce's REST API can read and update objects such as opportunities, accounts and tasks, and SOQL is used to retrieve data. Batch calls can reduce the number of requests.
  • Limits. Salesforce has limits on the number of API calls per day. Plan for it, so an integration does not run out in the middle of the day.
  • Security. Limit access to the objects and fields that are needed, use audit logging, and build error handling so that a failure does not leave half-finished updates behind.
  • Middleware. For large connections between several systems there are integration tools (for example MuleSoft). They are powerful but require resources. For simple connections there are lighter tools, which do not always have the same error handling.

How do you implement it, step by step?

Step 1: Data foundation. Look at your Salesforce data. How many fields are empty? How many duplicates? What should be updated automatically, and what requires human approval?

Step 2: Access and security. Create an integration with minimal permissions, and set up logging. Clarify access to telephony, calendar and mail.

Step 3: Conversation to fields, at small scale. Start with one team and one type of conversation. Check the quality, and adjust the rules until the results can be relied on.

Step 4: Pipeline check and competitor insight. Build on the data, which is now better.

Step 5: Expand. Add more sources (mail, meetings), and connect other systems when it makes sense.

Change management is just as important as the technology. Salespeople who know that the system removes their most tedious tasks, rather than monitoring them, are the ones who get the most out of it.

What can Agent360 do?

Agent360 does not sell a finished Salesforce product. We build solutions in your own systems, as part of Your AI department, and if it is possible in your Salesforce, we start there:

  • We first look at how your sales and your Salesforce are used today, and which licence and access you have.
  • We finish one thing properly before we move on. Examples from the use cases page: a meeting recorder that writes the meeting out and turns the promises into tasks with an owner, and a sales cockpit where the conversation is written out in Danish, AI suggests the outcome and drafts a mail for the salesperson to approve.
  • You decide what the system may do on its own. Everything else waits for approval.
  • If we lack access to a system, we contact the supplier for you and help get it.
  • The price depends on the task, and you receive a quote before you commit.

If you want to learn to set up AI and build with it yourself, there is the personal AI mentoring programme: three individual meetings, fixed price, no subscription. Read about the programme.

FAQ: Salesforce and AI

Q: Can AI automatically update Salesforce after a call? Yes, in principle. The conversation is written out, the information is found and placed in the right fields, and the salesperson approves or corrects it. How much can happen without approval depends on how well the rules have been tested on your conversations.

Q: What do Salesforce's AI features cost? It depends on the licence, number of users and usage, and the prices change. Check the current terms with Salesforce, and make a calculation that also covers implementation and operation.

Q: What is the difference between Einstein and Agentforce? Roughly speaking, Einstein is a collection of predictive features (scoring, recommendations), while Agentforce is Salesforce's solution for agents that carry out tasks themselves within specific flows. Names and packages change, so check with Salesforce what applies right now.

Q: Is AI integration with Salesforce GDPR-compliant? It can be, but it requires active choices: a lawful basis, information, data minimisation, erasure and clear agreements with processors. Get a legal assessment of the set-up.

Q: What is a pipeline check? It is a review of all open deals, where the stage recorded is compared with the actual activities: the latest contact, whom you are talking to, and whether promises have been kept. The purpose is a more realistic picture of the pipeline.

Q: Should we use MuleSoft or a lighter tool for the integration? If the integration touches critical systems and large volumes of data, an integration tool may be right. For simple, small connections, lighter tools are often enough. Assess based on criticality, data volume and error handling.

Q: How long does it take to get started? It depends on data quality, access and scope. A delimited first task can go into operation faster than a broad integration. The most important factor is often not the technology, but the data quality and getting the team on board.

Conclusion

Salesforce is only as good as the data in it. In many sales departments this is not a technical problem but a capacity problem: nobody has time to type. AI can take a large part of that, if it is set up with clear frameworks and with a person who approves what is important.

Would you like to hear what it could look like in your Salesforce?

Book a free 20-minute consultation. Tell us where the time disappears in your CRM, and get an honest view of whether it is worth building, and what it would require.

Related articles:

Should your CRM keep itself up to date?

We build the layer that moves data into the CRM you already use, without anyone having to type it in. We look at your current set-up first.

How we built it · CRM & customer overview

Customer CRM that updates itself

An internal page per customer with contacts, the agreement, the projects and a timeline where emails, meetings, email drafts and tasks land by themselves. Above the pages sits the board with all customers, owner and latest contact, and the system suggests next steps based on fixed rules.

  1. Email, meeting, task
  2. Lands on the customer’s page
  3. You take the next step
See the customer’s own page →