The artificial intelligence in CRM examples on this page are the ones that write something useful into the contact record without anyone typing it. On the levelpartner.ai platform that means three working builds: a call transcript that starts a workflow the moment the call ends, an AI Summarize step that turns that transcript into a short note on a contact field, and a duplicate clean-up that makes sure the summary lands on the right person. Each example below says what to build, why it pays back, and where it breaks. The test for every one is the same: does the output end up where a colleague already looks?
Why AI that writes into the record beats AI on a separate screen
Most CRM AI demos show a new panel: a prediction, a chat window, a dashboard of insights. In a 60-person firm those panels get opened for a fortnight and then ignored, because nobody's job includes checking them. AI that writes a field, a note or a tag is different. The account manager opening the contact tomorrow sees it without changing habits.
There is a second reason. Every later automation reads fields. A routing rule, a report or a follow-up message can use "last call summary" once it exists as a value. It cannot use a paragraph sitting in someone's head. So the examples here start with getting good text into the record, and only then use it.
Example 1: a finished call starts a workflow
The Transcript Generated trigger fires when the platform has finished transcribing a call. It works for calls handled by an AI voice agent, calls through the IVR and normal calls on platform numbers, once transcription is switched on for that call type. Voice agent calls are transcribed by default. Ordinary phone and IVR calls need transcription enabled in their settings, and the trigger links you there if it is off.
This is the starting point for everything else, because it turns a conversation into an event the CRM can act on. Before it, a call was a line in the activity log. After it, a call can write a summary, create a task, move a deal or alert a manager.
- Go to Automation, then Workflows, and click Create Workflow.
- Choose Start from Scratch, or open an existing workflow.
- Click Add New Trigger and select Transcript Generated from the trigger list.
- Turn on transcription for the call type you want, using the link in the trigger settings if it is off.
- Set the filters: Call Type (required), and optionally Call Direction and a minimum Call Duration.
- Click Save Trigger, add your workflow actions, and switch the workflow from Draft to Publish.
Set a minimum call duration. Without it, voicemails, wrong numbers and dropped calls each start the workflow and fill records with summaries of nothing. A floor of around half a minute removes most of them.
Only calls recorded and transcribed inside the platform fire this trigger. Calls imported or synced from an outside phone system do not. If a transcript fails entirely, the trigger does not fire; if it is partial, the workflow runs and the missing parts are blank.
Example 2: AI Summarize writes the call note
A full transcript is too long to read before a callback. The AI Summarize workflow action takes any text you give it, such as the transcript value from the trigger, and returns a summary at the length you set: a number of words, characters, sentences or paragraphs. The result is stored as a value you can use in the next step, so you can write it to a contact custom field, put it in an internal notification or send it by email.
- Search for AI Summarize in the workflow actions menu and add it to the workflow.
- Enter an Action Name, for example "Summarise sales call".
- Set Max Length Input to the summary length you want, such as three sentences.
- Under Input Text, choose a Dynamic Variable and pick the call transcript from the custom value picker.
- Click Save.
Then add an update-contact step that writes the summary into a field such as "Last call summary". Keep the field near the top of the contact view so people actually see it.
Say the firm takes around 300 client and prospect calls a month and roughly a third of them get any note at all. With the trigger set to normal calls over 30 seconds and a three-sentence summary written to "Last call summary", every qualifying call would leave a note, which is about 200 more notes a month. An account manager covering a colleague's client would read three sentences instead of asking for a handover.
Getting summaries people trust
The model decides what matters, and you cannot tell it to leave things out. What you can control is the input and the length. A shorter limit forces the summary to the agreed next step and the date; a longer one drifts into small talk. You can chain two summarise steps, a long one to a note and a short one to a field, if both are useful.
Test on a handful of real calls before you publish. Summaries are reliable on the main points and less reliable on exact numbers, names and dates. Where a figure matters, such as a quoted fee, the person on the call confirms it.
Example 3: duplicate clean-up before AI touches the data
The two examples above write onto a contact. If the same client exists twice, once from a web form under their work email and once from a phone enquiry under their mobile, the summary lands on one record and the next person opens the other. AI makes the duplicate problem worse, because it creates confident notes in the wrong place.
So the third example is not AI at all. It is the clean-up that makes the AI examples work. The platform's Manage Duplicates tool groups contacts that share an email, phone number or name and lets an admin merge them.
- Go to Contacts.
- Click the three-dot menu.
- Click Manage Duplicates.
- Choose a Find By option: Email, Phone or Name.
- Click Find Duplicates and wait for the groups of matches.
- Click the down arrow beside a group to see every contact in it.
- Click Review to compare the conflicting fields side by side.
- For each conflicting field, select the value to keep.
- Combine tags or other details as needed.
- Choose the primary email and primary phone number.
- Confirm that no duplicate opportunities will be created.
- Select up to 10 records to merge.
- Choose one record as the Master.
- Click Merge Contacts.
- In the confirmation dialog, tick the acknowledgement box.
- Click Merge Duplicates.
Merges are permanent. Only admins can use the tool, which is the right control: one person who knows the client base approves each group. On a custom field, a value from another record only fills the master's field if the master's field is empty, so pick the master with the most complete history.
To stop the problem returning, check the duplicate setting under the business profile's contact preferences. With duplicates disallowed, a form or calendar submission that matches an existing email or phone updates that contact instead of creating a new one.
How the three examples fit together
Run them in the reverse of the order you read them. Clean the duplicates first, so there is one record per client. Then publish the transcript workflow with summaries going to a field. After a month, you have a field on most active contacts that later automations can read.
That is where the more ambitious uses start to work. A follow-up message can open with the agreed next step. A manager can filter contacts by calls that mention a complaint. An AI agent answering a later chat has the context of the last call. None of that is possible while the notes are missing or split across two records. For what happens when AI makes a decision inside a process rather than writing a note, see AI as a step in a business process.
Examples that look good and usually fail
AI deal forecasts on small pipelines
A firm closing a few hundred deals a year does not have the volume or the consistent data a forecast model needs. Stages get skipped, close reasons are missing, and the model learns noise. It produces a confident number that is no better than the sales director's view and harder to argue with. Clean stages and a simple weighted pipeline will tell you more.
Automatic merges
Some teams want the platform to merge duplicates on its own. A merge that joins two different people with the same name destroys both histories, and it cannot be reversed. Keep merging as a reviewed task, done in batches of up to ten records at a time.
Summaries of everything
Summarising every email, chat and call creates more text than anyone reads. Start with calls, where nothing else captures what was said, and add other channels only where a person asks for it.
Publishing the summary workflow on all call types at once with no duration filter. The first week fills hundreds of records with notes about voicemail greetings, and people learn to ignore the field. Start with one call type and a duration floor, read a sample, then widen it.
What these examples replace
They replace the note-taking a rep skips when the next call is waiting, the handover email when a colleague is on leave, and the half day a quarter someone spends merging contacts by hand. They do not replace the conversation itself or the judgement about what to do next. For drafting the next message from what the record holds, see AI follow up, and for the sales side of the same idea, the AI sales assistant guide. If you want an agent that answers chat and calls and updates fields as it goes, start with setting up AI agents.
Troubleshooting
The workflow never fires after a call
Check that transcription is on for that call type and that the call went through a platform number. Calls from an outside phone system are not transcribed and do not count. Then check the Call Type filter matches the call.
Summaries appear on the wrong contact
The call was matched to a duplicate record. Run Manage Duplicates by phone, since calls match on number, and merge the pair.
The custom field cuts the transcript off
Custom fields hold transcripts only up to a limited size. Store the full transcript in a note and keep the field for the short summary.
Two workflows send the same message
More than one workflow uses the same transcript event with overlapping filters. Give each a distinct filter, such as call direction, or merge them into one workflow with branches.
Frequently asked questions
Which artificial intelligence in CRM example should we build first?
Clean duplicates, then build the call transcript workflow with AI Summarize. It needs no change in how people work, and it creates the text every later AI step depends on.
Does AI Summarize cost extra?
It is a premium workflow action billed per run as usage, at cost. Filtering out short calls keeps the number of runs to the calls that matter.
Can we summarise calls from our existing phone system?
Not with this trigger. Only calls recorded and transcribed inside the platform start it, so the calls need to run through platform numbers or an AI voice agent.
Can a team member merge duplicates?
No. Manage Duplicates is limited to admins, which keeps irreversible merges with someone who knows the client base.