An online shopChat
Every “where’s my order?” answered in the chat.
The AI agent looks it up in your order system and the courier’s tracking, and sends the link.
Ana reads a sample of the day’s chats each morning.
Your people serve far more customers with AI agents.
AI agents answer the everyday questions in chat, text and email, find the answer while your people are on a call, and write up every call when it ends. Your people manage them, check their work and keep the conversations that need a person.
The first step is one day, on your own conversations. AWS funding typically covers the workshop. We rebuilt the contact center of one of the world’s largest cinema chains on Amazon Connect, which takes more than 40,000 calls a day. An AWS Premier Tier partner, one of about 300 among more than 130,000, and an Amazon Connect service delivery partner.
Every customer’s line ends at one tree, where AI agents do the work and your people check it.
Answering is easy.
The work around it isn’t.
In a business of 50 or 250 people, every conversation with a customer comes with work nobody sees.
- Someone finds the customer in three systems before they can help, while the customer waits.
- They hunt for the answer in a policy, or put the customer on hold to ask a colleague.
- Afterwards they type it all up: the notes, the tags, the follow-up email and the CRM. That adds minutes to every call.
- The same twenty questions arrive by chat, text and email, and get answered one at a time.
- A busy week means hiring, because the only way to get through more is more people.
AI agents take the work around the conversation. Your people keep the conversation itself and check what the AI agents did, so they get to the next customer sooner.
Tuesday, 10:12A bill up $42
of Rosa’s time
for one call
Example
The work AI agents take on.
The conversations your people keep.
Each job gets its own AI agent, working in your own systems, and your people decide how far it goes: from suggesting the answer during a call to finishing the job on its own. Six examples, from six kinds of business.
A physical therapy clinicPhone, 7:52 pm
Bookings moved after the front desk goes home.
On the phone, the AI agent checks who’s calling, finds the booking, offers the free slots and moves it.
The front desk sees every change at 8 the next morning.
An appliance repair companyText
Every visit confirmed by text.
The AI agent reads the technicians’ schedule and the van’s position, and writes the customer’s answer back to the job.
The dispatcher sees the reply on Luis’s job and plans around it.
A gymEmail, drafted
Account changes drafted by your rules.
The AI agent drafts the change and the reply from your terms, with the clause it followed. A person sends it.
Mia reads it and sends it with one press, or changes it first.
An internet providerPhone, Rosa answers
The answer on screen during the call, the notes after it.
Rosa talks to the customer. An AI agent puts the customer’s record and the answer on her screen, then writes up the call and the follow-up email.
Rosa keeps the conversation and checks the write-up in a minute.
A travel agencyThe morning after
The calls worth coaching, picked by morning.
An AI agent scores every call, people check some of the scores, and the few worth talking about are waiting in the morning.
The supervisor coaches on the calls that matter, not just the few she had time to hear.
AI agents take on more jobs like these: returns and exchanges, quotes for a repair, appointment reminders, a copy of an invoice, a changed delivery address, a password reset and a warranty claim. Examples.
We count jobs finished,
not calls deflected.
A call counts as deflected when the customer stops asking, and that can mean they gave up. We count the job instead: the booking moved, or the question answered right. Our engineers make six decisions with you to get there on Amazon Connect, the way AK AI Factory builds every AI system. Each one stops a failure your customers would otherwise meet, and two of them keep your people in charge.
- 1
Ground: it answers from your policies and your records.
WhyAn AI agent that answers from memory can promise a refund your policy doesn’t allow.
- It answers from your policies, prices and order records, and keeps the source of every answer.
- Once it knows who’s asking, it sees that customer’s history in your CRM, and no one else’s.
- Where your policies don’t hold the answer, it says so and hands over.
- 2
Orchestrate: it does the job in your systems.
WhyAn AI agent that can’t act in your systems can only say, “I’ve noted that, and someone will call you back.”
- Each job, from tracking an order to moving a booking, is a short list of tools into your systems, each with its own permission.
- Checking who’s calling follows a fixed path, and it never improvises.
- Amazon Connect gives a tool 30 seconds to answer, so a slow system gets a faster path.
- 3
Control: you decide how far each AI agent goes.
WhyNo AI agent should do more than you agreed it could.
- Each job gets one of three settings. The AI agent suggests to your person during the call, drafts for your person to send, or finishes the job and reports it. You choose, and can change it any day.
- Each AI agent has its own identity and only the tools its job needs, so the one that tracks orders can’t issue refunds.
- Its guardrails block card numbers and mask personal details, and your customers’ data stays in your own AWS account.
- A refund over your limit waits for your yes, from a named person. There’s always a way to a person, and a customer who asks for one gets one.
- 4
Prove: tested on your real conversations first.
WhyA demo can work and still fail real customers.
- Before we build, you and our lead agree one scorecard: jobs finished, today’s baseline, the target and what counts as a failure.
- We test it on real past conversations whose right outcome you know, and every change replays them before it goes live.
- A job moves up a setting on evidence: from drafting to finishing on its own only when the scorecard says it’s ready, and back down if it slips.
- 5
Run: your people manage it from one queue.
WhyWithout the story, your person has to ask, “Sorry, can you tell me again what the problem is?”
- What the AI agents drafted, what they weren’t sure of and a sample of what they finished arrive as tasks in one queue, routed to your people like any conversation.
- The hand-over carries the story: what the customer wants, what the AI agent tried and why it stopped.
- We set Amazon Connect to record and transcribe every conversation, so you can see what any AI agent did and why, and show a customer exactly what was done with their details.
- 6
Improve: coaching the next morning.
WhyCoaching that arrives days late lets the same mistake happen twice.
- Sonic, our call-scoring accelerator, scores every call, and people check a sample.
- Each night it picks the few calls worth a coaching conversation, each with an open question.
- When your team corrects an AI agent, the correction becomes a test it has to pass from then on.
One day to a plan.
About 90 days to go live.
We start with a one-day workshop, our contact center discovery, with the engineers who’d do the work. AWS funding typically covers the workshop. After it, every step has a date, and there’s a go-or-stop decision in week 4, on a scorecard agreed before we build.
We’ll tell you what we think, even when it isn’t what you hoped.
- Keep people answering first if most of your callers need a person. AI agents then stay behind your people, finding answers and writing up, and never pick up first.
- Don’t move yet if your contract has years to run. You leave with the move priced, to take into your renewal.
- Hire instead if one more person would cost you less. We’ll show you the sums.
- One dayIn the morning of the workshop we map what you run today: the channels, the call flows, the work after each call, your CRM and where your contract stands. In the afternoon we design the target on Amazon Connect. You leave with a findings report, a target architecture and a costed roadmap, yours to act on with us or without us.
- Week 4You decide go or stop, on your real conversations. If it’s stop, you keep the map, the baseline and the scorecard.
- About 90 daysIt goes live with a few customers first. Then your team runs it, or we do.
Contact centers we’ve already put right on Amazon Connect.
One of the world’s largest cinema chains
Onshore and offshore centers, and opening weekends to get through
40,000+calls a day, with AI agents helping its people
A rushed migration to Amazon Connect was failing a week before go-live. We stabilized the launch, then built in AI agents that put answers in front of its people during the call and write the summary after it, with voice AI agents, chat automation and live data.
- 30% faster call handling across the contact center.
- Three chat workflows run with no person in the loop.
- Coaching feedback arrives the same day.
- A major film release came and went with no emergency hiring.
Another Amazon Connect client
Its contact center, on Amazon Connect
30%faster email responses
The Amazon Connect work we did for another client, who isn’t named.
- Email answered 30% faster.
- Operating costs down 25%.
- 89% customer satisfaction (CSAT).
- 82% of questions resolved in the first contact.
We’re an AWS Premier Tier Services Partner and an Amazon Connect service delivery partner, both checked by AWS. The Amazon Connect estates we run hold 99.9% availability.
“Kudos to Armakuni for demonstrating the speed, precision, and partnership needed to turn a high-speed challenge into a success story.”
What owners say to put it off.
And what we tell them. Got another? Bring it to the workshop.
Our customers want a person, not an AI agent.
Then a person answers. On the phone, AI agents can stay behind your people: the customer’s record and the answer on screen during the call, the notes and the follow-up written after it. Where customers are happy to sort it themselves, like an order’s whereabouts in chat at 11 pm, the AI agent finishes it. It says what it is at the start, and asking for a person always works.
We’re mid-contract on another platform.
That’s the best time to look. The workshop prices the move before your renewal talks start, so you renew or leave with a costed alternative in hand.
When you move, FlowShift, our accelerator for moves, reads your old call flows, converts most of them automatically and flags the rest for your team to check. The old system keeps answering until the new one has proved itself, so there’s no big-bang weekend.
There’s no budget for it this year.
AWS funding typically covers the workshop, and AWS has funded part of the work itself for several of our clients. At the workshop we’ll tell you whether yours could qualify.
Running it is paid for by use. Amazon Connect charges by the minute or the message, and most of its AI is in that price. The gateway into your own systems and the knowledge your AI agents read are billed by use on top, and the scorecard counts them in the cost of each finished job.
We’re happy with what we have.
Then the workshop costs you a day, not a project, and you leave knowing what the work around each conversation costs you today, in minutes and in people. Some businesses find their numbers are fine, and we’ll say so.
Our own team can build this.
Some can, and we build it with them, inside your business, so they run it afterwards with the runbooks and the record.
What we add is having done it on Amazon Connect at 40,000+ calls a day, and the Connect accelerators we bring on day one. If you have no IT team, we build it with your people, and then your team runs it or we do.
Bring us a week of your conversations.
Spend one day with the engineers who’d do the work. You leave knowing which jobs AI agents should take first and how far each should go, and we’ll say plainly if it isn’t worth doing. AWS funding typically covers the workshop.