[Customer service & support]() / [Use cases]() / [Agent assist]() [How it works]() [What you get]() [Scope]() [Start a review]() [Customer service & support]()[Knowledge]() # Give your team source-linked answers and the next step. While the customer waits, your agent sees the answer, the policy it came from and the next action, and decides what to send. [Start with a support review]() [How it works]() My jacket arrived 26 days ago. Can I still return it? Yes. You’re inside our 30-day return window, with four days left. I can send a return label now, or an exchange if you’d like a different size. RET-13, effective Oct 15 · suggested reply, the agent approves What the agent sees: a suggested reply with the policy it came from. They approve, edit or ignore it. ## Sound familiar? - Agents search three places while the customer waits. - Newer agents ask the same colleague, and that colleague falls behind. - The right answer depends on which agent you got. Time spent hunting for the answer is time not spent on the customer, and new agents learn from whoever they ask. ## What changes. ShowTodayWith the engine Finding the answer Search three tools Shown with its source Next step The agent works it out Suggested, with the order and policy attached New agents Shadow a veteran Coached from the same sources everyone uses Wrong answers Found by a complaint Found when agents edit the suggestion ## How it works. 5 steps, 4 stages. - AI**2** - Rules and automation**1** - Your team**2** - Approval 1. 1 Read 1. Rules and automation ### Pull the order, history and policy The context the agent would have gathered, in one view. - Shopify - Gorgias 2. 2 Suggest 1. AI ### Draft the answer and the next step Written from the sources in force, each one linked. 3. 3 Send 1. Your team Approval ### Approve, edit or replace The agent decides what goes to the customer. 4. 4 Learn 1. AI ### Group the edits by cause Wrong source, missing fact or a coaching moment. 2. Your team ### Fix the source or coach Your CX lead acts on the pattern each week. Agent edits show where a source is wrong ## What you get. Real work, every week. What the agent sees Order O-8402 Signal Source Value Delivered Shopify 26 days ago Return window RET-13 30 days Contacts this month Helpdesk None Suggested next step Policy Offer a label or an exchange The context behind the suggestion: the order, the policy and the next step, in one view. Edited suggestions This week - Returns windowWording changed, policy kept **4 edits**Fine - Waterproof claimAgent removed a claim we don’t make **2 edits**Fix source - Shipping timesAgent added the carrier’s delay **3 edits**Coach Where agents change a suggestion tells you where a source is wrong, or where coaching helps. ## How you’ll know it’s working. 1. Answer accuracy against the source Reviewed on a sample each week, by someone who owns the policy. 2. Time to a correct answer How long the customer waits. 3. Adoption and edits, by tenure Who uses it, and who corrects it. Baselines agreed before we build. ## Works with what you have. Support - Gorgias - Zendesk - Intercom - Assembled Commerce - Shopify We configure first Zendesk’s Copilot, Gorgias and Intercom suggest replies, and Assembled has its own copilot. We pick what fits your helpdesk, then write the sources and rules it uses. We build when The suggestions are fluent but can’t show their source, or can’t tell which policy applies. ## What a first build includes. Built in your accounts, around your team. Scope, fee and baseline agreed before we start. 1. 1 Design 2. 2 Build 3. 3 Run it, or hand it over [Start with a review]() [Scope this build]() ### We build - Source-linked suggestions inside your helpdesk - Rules for which policy applies to which order - A weekly edit review that finds wrong sources - Coaching examples drawn from real conversations ### We start from - Helpdesk and Shopify access - Policies, macros and product facts - A month of conversations and agent edits ### From your team Owner CX lead Involves - Customer service - Quality - Engineering ## Questions teams ask. Does it actually help agents? A peer-reviewed study of 5,172 support agents found AI suggestions raised issues resolved per hour by about 15% (Brynjolfsson, Li and Raymond, QJE 2025). Gains were largest for newer agents, with small quality declines among the most experienced. That was business-software support in 2020 and 2021, so we measure your own cohorts. Will agents have to use it? It works as a suggestion. Agents approve, edit or ignore it, and their edits are the signal that tells us where a source is wrong. Who owns it once it’s running? You do. It’s built in your accounts, documented, with training for the people who run it. Ownership usually sits with your CX lead. We can keep running it with you, or step back. How do we start? Usually with a support review: two to four weeks, a fixed fee credited to the build, and a plan that’s yours. If the job is already clear, we can scope the build directly. ## Part of your support engine. One of 12 use cases, in 6 parts. [See the whole support engine]() 1. [Order changes **2**]() 2. [Delivery & returns **2**]() 3. [Product help **2**]() 4. [Knowledge **2**]() 5. [Quality & proof **2**]() 6. [People & coverage **2**]() In knowledge - [**Knowledge upkeep**Keep every answer current when policies and products change]() - **Agent assist**You’re here Works with - [**Product diagnosis**Diagnose product problems and hand specialists a complete case]() - [**Coverage planning**Plan coverage across your team, AI and your BPO]() ## Put the right answer in front of your agents. Start with a support review. [Start with a support review]() [**Talk it through first**15 minutes]() We reply within one business day. --- Source: [https://www.thekeating.co/use-cases/agent-assist/]() [Contact Keating & Co](mailto:nelson@thekeating.co) · [Agent guidance](https://www.thekeating.co/llms.txt)