Transcript:
0:02 [Music]
0:05 Are you tired of AI agents giving you
0:07 different answers to the same question?
0:09 Today, we're exploring how North52's
0:11 decision suite can bring deterministic,
0:13 100% consistent results to your
0:15 Microsoft database AI agent. Let's dive
0:18 into how we do this with XRM formula
0:21 sample 294. Here's the challenge. AI
0:25 large language models are inherently
0:27 probabilistic. They're fantastic for
0:29 creative tasks, but when you need fast,
0:31 consistent, accurate results for
0:33 business critical decisions, that
0:35 uncertainty becomes a problem. But what
0:38 if you could combine the power of AI
0:40 with the reliability of deterministic
0:42 business rules? That's exactly what
0:44 we're going to show you today.
0:46 Deterministic rules are essential for
0:48 your database AI agent for four key
0:50 reasons. Compliance. Ensure regulatory
0:54 requirements are always met. No
0:56 exceptions.
0:57 Data quality. Maintain consistent data
1:00 validation and formatting every single
1:02 time.
1:03 Business logic. Apply your
1:05 organization's specific rules without
1:07 any variation. And reliability. Remove
1:11 uncertainty from critical decision
1:12 points that could impact your business.
1:14 Let's look at a real example. Imagine
1:17 you're processing credit card
1:18 applications. You have specific
1:20 eligibility rules for platinum, gold,
1:22 and silver cards based on annual income,
1:25 total assets, and credit scores.
1:27 With traditional AI, you might get
1:29 different recommendations for identical
1:31 applications. With North52's
1:33 deterministic rules, every customer with
1:35 the same financial profile gets exactly
1:37 the same result every single time.
1:41 Here's how the North52 solution works
1:43 in three simple steps. First, your
1:46 co-pilot agent is set up with a tool
1:48 that calls a custom action. Second, a
1:51 formula runs on that custom action,
1:53 evaluating eligibility based on the
1:55 values provided by the agent. Finally,
1:57 the system returns a definitive result,
2:00 whether they're eligible, what card type
2:02 they qualify for, and their exact credit
2:04 limit. Let's build this step by step.
2:07 First, we create a global unbound action
2:10 with specific parameters. For inputs, we
2:13 need annual income, total assets, and
2:16 credit score all as required integers.
2:21 For outputs, we define result as a
2:24 required boolean plus optional credit
2:27 limit and card type fields.
2:30 Next, we create our formula in the North
2:32 52 app using the decision suite.
2:34 Navigate to business process activities,
2:37 then formulas, and then create a new
2:39 formula. Set your source entity to N52
2:42 command formula type to action and
2:45 select the decision table editor. Name
2:48 it action hyphen credit card eligibility
2:51 and link it to your custom action. The
2:54 magic happens in the decision suite's
2:55 decision tables. We'll create three
2:58 separate sheets, one each for platinum,
3:01 gold, and silver card eligibility. Each
3:04 table defines exact thresholds for
3:06 income, assets, and credit score.
3:09 Finally, we connect everything to your
3:11 co-pilot agent. Add a new tool. Select
3:14 Microsoft data versse. Then perform an
3:17 unbound action in selected environment.
3:20 Configure your connection. Provide a
3:22 clear name and description so the agent
3:24 can identify easily. Then map your input
3:27 parameters. Make sure to enable should
3:31 prompt user so that the agent asks for
3:33 the required information when it's
3:35 missing. Now for the exciting part,
3:38 testing. Your agent can now handle both
3:40 natural language requests and simple
3:42 responses. Whether someone asks, "Can I
3:45 get a credit card with 80K income, 200k
3:48 assets, and 750 credit score?" or just
3:51 provides the numbers, your agent will
3:53 use the deterministic rules to provide
3:55 exactly the same answer every time. The
3:58 AI handles the conversation naturally,
4:01 but the business decisions are rock
4:02 solid reliable. Remember these key
4:05 points. Test your rules thoroughly. Use
4:07 clear parameter names and document your
4:10 decision logic. The beauty of this
4:11 approach is that you get the
4:12 conversational intelligence of AI
4:14 combined with the reliability of
4:16 business rules. That's how you implement
4:18 deterministic rules for your Microsoft
4:20 database AI agent. Using North52's
4:23 decision suite, your agents will now
4:25 provide consistent, reliable results
4:27 while maintaining their AI powered
4:29 conversational abilities.
4:32 If you're ready to try this yourself,
4:34 check out the full knowledge article
4:36 link in the description. And don't
4:37 forget to subscribe for more North52
4:39 tutorials.