<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Polat Aydın on Medium]]></title>
        <description><![CDATA[Stories by Polat Aydın on Medium]]></description>
        <link>https://medium.com/@polataydin?source=rss-4cc783c23b74------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*OqvQq_OdmStFqHGYiZo-wA.png</url>
            <title>Stories by Polat Aydın on Medium</title>
            <link>https://medium.com/@polataydin?source=rss-4cc783c23b74------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 01:05:20 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@polataydin/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Rebuilding the routing diagnostics screen Microsoft retired]]></title>
            <link>https://medium.com/@polataydin/rebuilding-the-routing-diagnostics-screen-microsoft-retired-4c8b8c5d8315?source=rss-4cc783c23b74------2</link>
            <guid isPermaLink="false">https://medium.com/p/4c8b8c5d8315</guid>
            <category><![CDATA[powerplatform]]></category>
            <category><![CDATA[dynamics-365]]></category>
            <category><![CDATA[pcf]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[customer-service]]></category>
            <dc:creator><![CDATA[Polat Aydın]]></dc:creator>
            <pubDate>Wed, 07 Oct 2026 19:38:06 GMT</pubDate>
            <atom:updated>2026-10-07T19:45:54.369Z</atom:updated>
            <content:encoded><![CDATA[<p>A PCF control that tells you what unified routing actually did with a record, and the five decisions that shaped it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3vLD6AYFF63EkxMEqNpZYg.png" /></figure><p>The lifecycle of a routed case: the runs, the sessions behind them and where the record sits now.</p><p>Unified routing in Dynamics 365 Customer Service is a black box with one small window. Microsoft used to ship a diagnostics screen that let you see inside it. That screen is gone, and what replaced it does not answer the question anyone actually asks: this case went to the wrong queue, or nobody picked it up for two days, so what happened?</p><p>Answering that today means opening several tables by hand, then writing a KQL query against Application Insights, then lining the two up by eye. I have done that enough times to build the thing instead. It is a PCF control, it is MIT licensed, and the link is at the bottom.</p><p>This is a write-up of what the data looks like and the decisions that mattered, because most of them were not obvious to me when I started.</p><p>Where the answer actually lives</p><p>Routing writes its story across four Dataverse tables and one Azure resource.</p><p>msdyn_ocliveworkitem is the routing attempt. One row per time the record entered unified routing. A case routed, resolved, reopened and routed again has two rows, and they are unrelated to each other beyond pointing at the same record.</p><p>msdyn_ocsession is the offer. A session exists only once an agent has been chosen, so the absence of a session is itself the answer to “why is nobody working this”: nobody was ever notified.</p><p>msdyn_sessionparticipant holds the acceptance. A participant row with addedon set and joinedon empty is an offer that timed out, and it is the only place that distinction is recorded.</p><p>queueitem says where the record sits now.</p><p>And Application Insights holds everything about the decision itself: which classification rules ran, which route to queue rules matched, what the assignment attempt was looking for, and why it found nobody. None of that is in Dataverse at all.</p><p>Decision 1: do not trust the field that looks like the answer</p><p>msdyn_statuschangereason reads like exactly what you want. It says things like “AwaitingAgentAcceptance” in plain English.</p><p>It is a trap. The platform stamps it at the last status change and never revises it. In record routing with push distribution, it gets written when the work is handed to an agent and nothing is written after that. So a case someone has owned for a week still reads “AwaitingAgentAcceptance”, and a control that shows it is confidently reporting the opposite of what happened.</p><p>The control does not read it for record routing. The state is derived from the columns the platform does maintain: open with no active agent means still queued; an active agent with no acceptance means offered; accepted and not closed means being worked. It is more code, and it is right.</p><p>Decision 2: the browser cannot query Application Insights, so do not try</p><p>A PCF control has context.webAPI and nothing else. There is no credential it can use against Azure Monitor, and there should not be.</p><p>So the control does not query Application Insights. It writes a row.</p><p>It creates a record in a diagnosticrequest table with the conversation id and a time window. A Dataverse-triggered cloud flow picks that row up, runs the KQL against the Application Insights resource named by three environment variables, and writes the results back as child event rows. The control polls the request row and renders the events when it reaches a terminal status.</p><p>The Azure credential lives in a connection reference, which is where it belongs. The user needs no Azure access at all. The cost is latency and a row per trace, so results are cached per routing attempt and asking again is a deliberate click.</p><p>Decision 3: never compute a date</p><p>This one cost me the most time, and the lesson is embarrassingly simple.</p><p>Dataverse stores datetimes in UTC. The platform displays them in the time zone on the user’s personal settings, which is frequently not the browser’s. If your control formats dates itself, it disagrees with the audit history sitting in the next tab, and the person reading both has no way to tell which one is lying.</p><p>I tried to resolve the offset from the PCF user settings. The host returned the browser’s zone. I tried deriving it by formatting a known instant through the platform formatter and reading the hour back. That worked until the platform prefixed the date, my parser read “10/02” as 10:02, and every timestamp on screen shifted by ten hours and two minutes.</p><p>The fix was to stop computing. Every Dataverse datetime comes back twice: the raw UTC value, and the platform’s own rendering under the FormattedValue annotation. The control reads that annotation and prints it unchanged. The raw value is used only for sorting and for measuring durations, where the time zone plays no part.</p><p>Do not reimplement something the platform already hands you.</p><p>Decision 4: a shared timeline was the wrong picture</p><p>The first version drew every routing attempt on one shared time axis, each bar at its true position. It looked right on paper.</p><p>Then a real case arrived: routed on 17 September, closed on the 23rd, routed again on 5 October. Eighteen days of axis, twelve of them a blank band in the middle that carried no information at all, and the second run squeezed into a sliver one percent wide with its labels clipped to a single letter.</p><p>The axis was answering a question nobody asks. Nobody wonders how far apart two routings were in pixels; they wonder where the time went inside a run. So each run now gets its own full-width bar, split into time in queue, time offered and time worked, each segment as wide as its own words with the remaining space shared by duration. A phase that took no time is still readable, and the one that took 137 hours is still visibly the long one.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DnzpsAmHmvo6ajR5B1J47w.png" /></figure><p>Each run gets its own bar: time in queue, time offered, time worked.</p><p>Decision 5: resolve the ids</p><p>Application Insights logs guids and nothing else. The queue is a guid. The workstream is a guid. The required skill is a guid, and the assignment rule set that ran is a guid.</p><p>A diagnostics screen that shows you 0e0914f1-f95f-f111-a826–002248c89e76 where the answer is “Billing” has not actually told you anything. The control collects every id in a trace, looks the names up in Dataverse, and makes each one a link that opens the record. A table the user cannot read is not an error; the id simply stays on screen as written.</p><p>The same applies to the agent who was picked. The trace names them and logs an empty skill list, so the skills are read from the schedule tables and folded in. Set against the criteria the assignment was looking for, that is usually the whole explanation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NyB55PexCCwlz7595s_OYQ.png" /></figure><p>The routing trace, stage by stage. Open one and you see the rules, the criteria and the agent who was picked.</p><p>Installing it</p><p>Import the managed solution, fill in three environment variables for the Application Insights resource, connect Dataverse and Azure Monitor Logs, turn on the flow, then put the custom page on a sitemap.</p><p>That last step is all the setup there is. The control reads Dataverse itself rather than a bound column, so it does not go on a form or a view, and it covers every routed record and conversation in the environment at once. Cases, emails, conversations on any channel, custom tables.</p><p>Diagnostics has to be switched on for unified routing, otherwise Application Insights has nothing to read.</p><p>Code, screenshots and the solution are here:</p><p><a href="https://github.com/polataydin/pcf-routing-diagnostics">https://github.com/polataydin/pcf-routing-diagnostics</a></p><p>If you use it and something is wrong or missing, please open an issue.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4c8b8c5d8315" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Triggering Customer Insights — Journeys Custom Triggers from C#]]></title>
            <link>https://medium.com/@polataydin/triggering-customer-insights-journeys-custom-triggers-from-c-0a24a0fe50b6?source=rss-4cc783c23b74------2</link>
            <guid isPermaLink="false">https://medium.com/p/0a24a0fe50b6</guid>
            <category><![CDATA[customer-insight]]></category>
            <category><![CDATA[dynamics-365]]></category>
            <dc:creator><![CDATA[Polat Aydın]]></dc:creator>
            <pubDate>Wed, 22 Apr 2026 09:28:48 GMT</pubDate>
            <atom:updated>2026-04-22T09:29:12.472Z</atom:updated>
            <content:encoded><![CDATA[<p>When you create a custom trigger in Dynamics 365 Customer Insights — Journeys and click <strong>Go to code snippet</strong>, Microsoft gives you a JavaScript snippet for a website integration. That works for browser-originated events. For anything server-side — a plugin reacting to a Dataverse change, an Azure Function receiving a webhook from a payment provider, a background worker processing an IoT queue — you need to call the trigger from C#.</p><p>The official documentation doesn’t spell this out end-to-end. It mentions in passing that a custom trigger is a Dataverse Custom API, but if you’ve never wired one up from C#, you’ll spend a few hours figuring out which parameters are required, where the request name comes from, and why your trigger fires but the journey never advances.</p><p>This post is a focused walkthrough of the C# integration. It assumes you’ve already created the custom trigger in the Customer Insights — Journeys UI and defined its attributes. Everything below is about what happens on the server side after that.</p><p><strong>What a custom trigger actually is</strong></p><pre>msdynmkt_purchasecompleted_103112499</pre><p>The msdynmkt_ prefix is fixed. The middle section is derived from the trigger name you chose. The numeric suffix is a unique ID the platform assigns.</p><p>Because it’s a standard unbound custom API, anything that can call IOrganizationService.Execute() — a plugin, a custom workflow activity, a console app, an Azure Function using ServiceClient, a web job — can raise it. No special SDK, no ingestion key, no HTTPS endpoint to manage.</p><p>The implication is worth stating clearly: <strong>a custom trigger is just an </strong><strong>OrganizationRequest with a specific name and a specific set of parameters.</strong> Once you see it that way, the rest is plumbing</p><h3>Finding the RequestName</h3><p>You cannot guess the RequestName and you cannot hardcode it from the trigger’s display name. You have to retrieve it from the trigger itself. Two ways:</p><p><strong>From the UI:</strong> open the trigger in Customer Insights — Journeys, click <strong>Go to code snippet</strong>, and copy the function name from the first line of the generated JavaScript. It’s the msdynmkt_* string.</p><p><strong>From Dataverse:</strong> query the customapi table directly. Every custom trigger you create in Customer Insights — Journeys generates a row here; filter by uniquename LIKE &#39;msdynmkt_%&#39; and the uniquename field is your RequestName.</p><p>Store it in configuration, not in source code. The numeric suffix differs between environments — the RequestName in your dev sandbox is <strong>not</strong> the same as the one in UAT or production, because the trigger is recreated in each environment with a new ID. I’ve lost more time than I want to admit on plugins that worked in dev and failed silently in UAT for exactly this reason.</p><h3>The required system parameters</h3><p>Every custom trigger invocation needs a set of system parameters in addition to whatever custom attributes you defined. Miss any of these and the trigger either fails outright or fires into the void:</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/f226f424add6dc64eec5f6ceaf4f8002/href">https://medium.com/media/f226f424add6dc64eec5f6ceaf4f8002/href</a></iframe><p>The timestamps must be UTC. If you pass local time, the journey’s time-based branches and wait conditions will compute incorrectly. On a server running in a non-UTC time zone this is easy to get wrong.</p><p><strong>On </strong><strong>bindingid:</strong> if your trigger is an exit or branching signal — &quot;cancel the appointment&quot; or &quot;finish the course&quot; — and the contact could reasonably be in multiple instances of the same journey, you need a bindingid that uniquely identifies which instance this signal belongs to. The format follows the Dataverse row reference pattern (tablename(guid)), for example appointments(b6a8...). Without it, a cancellation on appointment #2 will also cancel the journey instance tied to appointment #1.</p><h3>Custom attributes</h3><p>The attributes you defined when creating the trigger (e.g., ExamCode, VoucherExpiry, EnrollmentOrder as an entity reference) also show up as parameters with the msdynmkt_ prefix. You add them to the OrganizationRequest.Parameters collection the same way you&#39;d add any other action input.</p><p>One thing to remember, again: <strong>any attribute with a null value will cause the trigger to fail at journey runtime.</strong> The failure mode is silent — the trigger call itself succeeds, but the journey logs a System Failure and the contact never progresses. Always default nullable fields to a sentinel value before adding them to the parameters collection.</p><h3>Calling from a plugin</h3><p>This is the most common server-side scenario: a Dataverse event occurs (an order is paid, a case is resolved, a field changes), and a plugin raises a custom trigger to kick off a journey with the full business context attached.</p><pre>using System;<br>using Microsoft.Xrm.Sdk;<br><br>public class OrderPaidPlugin : IPlugin<br>{<br>    private const string TriggerRequestName =<br>        &quot;msdynmkt_orderpaid_103112499&quot;; // load from config<br><br>    public void Execute(IServiceProvider serviceProvider)<br>    {<br>        var context = (IPluginExecutionContext)serviceProvider<br>            .GetService(typeof(IPluginExecutionContext));<br>        var factory = (IOrganizationServiceFactory)serviceProvider<br>            .GetService(typeof(IOrganizationServiceFactory));<br>        var service = factory.CreateOrganizationService(context.UserId);<br><br>        if (!context.InputParameters.Contains(&quot;Target&quot;)<br>            || !(context.InputParameters[&quot;Target&quot;] is Entity order))<br>            return;<br><br>        var contactRef = order.GetAttributeValue&lt;EntityReference&gt;(&quot;customerid&quot;);<br>        if (contactRef == null) return;<br><br>        var now = DateTime.UtcNow;<br><br>        var request = new OrganizationRequest(TriggerRequestName);<br>        request.Parameters.Add(&quot;msdynmkt_signaluserauthid&quot;, contactRef.Id.ToString());<br>        request.Parameters.Add(&quot;msdynmkt_profileid&quot;, contactRef.Id.ToString());<br>        request.Parameters.Add(&quot;msdynmkt_signaltimestamp&quot;, now);<br>        request.Parameters.Add(&quot;msdynmkt_signalingestiontimestamp&quot;, now);<br>        request.Parameters.Add(&quot;msdynmkt_bindingid&quot;, $&quot;orders({order.Id})&quot;);<br><br>        // Custom attributes defined on the trigger<br>        request.Parameters.Add(&quot;msdynmkt_ordertotal&quot;,<br>            order.GetAttributeValue&lt;Money&gt;(&quot;totalamount&quot;)?.Value ?? 0m);<br>        request.Parameters.Add(&quot;msdynmkt_productcode&quot;,<br>            order.GetAttributeValue&lt;string&gt;(&quot;productcode&quot;) ?? &quot;unknown&quot;);<br><br>        service.Execute(request);<br>    }<br>}</pre><p>A few production notes on this pattern:</p><p><strong>Register the step as post-operation, asynchronous.</strong> Synchronous execution blocks the originating transaction on the journey platform’s response time. Async keeps the user-facing operation fast and the trigger dispatch isolated.</p><p><strong>Don’t raise from pre-operation.</strong> The record whose creation or update is raising the trigger may not yet be queryable by downstream services, and if the outer transaction rolls back you’ve dispatched a signal for an event that never happened.</p><p><strong>Guard against infinite loops.</strong> If your journey eventually writes back to the same table that triggers your plugin (through a Power Automate flow or another action), you need a guard condition — an inLoop check or a depth counter — to prevent recursive triggers.</p><h3>Calling from an external .NET application</h3><p>For external callers — an Azure Function, a worker service, a console utility — use ServiceClient from the Microsoft.PowerPlatform.Dataverse.Client package. The invocation pattern is identical once you have an IOrganizationService.</p><pre>using Microsoft.PowerPlatform.Dataverse.Client;<br>using Microsoft.Xrm.Sdk;<br><br>var connectionString =<br>    &quot;AuthType=ClientSecret;&quot; +<br>    &quot;Url=https://yourorg.crm4.dynamics.com;&quot; +<br>    &quot;ClientId={appId};ClientSecret={secret};&quot;;<br><br>using var service = new ServiceClient(connectionString);<br>if (!service.IsReady) throw new InvalidOperationException(service.LastError);<br><br>var now = DateTime.UtcNow;<br>var contactId = Guid.Parse(&quot;b6a8...&quot;);<br><br>var request = new OrganizationRequest(&quot;msdynmkt_orderpaid_103112499&quot;);<br>request.Parameters.Add(&quot;msdynmkt_signaluserauthid&quot;, contactId.ToString());<br>request.Parameters.Add(&quot;msdynmkt_profileid&quot;, contactId.ToString());<br>request.Parameters.Add(&quot;msdynmkt_signaltimestamp&quot;, now);<br>request.Parameters.Add(&quot;msdynmkt_signalingestiontimestamp&quot;, now);<br>request.Parameters.Add(&quot;msdynmkt_bindingid&quot;, contactId.ToString());<br>request.Parameters.Add(&quot;msdynmkt_ordertotal&quot;, 499.00m);<br>request.Parameters.Add(&quot;msdynmkt_productcode&quot;, &quot;ITIL-FOUND-2025&quot;);<br><br>service.Execute(request);</pre><p>The application registration used here needs the standard Dataverse user permissions plus access to the custom API. In practice, give the app user the same security role you’d give a system integration account — and no more.</p><h3>Batching with ExecuteMultipleRequest</h3><p>If you’re processing a batch — replaying a day of missed events, migrating historical data into journeys — don’t call Execute in a loop. Use ExecuteMultipleRequest:</p><pre>var executeMultiple = new ExecuteMultipleRequest<br>{<br>    Settings = new ExecuteMultipleSettings<br>    {<br>        ContinueOnError = true,<br>        ReturnResponses = true<br>    },<br>    Requests = new OrganizationRequestCollection()<br>};<br><br>foreach (var order in ordersToReplay)<br>{<br>    var request = BuildTriggerRequest(order); // same as above<br>    executeMultiple.Requests.Add(request);<br>}<br><br>var response = (ExecuteMultipleResponse)service.Execute(executeMultiple);<br><br>foreach (var item in response.Responses)<br>{<br>    if (item.Fault != null)<br>    {<br>        // log the fault, the request index is item.RequestIndex<br>    }<br>}</pre><p>The default batch size limit is 1,000 requests per ExecuteMultipleRequest. For larger replays, chunk the input.</p><h3>Gotchas specific to C# invocation</h3><p>A short list of things that burned time for me or people I’ve reviewed code for:</p><p><strong>Environment-specific RequestName.</strong> Covered above, but worth repeating — never hardcode the msdynmkt_* name. Pull it from config, keyed by environment.</p><p><strong>UTC timestamps.</strong> The platform treats the timestamps as UTC. DateTime.Now on a server in Athens or Istanbul will produce wrong time-based branch behavior. Always DateTime.UtcNow</p><p><strong>Silent null failures.</strong> The C# call succeeds even when a parameter is null. The journey fails silently. Wrap your parameter assembly in a defensive block that substitutes sentinels for nulls and logs which substitution happened.</p><p><strong>bindingId format.</strong> For row-based binding, the format is tablename(guid) with the guid in plain form (no braces). Get this wrong and the trigger either binds to nothing or binds too broadly.</p><p><strong>Permissions.</strong> A plugin running in the calling user’s context will only raise the trigger if that user has execute permission on the custom API. In practice, this means app users and system accounts need the appropriate security role; end users usually don’t, which is a feature, not a bug.</p><p><strong>Don’t call from pre-validation or pre-operation stages.</strong> The transaction isn’t committed, downstream services don’t see the record yet, and a rollback on the outer operation won’t roll back the trigger dispatch.</p><p><strong>Test the trigger dispatch in isolation before wiring up the journey.</strong> Create a minimal test flow that listens for the trigger and writes to a log table. Fire the trigger from C# and confirm the log row appears with all expected attributes. Only then attach the real journey.</p><h3>Closing</h3><p>The C# path to custom triggers is simpler than the documentation makes it look — it’s just a Custom API call with a handful of required parameters — but the silent failure modes are unforgiving. Timestamps must be UTC, attributes cannot be null, the RequestName differs per environment, and bindingId either isolates your trigger to the right journey instance or lets it fire across every instance the contact is in.</p><p>If you get those four things right, the rest is just plumbing the signal through from wherever your business event originates to the Execute call. And once you&#39;ve done it from a plugin, you&#39;ve done it from everywhere — the pattern is identical across plugins, Azure Functions, worker services, and console tools.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0a24a0fe50b6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Skill-Based Routing in D365 Omnichannel: Lessons from the Field]]></title>
            <link>https://medium.com/@polataydin/skill-based-routing-in-d365-omnichannel-lessons-from-the-field-8aa976fa108f?source=rss-4cc783c23b74------2</link>
            <guid isPermaLink="false">https://medium.com/p/8aa976fa108f</guid>
            <category><![CDATA[management-skills]]></category>
            <category><![CDATA[omnichannel]]></category>
            <category><![CDATA[d365]]></category>
            <category><![CDATA[microsoft-dynamics]]></category>
            <category><![CDATA[customer-service]]></category>
            <dc:creator><![CDATA[Polat Aydın]]></dc:creator>
            <pubDate>Tue, 21 Apr 2026 16:28:00 GMT</pubDate>
            <atom:updated>2026-04-21T17:02:45.088Z</atom:updated>
            <content:encoded><![CDATA[<h3>Skill-Based Routing in D365 CRM Omnichannel: Lessons from the Field</h3><p>Skill-Based Routing (SBR) in Dynamics 365 Omnichannel looks straightforward on paper. You create skills, assign them to agents, attach them to work items, and the routing engine does the rest. In practice, the gap between a working demo and a production-ready implementation is significant. This article shares the design decisions, pitfalls, and operational realities I&#39;ve encountered across enterprise deployments.</p><p><strong>The Architecture You’re Actually Building</strong></p><p>Before touching any configuration, it helps to understand what you’re wiring together. SBR is not a standalone feature — it sits across three interconnected subsystems:</p><p><strong>1. Unified Routing Engine</strong> — evaluates incoming work items against assignment rules and skill requirements before placing them in a queue.</p><p><strong>2. Bookable Resource Characteristics (BRC)</strong> — the data model where skills are actually stored. Skills don’t live directly on agent records; they live as BRC records in the Universal Resource Scheduling (URS) layer, linked to Bookable Resources.</p><p><strong>3. Skill Attachment Rules</strong> — determine which skills get tagged onto a <em>work item</em> at intake. They don’t assign agents. They describe what the work item needs.</p><p>The end-to-end flow:</p><p>Incoming work item<br> → Triggers Work Stream<br> → Skill Attachment Rules evaluated → skills tagged to work item<br> → Unified Routing matches work item skills to agent BRC skills<br> → Exact Match or Closest Match logic applied<br> → Work item placed in queue / assigned to agent</p><p>Most configuration errors happen because people conflate skill tagging (step 3) with agent assignment (step 4). Keep them separate in your mental model — it will save you hours of troubleshooting.</p><p><strong>Real-World Scenario: Global Certification Support</strong></p><p>Throughout this article I’ll use a realistic enterprise scenario. The company operates globally and provides certification exams. Customers contact support via Chat and Voice with the following routing requirements:</p><ul><li>Technical exam issues → Technical Support agents with product-specific skill</li><li>Registration issues → Operations agents</li><li>Complaints and escalations → Senior agents with Escalation Handling + language match</li><li>VIP customers → Priority queue regardless of topic</li></ul><p>Let’s architect this properly, step by step.</p><h3>Step 1: Skill Taxonomy Design — Do This Before Touching the System</h3><p>The most skipped step. Before creating a single skill record, design your taxonomy offline.</p><p>Skills in Omnichannel fall into two implicit categories — the system doesn’t enforce this distinction, but your design should:</p><p><strong>Functional skills</strong> (what the agent can do):</p><ul><li>FUNC_Technical Support - Exam Platform</li><li>FUNC_Candidate Registration</li><li>FUNC_Escalation Handling</li><li>FUNC_Billing and Payments</li></ul><p><strong>Language skills</strong> (what language the agent can serve):</p><ul><li>LANG_English</li><li>LANG_Greek</li><li>LANG_German</li></ul><p>Using a prefix convention (FUNC_, LANG_) becomes essential once you have 30+ skills. Without it, skill lists become unmanageable and attachment rules become error-prone.</p><p><strong>Why the separation matters:</strong> A work item from a German-speaking customer with a technical issue needs an agent with <em>both</em> FUNC_Technical Support AND LANG_German. If you treat these as a single combined skill, your taxonomy explodes. If you don&#39;t separate them at all, you&#39;ll route German customers to English-only agents.</p><p><strong>Rating model design:</strong> The default rating model ships with five values: Poor (1), Fair (2), Good (3), Very Good (4), and Excellent (5). For functional skills, this granularity is useful — you may want only Good or above on standard cases, and Excellent reserved for complex escalations. For language skills, ratings are effectively binary — an agent either covers a language or they don&#39;t. A practical approach: default all language BRC assignments to Excellent and require Excellent in your language attachment rules, keeping the threshold unambiguous.</p><p><strong>Rating model design:</strong> The default rating model includes values like Beginner, Proficient, and Expert. For functional skills, this is useful — you may want only Proficient or above on complex technical cases. For language skills, ratings are effectively binary. A practical approach: create a dedicated rating model for languages with a single value (Fluent), or default all language BRC assignments to Expert and require Expert in all language rules.</p><h3>Step 2: Enabling Skill-Based Routing</h3><p>Navigate to <strong>Customer(Copilot) Service admin center → Routing → Skill-based routing</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bgdQjUpymIloweh7cGE8Kw.png" /></figure><p>Toggle SBR on. Once enabled, the <strong>Rating Model</strong> section becomes active. Review the default model and decide whether to keep it or create custom models per your taxonomy design above.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kjPlk5QrowVsyR53qQoNUw.png" /></figure><h3>Step 3: Creating Skills</h3><p>Navigate to <strong>Customer(Copilot) Service admin center → User management → Skills</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IGbdJxoKNt1AS1iNz1dQ3Q.png" /></figure><p>Create each skill from your taxonomy. For each record, set:</p><ul><li><strong>Name</strong> — follow your naming convention exactly</li><li><strong>Type</strong> — Skill or Certification. Use Certification only for credentials that are time-bound or externally issued. Use Skill for everything else.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TBCniJsUSwmIsywfE4pXYg.png" /></figure><h3>Step 4: Assigning Skills to Agents via Bookable Resource Characteristics</h3><p>This is where most implementations hit their first real problem.</p><p>Skills are assigned to agents by creating <strong>Bookable Resource Characteristic (BRC)</strong> records — not by editing the agent’s User record directly. Each BRC links a Bookable Resource (the agent) to a Characteristic (the skill) with a Rating Value.</p><p>From the Skill record, go to the <strong>Agents</strong> subgrid and add a new BRC:</p><p>Skill record with Agents subgrid open, showing the “Add Bookable Resource Characteristic” side panel with Bookable Resource lookup and Rating Value dropdown.</p><p>Saved BRC record showing agent name, skill name, and rating value (e.g., “Expert”).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xg7U7aqdeSm6Vp5CWfV7Dw.png" /></figure><p><strong>Critical: Agents must have a Bookable Resource record to receive skill assignments.</strong> If an agent was provisioned via Entra ID sync but doesn’t have a corresponding Bookable Resource, they will never appear in skill-based routing assignment — silently. This is a common gap in greenfield implementations, especially with bulk user imports. Always audit Bookable Resources after user provisioning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oHWAQVKQpOvztd0j76B0KA.png" /></figure><p><strong>Step 5: Configuring Skill Attachment Rules in the Work Stream</strong></p><p>Open your Work Stream in <strong>Customer (Copilot) Service admin center → Workstreams</strong>, then navigate to the <strong>Work classification</strong> section.</p><p>Under <strong>Work classification</strong>, create a ruleset and add rules inside it. Each rule evaluates conditions on the incoming Case record and attaches skills as output. We’re using out-of-the-box Case fields — no custom fields or context variables required.(If workstream is based on chat then you can use context variables)</p><p>I assume that you have already created a workstream for your implementation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FmI2GFCk1M1XFSORUQHrUg.png" /></figure><p><strong>Rule: Technical Issue</strong></p><ul><li>Condition: Routed record (Case) → Case Type equals Problem</li><li>Output skill: FUNC_Technical Support - Exam Platform, Rating: Expert</li></ul><p><strong>Rule: General Inquiry</strong></p><ul><li>Condition: Routed record (Case) → Case Type equals Question</li><li>Output skill: FUNC_Escalation Handling, Rating: Good</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Z2ySbH2LxRZJHKEjm5HCUA.png" /></figure><p><strong>Step 6: Exact Match vs. Closest Match — The Decision That Defines Your Queue Design</strong></p><p>This is the most consequential configuration choice in any SBR implementation.</p><p>Set this in the Work Stream under <strong>Work classification → Matching logic</strong>:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/679/1*d03fc3cb-lFw8OCZ20sl7A.png" /></figure><h3>Exact Match</h3><p>The routing engine assigns a work item only to an agent who has <strong>all</strong> required skills at the required rating or above. If no such agent is available, the work item waits.</p><p><strong>Use for:</strong> Compliance-sensitive scenarios, specialized technical skills where the wrong agent handling the case is not acceptable, escalation handling where only senior agents should be assigned.</p><p><strong>Operational risk:</strong> Queue pile-up. If your skill supply is too narrow or agent availability drops, work items age indefinitely. This becomes a workforce management problem that SBR makes visible — which is actually useful, but needs to be managed proactively.</p><h3>Closest Match</h3><p>The engine scores available agents by how many required skills they match and at what rating. It prefers a full match but will assign a partial match if no exact match exists.</p><p><strong>Use for:</strong> General support scenarios where any capable agent is better than waiting, transitional phases where skill coverage is still being built out.</p><p><strong>Operational risk:</strong> Mismatches that are invisible in standard reporting. A work item assigned via closest match may have gone to an agent missing a key skill — and you won’t know unless you build explicit reporting on it.</p><p><strong>Important architectural constraint:</strong> You can only set one matching mode per Work Stream. If you need Exact Match for language and Closest Match for functional skills on the same channel, you need separate Work Streams routed to separate queues. This has real implications for your queue architecture and reporting design.</p><p><strong>Step 7: Testing Your Routing Logic</strong></p><p>Never test SBR with real agents in a production environment and never rely on manual observation alone.</p><p><strong>Use dedicated test agent accounts:</strong> Create users with specific, controlled skill sets. Testing with real agents whose availability you can’t control produces unreliable results.</p><p><strong>Use Conversation Diagnostics:</strong> Within any conversation record, the Diagnostics panel shows which classification rules fired, which skills were attached, and the reason for the routing engine’s assignment decision. This is your primary debugging tool.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2t-1mSuBU4MHfDIowmKszw.png" /></figure><p><strong>Timing gotcha:</strong> Skill changes are cached by the routing engine. After modifying a BRC record, allow a few minutes before retesting. In some cases, toggling the agent’s presence (offline → online) forces a cache refresh. Never assume a null result immediately after a config change means the configuration is wrong.</p><h3>Step 8: Operational Governance — Keeping Skills in Sync</h3><p>Once you go live, the hardest challenge is not the initial configuration — it’s keeping agent skills accurate as your workforce changes.</p><h3>Skills Hub — Managing at Scale</h3><p>Skills, skill types, proficiency scales, and intelligent skill finder models are all important parts of skill-based routing. The skills hub is the one-stop place to manage these attributes across your entire call center. It provides an overview of all configured skills and the number of agents associated with each one — making it easier to spot skills with no agents assigned, which directly causes routing failures.</p><h3>Data Sharing Across Apps</h3><p>One operational consideration that catches teams off guard: the data such as skills, rating model, and rating value that is created, updated, and used in Dynamics 365 Contact Center is shared with other model-driven apps (Field Service and Project Service Automation) installed in your environment. If your organization uses Field Service or PSA alongside Customer Service, skill records are shared — changes in one app affect the others. Coordinate with those teams before modifying shared rating models or skill definitions.</p><p>Source: <a href="https://learn.microsoft.com/en-us/dynamics365/customer-service/administer/overview-skill-work-distribution">Overview of skill-based routing — Microsoft Learn</a></p><h3>Closing Thoughts</h3><p>Skill-Based Routing is one of the most powerful capabilities in Dynamics 365 Omnichannel. The system will route something almost all of the time — the real question is whether it’s routing the right work to the right agent, and whether you have the observability to know when it isn’t.</p><p>Design the taxonomy before opening the admin center. Understand what Bookable Resources are and where they sit in the data model. Choose your matching mode deliberately. And build for operational maintainability from day one — because the people who will maintain this configuration after you leave are not architects.</p><p>Tags: Dynamics365, Omnichannel, CustomerService, MicrosoftDynamics, PowerPlatform, SolutionArchitect, UnifiedRouting</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8aa976fa108f" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>