Present content clearly

Use language that is easy to understand and appropriate for a wide audience.

Use a natural, conversational style, so the system isn’t mechanical or robotic.

“Don’t save the transaction.”

“Do you want to cancel it?”

“Don’t save the transaction.”

“If you want to cancel this transaction, say or enter cancel.”

Use short, simple words that users can easily understand.

“Plan business trip to China.”

“You’ll need a visa to travel to China. Here’s the link to apply:”

“Plan business trip to China.”

“OK, this is the travel requirement for China.”

Avoid jargon and industry-specific terms.

“Have we paid for the laptops that were just delivered?”

“No, not yet. Would you like to send the payment now?”

“Have we paid for the laptops that were just delivered?”

“No, not yet. Would you like to remit the payment now?”

Never show system codes to the user.

“Add last night’s dinner to the expense report.”

“OK, I’ve added “Hotel Dinner” to your expense report.”

“Add last night’s dinner to the expense report.”

“OK, I’ve successfully itemized the $HOTEL1 expense.”

Use contractions for a friendlier, more natural tone.

“What is my hotel reservation?”

“I’ve found your hotel information. It’s in your ‘Business trips’ folder.”

“What is my hotel reservation.”

“I have found your hotel information. It is in your ‘Business trips’ folder.”

Avoid colloquialisms and slang words or phrases. They may not be understood by all users and can become outdated.

“Start purchase request.”

“I’m having some trouble with that request. Try again in a few minutes.”

“Start purchase request.”

“I’m in a bit of a pickle. Try again in a few minutes.”

Spell out acronyms for the first time.

“My child is sick and I’ll need to take the next two weeks off.”

“You can take up to 12 weeks under the Family and Medical Leave Act (FMLA).”

“My child is sick and I’ll need to take the next two weeks off.”

“You can take up to 12 weeks under FMLA.”

Use second person pronouns and active voice.

“Submit my expense report.”

“Please itemize all your credit card expenses before you submit your expense report.”

“Submit my expense report.”

“All credit card expenses should be itemized before the expense report is submitted.”

Provide responses that apply to the user’s context. Don’t diverge from the user’s intent.

“Create a purchase request for 2 laptops”

“Sure, which laptop do you want?”

“Create a purchase request for 2 laptops”

“Are the laptops for you?”

Avoid language that suggests partiality, judgement, or motives beyond helping the user.

“What’s the weather like in Palo Alto today?”

“It’s sunny with a high of 75°F today.”

“What’s the weather like in Palo Alto today?”

“It’s a very gorgeous afternoon with a high of 75°F today—perfect for taking a walk!”

Focus on the user and the task at hand. Don’t brag or focus on the system.

“Travel arrangements for my trip tomorrow.”

“You’re all set. Your flight’s booked and you can pick up your car from the rental agency. Your hotel confirmation is in your inbox.

Do you need help with anything else?”

“Travel arrangements for my trip tomorrow.”

“I’ve made all your travel arrangements. I booked your flight and I got a good deal on the rental car. I also reserved a room at your favorite hotel and the confirmation is in your inbox.

Do you need my help with anything else?”

Don’t use emojis. They’re too informal, and may suggest that the system has emotions – which it does not.

“Was my leave request approved?“

“Yes, the approval is in your inbox.”

“Was my leave request approved?“

“Yes, the approval is in your inbox 🙂 “

Use humor sparingly. You can use positive humor to deflect personal questions. If you use humor, make it general and not focused on the user or the system.

“Submit expense report.”

“I couldn’t submit the expense report because it didn’t have a cost center. Is it ABC-123?”

“Submit expense report.”

“Uh oh, looks like your expense report couldn’t be submitted. My bad!”

Set the user up for success

Evaluate the current context and analyze what could happen to create awareness of potential situations.

“Add 3 widgets to the order.”

“The items are out of stock until next Thursday. I’ll save your order as a draft, and notify you when they’re in stock.”

“Add 3 widgets to the order.”

“You’re trying to add items that are out of stock. The purchase order will not be created.”

Proactively provide relevant and meaningful information.

“Is there a direct flight to Frankfurt on Friday?”

“There’s a direct flight to Frankfurt (FRA) this Friday for $50 more. You’ll arrive at 7:00AM CET instead of 1:00PM CET.”

“Is there a direct flight to Frankfurt on Friday?”

“I see your current flight has 2 layovers. For only $50 more, you can take a direct flight to Frankfurt.”

Summarize data and information into concrete, consumable insights.

“Are there new sales leads this week?”

“Yes, here are the 3 top leads.”

“Are there new sales leads this week?”

“You have a few new sales leads. Would you like to see them?”

Recommendations provide predictive or available options. Use data to support recommendations, and prompt users with next steps.

“Find a gasket supplier in Georgia.”

Sure, there are 2 gasket suppliers in Georgia: ABC and CDE.

They’re both certified suppliers, but CDE has a higher performance score. Here’s the supplier profile of CDE:

“Find a gasket supplier in Georgia.”

Sure, there are 2 gasket suppliers in Georgia: ABC and CDE.

CDE is better. Here’s the supplier profile of CDE:

Clarify the request of the user when it’s ambiguous, or when the system doesn’t have enough information to complete a task.

“Update the sales order.”

“Which sales order do you want to update?”

“Update the sales order.”

“Please be more specific about which sales order you want to update.”

Recognize the context of the conversation as well as prior inputs and requests.

“When’s my next meeting?”

“You have a meeting at 2 PM today to discuss ‘Sales order with Avantel’.”

“Change the meeting time to 1 PM.”

“OK, I’ve changed it.”

“When’s my next meeting?”

“You have a meeting at 2 PM today to discuss ‘Sales order with Avantel’.”

“Change the meeting time to 1 PM.”

“Which meeting would you like me to change?”

The system should be able to forgive spelling errors and still understand the user’s intent.

“Wen’s my next meet?”

“You have a meeting at 2 PM today to discuss ‘Sales order with Avantel’.”

“Wen’s my next meet?”

“‘I didn’t get that. Can you say that in different words?”

Help users fix errors and make changes.

I ordered the wrong valve pump.

No problem, you have 5 days to change or cancel the order. Which valve pump did you want?

I ordered the wrong valve pump.

No problem, you have 5 days to change or cancel the order.

Adapt to the user’s context

The system should understand actions and behaviors that the user repeats, and suggest automation in recurring behavior patterns.

“Start weekly status report.”

“Sure, I’ve updated the sourcing events for this week. Would you like me to update them every week?”

“Start weekly status report.”

“Sure, how can I help?”

The system learns the phrases that users say and incorporates them into its behavior.

“Add ‘Avantel’ as a biz prtnr.”

“Do you mean ‘business partner’?”

“OK. I’ve added Avantel to your “Business Partners” contact list.”

“Add “Avantel” as a biz prtnr.”

“Sorry, but I don’t understand what you mean by ‘Biz prtnr’.”

The system should understand frequency and types of use, and eventually make information easier for users to access.

“Start expense report for last week.”

“Sure, here’s your expense report with your receipts from last week.”

“Start expense report for last week.”

“OK, how can I help?”

Be transparent

Provide context. Connect multiple points of data to tell a story or explain reasoning behind a statement.

“How are our sales this quarter?”

“Sales are down 10% due to a heavy storm on the East Coast.”

“How are our sales this quarter?”

“Sales are down 10%.”

Help users learn the capabilities and limitations of the system. Otherwise, they may continue to try actions that the system can’t do, and eventually stop using it.

“Delete the onboarding program.”

“I can’t delete it, but I can send a request to HR to delete it.”

“Delete the onboarding program.”

“Sorry, I can’t do that.”

Caution users about an action or entry that might cause an error or an unexpected or undesirable result. Provide feedback on the results of a specific action, especially if it can’t be undone.

“Delete this object.”

“Sure, but this object also appears in other instances, which may break them.”

“Delete this object.”

“OK, I deleted it.”

Cite sources. Provide a link to sources so users can reach the same conclusions on their own.

“What is SAP’s DA?”

“SAP DA is a digital assistant and bot integration hub for the enterprise.”

“What is SAP DA?”

Turn your conversations into actions with SAP DA. Complete all your enterprise tasks from one place, with voice or text. Get your work done wherever you are, whether on your mobile device or desktop, or even in popular channels, such as Slack.

• Cloud deployment
• Conversational user experience
• Single integration point for all your bots
• Extensibility with additional skills and capabilities

Would you like to hear more?

Avoid words that make it seem like the system has preferences

“Do I have any meetings this afternoon?”

“You have a Q2 preparation meeting with John and Bill at 2 PM.”

“Do I have any meetings this afternoon?”

“You have an important Q2 preparation meeting with John and Bill at 2 PM.”

Avoid extremes. Keep emotions neutral.

“I am taking a leave of absence for my wedding.”

“OK. Let me know which days you want off.”

“I am taking a leave of absence for my wedding.”

“Splendid!!! Go party!”

Keep errors emotionless.

“Submit the leave request.”

“I can’t seem to submit it right now. I’ll save it, and try to submit it later.”

“Submit the leave request.”

“Oh no! I’m really sorry, I can’t submit it.”

Acknowledge the user

Recognize the user’s tone and adjust responses accordingly.

“Submit my bid.”

“The auction just ended.”

“Nooooo!!!!!!”

“Would you like to email it to the seller, in case they accept it?”

“Submit my bid.”

“There was a problem submitting your bid. Please try again.”

“Nooooo!!!!!!” 

“There was a problem submitting your bid. Please try again.”

Affirm that the system understands the user’s request.

“Download the latest Excel file titled ‘Sales order with Avantel’ and delete the first two products on the list.”

“OK. I’ve download the most recent version of ‘Sales order with Avantel’ and deleted ‘HT-1001’ and ‘Astro phone 5’.”

“Download the latest Excel file titled ‘Sales order with Avantel’ and delete the first two products on the list.”

“OK. You have ‘Sales order with Avantel v2’ last modified 1:30PM today and ‘Sales order with Avantel’ v3 last modified 5PM today. Which one would you like me to modify?”

Adapt to sensitive information by confirming the user’s intent.

How much was my sales commission last month?

Would you like me to:

How much was my sales commission last month?

Your sales commission was $5000.

Be approachable

Be considerate of the user’s situation.

Remind me again tomorrow.

OK, I’ll remind you at 9 AM tomorrow.

Remind me again tomorrow.

OK, I’ll remind you at the same time tomorrow.

Be friendly, pleasant and polite towards the user.

Hi Jessica, how can I help?

What do you want?

Keep a professional distance. Avoid small talk or personal conversations.

“How are you?”

“Fine, thanks for asking. How can I help you?”

“How are you?”

“Fine, thanks for asking. How are you doing?”