Services · Claims

Claims: we do the work in every system and follow up with suppliers and carriers.

A ready-made service from Cloudfox. The AI agent takes the whole case, from the customer's email until it is closed, registers it in your systems and asks you in Teams when it is time to decide. We are responsible for it when it runs. Fixed price, in operation in 2–4 weeks.

The money backThe claims to the carrier and the supplier are sent on time and with complete documentation, and followed up until the compensation arrives. Less is written off.
Hours back in customer serviceThe agent registers in the systems, communicates with the customer and follows up with suppliers and carriers. You decide in every case.
Happier customersThe customer gets an answer the same day and word on what happens next, and the case is closed without the customer having to chase. The compensation goes out as soon as you have decided.
Fewer claimsThe causes show per supplier, carrier and batch. What keeps coming back becomes an action before the next claim: change the packaging or the carrier.
Book 30 minutes

The flow is being implemented at customers today. The customers are not named publicly.

ERP systems Business Central · Dynamics FO · Monitor · SAP · and more
Carriers PostNord · DHL · DSV (formerly Schenker) · Bring · UPS · FedEx · Jetpak · Bussgods · Green Cargo · DFDS · and more, via API or portal
The problem

The decision takes a minute. Everything around it takes weeks.

The customer emails. Before anyone can answer, one person jumps between the email, the CRM, the ERP, the document system, the carrier's portal and the quality system, logs in to each of them and enters the same information several times.

Then the follow-up begins. The damage must be reported to the carrier and the credit requested from the supplier. Replies are slow, documentation is missing, reminders must be sent and deadlines watched, often for weeks. The case sits waiting for someone to have time.

Four things often go wrong.

  1. The customer waits. Before anyone can answer, the order, the invoice and the shipment have to be found in several systems. The customer who emailed in the morning gets an answer in a few days.
  2. The deadline is missed. The carriers' deadlines for reporting concealed damage are short and counted in days. Once they pass, compensation is hard to get, and the damage ends up with you.
  3. The follow-up is dropped. Small claims against carriers and suppliers are written off, because no one has time to send a third reminder.
  4. The pattern goes unseen. The case is rarely registered in full in the quality system, and the rest stays in the email, the ERP and the carrier's portal. No one sees that the same supplier, batch or carrier keeps coming back.

It costs twice: customer service hours, in the registration and in the follow-up, and the compensation to the customer that never comes back from the carrier or the supplier. Often, everything also sits in one person's head. When that person is away, the cases stand still.

Today35 min
One person, six systems and weeks of reminders
Email3 min
CRM3 min
ERP5 min
Document system3 min
Carrier portal7 min
Quality system4 min
Follow-up, carrier and supplier10 min
With the agent1 min
The agent does the work and follows up, you make the decision
Email✓ the agent
CRM✓ the agent
ERP✓ the agent
Document system✓ the agent
Carrier portal✓ the agent
Quality system✓ the agent
Follow-up, carrier and supplier✓ the agent
Your decision in Teams1 min
Example of one case. The follow-up is reminders and documentation to the carrier and the supplier, spread over weeks. Your own figures are established in the baseline measurement.
Our solution

Four steps, in this order.

  1. We take over the case handling. The agent we implement at your company does the same job you do today, in the systems you already have: reads the claim, looks up the customer, fetches the order and the invoice, reports the damage to the carrier, registers the case in full and confirms to the customer. The decision is still yours. It arrives as a question in Teams, with all the documentation. One answer, and the agent updates every system: the ERP, the carrier, the quality system, the customer. If the agent is unsure, it hands the case to you, with the documentation gathered.
  2. We follow up until the compensation arrives. The agent stays with the case until it is closed: reminds the carrier until the damage is compensated, requests the credit from the supplier, watches the return and the deadlines. Cases that are written off today are followed through to a close.
  3. We find the patterns. Once the cases are complete, AI can read everything: the free text in the claims and the emails, and the data from every system. Then what no one had time to see becomes visible: which causes keep coming back, what the claims cost per supplier, batch, carrier and customer, and how much is recovered from subcontractors and carriers.
  4. We improve with AI. AI learns from every case: which decisions you made, what the carrier approved, which causes came back. The proposals get better every month, with a rationale, and more decisions can be approved with one click. What keeps coming back, the same supplier, batch, carrier or step in your own production, becomes an action before the next claim: change the packaging, move the volume, renegotiate, correct the production.
Customer service · Claims
CF
Claims agent09:16
Claim R-2291 · Nordvik Bygg AB
Email 09:14 · order 48213 · delivered 3 days ago
Damage
2 of 14 packages crushed on delivery, photos attached
Carrier
Reported in the portal 09:15 · deadline 7 days ✓
Quality system
Registered · cause: transport damage
Customer
Confirmation sent 09:15
Proposal
Replacement delivery of 2 packages tomorrow. The cost, SEK 6,480, is claimed from the carrier.
Same decision as in 11 of 12 similar cases this year.
ApproveChangeReject
One click, and the agent updates the ERP, the carrier, the quality system and the customer. Made-up data.

The agent takes the case handling. The team spends its time on what requires judgement: the difficult cases, the conversation with the customer who was affected, the supplier who keeps coming back, and solving the problems that cause the faults. That is where claims become fewer, not in the registration.

No new systems, no new logins. The agent logs in where you log in.

Do the maths

What the case handling costs you.

Move the sliders. The figures are calculated on your own numbers, not on an average.

The time
The money
420
hours per year on case handling today
224 400 SEK
freed up per year in time, with one minute per decision
162 000 SEK
recovered per year from carriers and suppliers
386 400 SEK
per year, time and money together
The default values are cautious examples. Your own figures are established in the baseline measurement, before implementation.
How we work

From the first call to operation.

  1. Walkthrough, 30 minutes. You tell us how you handle claims today: which systems, how many cases, where it gets stuck. We tell you what can be automated first.
  2. Mapping and baseline. We go through the cases with the person who handles them today and measure the time, the cost and the deadlines before the agent is implemented. That becomes the yardstick for the result.
  3. Implementation and testing. We implement the agent in your environment or in ours, together with your team, and test it against a key of your own cases. Fixed price, in operation in 2–4 weeks.
  4. Operation. We take application responsibility for the agent, always: cost cap, monthly report and a named person to call. If the agent runs at your company, you operate the platform yourselves or via your IT partner, as today. If it runs with us, we operate it.

What we need from you.

  • The person who handles the claims today, a couple of hours in the mapping.
  • Read and write access in your test environment, via API or login.
  • An hour with your IT, about your Microsoft or Google environment.
Key figures

This is what we measure.

  • The time — handling time per case, and hours per month in customer service.
  • The quality cost per month — the sum of time, credits and replacement deliveries, broken down per supplier, batch, carrier and customer.
  • Recovered from subcontractors and carriers — in SEK and as a share of the cost.
  • Claims per 1,000 orders — how often it goes wrong.
  • Fully registered cases — share of cases with classification, cause and cost filled in.
  • Missed carrier deadlines — number of cases where the carrier's deadline has passed.
  • Recurring causes — the same fault in several cases, which AI finds in the free text and the data: a supplier, a batch, a carrier, a step in production. Measured as number found, number addressed, and whether they stop recurring.
Common questions

What you usually ask.

  • What happens if the agent makes a mistake? It sends, credits and reports nothing without your approval. If it is unsure, it hands the case to you, with the documentation gathered. Every step is logged.
  • Does it work with our systems? With the systems you already have: the ERP, email and Teams, the CRM, the quality system and the carriers' portals. Where there is no API, the agent uses the portal, as you do.
  • Where does our data end up? In your Microsoft environment or in ours. The language model sees the case with the personal data masked, and it does not train on your data.
  • What does it cost? A fixed price for the implementation, which you get after the mapping. Operation has a monthly cost cap, so the AI calls are never a surprise.
  • How long does it take? 2–4 weeks to operation. From you it takes a couple of hours with the person who handles the claims today, an hour with your IT and access to a test environment.
For your IT

How we implement and run it.

Show the technical overview
  • Identity. The agent runs as a service account in your Entra ID, with the least possible permissions per system. No admin accounts, no shared passwords.
  • Hosting. Runs in your Microsoft environment* or in ours. What leaves it is the calls to the language model, see below.
  • Integration. API where the system has one: Microsoft Graph for email and Teams, and the native APIs in the ERP, CRM, document and quality systems. Where there is only a portal, the agent drives the browser via Playwright, the same clicks you make. Carriers vary: some have an API, others only a portal.
  • Credentials. API keys and the service account's secrets are kept in Key Vault in the environment the agent runs in, as are the logins to portals without an API. Never in a prompt, never in the code.
  • Decisions. The question arrives as a card in Teams, or in the system you already work in, the CRM or the ticketing system, with the documentation attached. Who may answer is governed by rules you set yourselves, for example amount limits. Nothing is sent, credited or reported without an approval.
  • Traceability. Every action is logged with account, time and case number. Portal steps are saved as screenshots.
  • The language model. Called via API, and what is sent is the case text and documentation, with personal data masked, nothing more. We make sure the provider does not train on your data and that retention of calls is agreed, down to zero if you wish. The simplest option is to run the model under your own Azure subscription: then calls and responses stay within Azure.
  • GDPR. Names, email addresses and phone numbers are replaced with placeholders before the call and put back in the response. The model sees the case, not the person.
  • Application responsibility. Always lies with Cloudfox, from day one. We answer for the agent: new requirements, new rules, new suppliers, new amount limits, a new carrier. That it does the right thing and stops when it is unsure. That APIs are versioned, models replaced and portals rebuilt. And what the AI calls cost, with cost cap and alerts.
  • Operations responsibility. Covers the platform: tenant, accounts, network, updates, that the resources are up. If the agent runs at your company, it lies where operations lie today, with your IT department or operations partner. If it runs with us, it lies with Cloudfox.

* If you have Google instead of Microsoft, the same setup applies: Google Workspace for email and chat, Google Cloud for hosting and keys.

Book 30 minutes.

Tell us how you handle claims today, and we will show you what can be automated first and what it gives in hours and money.

We use the information to reply to you. How we process personal data (in Swedish)

Your contact
Pontus Granborg
Senior AI Solution Architect, Cloudfox

You talk directly to the person who implements the agent at your company, from the first call to operation.

Email
pontus.granborg@cloudfox.se
Phone
+46 70 090 17 06
LinkedIn
linkedin.com/in/pontusgranborg