We’re a hangover cure. Yes, really.

In the US $9.8 billion will be spent on Black Friday.  Retailers will discount up to 38% to encourage purchases, but then will pay the price of returns. Retailers are making it easier to make impulse purchases, but also are making it easier for customers to return. 42% of consumers have regretted at least one of their Black Friday purchases. 16% of fashion purchases will be returned, but this climbs to 23% for GenZ and 24% for Millenials. Retailers listened to shoppers’ feedback from past years and were more transparent regarding their discounts and return policies. This reduced the returns.

Executives are realizing that you can’t spell holidays without AI. Retailers are embracing this innovative technology to personalize shopping experiences and increase customer profitability during the holiday season and beyond

Rob Garf, VP and GM of Retail, Salesforce

You have 4 days ‘til Black Friday to make sure your returns policies are consistent. You could easily get an agent up and running to be able to answer policy and purchasing questions by then.  Failing that, you have a week until Returns Tuesday to get that agent up and running.

Too soon?  Not really. 

You have time if you follow this proven, governed process. This is a process we use internally to build complex agents in less than a day. It will be explained in detail in our new ebook Ultimate Guide to Creating Agents (coming soon)

Live by Black Friday

Design

Map process

It all starts with understanding the existing returns processes and working out what can be agentified. I described the overall process (below) into my iPhone and then got Elements AI to draw the UPN process diagram.

We have been promoting the whole notion of design thinking and process thinking. When it comes to AI, having the mindset of process pattern thinking, I think it’s going to be really critical, especially as we are held accountable to having ROI at the end of the day with improving these technologies to improve processes.

Juan Perez, CIO of Salesforce

Here is my description of the process. I just opened up my iPhone and started talking. Here is what I said. It is not complete or 100% correct, but it is a great starting point.

The process starts with the customer hitting an agent asking for a product return. 

The agent needs to determine whether the return is for a replacement for cash,  an exchange, or a return and replacement for warranty purposes 

To do that the agent needs to establish who the customer is and what the order was so they can the agent can look at the policies for that particular product and order to determine the action

Based on the policy, there could be 5 possible outcomes each drawn as a separate set of process boxes

1. Policy says no return

2. Policy says cash refund but could offer a credit if the product is returned

3. Cash refund if the product is returned

4. Warranty issue and therefore the product needs to be returned and a replacement of the same part is sent out 

5. Replacement for exchange. So need to check whether that is in stock before it can be sent, 

Ask customers who want to return – in store or shipping?

If in store, provide the nearest store that accepts returns

If shipping

– Send RMO, 

– confirming who pays for shipping based on policy

– confirm return receipt

Once product is received (instore) or shipped and received, then the follow-up action will be initiated

This is the UPN diagram that AI automatically geneerated.

Refine the process

When you think through the process it is slightly more nuanced. AI did a great job getting me started. So I refined it based on my understanding of the process. I worked out which was a user, when it was the agent talking to the user, when the agent was taking action, and when it was a human. Looking at the process I decided where Actions were required, and where the agent was working with the user and the agent would need Instructions.

Write instructions and actions

I used AI to look at the diagram and write the Instructions and Actions. For each Action I asked it to recommend the metadata type and data required. Is it correct? Not 100%. But it is a great start. These are stored against each step on the process diagram so that they are version-controlled because they will change over time. I also got AI to look at the UPN diagram and write the first draft of the Topic Classification Description that Agentforce’s Atlas Reasoning Engine will use to decide whether to call the Topic. And also the Topic Scope that the Atlas Reasoning Engine uses to determine the guardrails.

INSTRUCTIONS CREATED

ACTIONS CREATED

Validate Actions and Instructions

The next step is to look through the AI-generated AI Actions and Instructions and refine them.  What is missing? What guardrail instructions are needed based on company policies that are not in the process?

Create user stories for each Action

Then I needed to write a user story to build the metadata for each Action. Elements AI automatically generates user stories with acceptance criteria and suggested metadata implementation saving me a ton of time. The user stories sync with Jira as we use Jira internally. But it could be whatever ticketing system you use.

Build agent

Build

Creating the agent is simple, and this returns process is delivered by a Topic as part of a Customer Service Agent. So you build the Topic by copy/paste the Topic info, Instructions and Actions using Agent Builder.

For each Action we also need to build a metadata item, or find an existing metadata item.  Elements AI will look through your org for existing metadata, but the results depend on how good your descriptions are. We also need to upload the policies into Agentforce knowledge which is exposed as a Data Cloud DMO so that they are visible to Prompt Template. 

Train/test agent

Now it is live we need to start throwing some test scenarios at it. As you refine it make sure that you track the changes to your Instructions. If these are updated on the process map you have version control.

Monitor

Once live you need to monitor the results and fine-tune the agent.

What are you waiting for?

You know that Black Friday is coming.

You know that more transparent policies will reduce returns. 

You also know that Returns Tuesday is coming.

You also know you don’t have the staffing to handle the peak for Friday and Tuesday.

This is the perfect use case for an agent. And today is only Monday!!!!