CASE STUDY · AGENTIC WORKFLOW

How a Financial Advisory Firm Put Its Shared Inbox on Autopilot

How a Financial Advisory Firm Put Its Shared Inbox on Autopilot

How a Financial Advisory Firm Put Its Shared Inbox on Autopilot

Cloudify built an agent that reads every email from the providers this firm works with, does the matching work in HubSpot, and colour-marks each message so the team sees what still needs a person.

Cloudify built an agent that reads every email from the providers this firm works with, does the matching work in HubSpot, and colour-marks each message so the team sees what still needs a person.

CLIENT

PensionsRådgiverne

INDUSTRY

Pension advice

SYSTEMS

Outlook · HubSpot · AI classification

SYSTEM OVERVIEW

A closed loop for routine emails.

AUTOMATED

SOURCE

Outlook inbox

Pension-company emails

SOURCE

Outlook inbox

Pension-company emails

AI AGENT

Ny Robot

Reads · understands · acts

AI AGENT

Ny Robot

Reads · understands · acts

DESTINATION

HubSpot

Deals updated automatically

DESTINATION

HubSpot

Deals updated automatically

Routine emails are handled, logged and colour-marked before anyone opens the inbox.

0%

Of messages had the admin done without a person

0 sec

From arrival to done, against four to five minutes

0.0

System actions carried out per message

0%

Of the agent’s decisions kept by the team

What happened to each email

1,348 · fully handled

447 with work done in HubSpot, 901 automatically moved away

738 · pre-sorted for the team

Customer found and email saved on the deal; a person decides the next step

403 · needs a person

The agent could not identify the customer, so it leaves the email to the team

"The AI agent has transformed how we manage our day-to-day operations"

Customer and Administration Manager at the client · August 2026

TESTIMONIAL

01 / 03

TIME SAVED

Give Your Team Hours Back Every Week

Give Your Team Hours Back Every Week

We estimated the time saved from two numbers we know: how long a person takes per message, and how many messages the agent did the admin for.

Four in five messages now arrive with the work already done. The team reads a decision, not an inbox.

A PERSON

Time per message

4–5 minutes

Available

Mon–Fri, 8 to 4

Hours a week

40

THE AGENT

Time per message

About 11 seconds

Available

Every day, all day

Hours a week

168

2,489

Messages handled in twelve weeks

84%

With the admin done by the agent

6,400

Actions carried out in the connected systems

0

Records linked to the wrong customer

Low end

Only the 1,348 messages closed with nobody involved, at 4–5 minutes each.

High end

All 2,086 messages where the agent found the customer, filed the email and created the task.

Not counted

The 403 messages left for a person, and the time a person still spends deciding on yellow ones.

This is an estimate, not a timed measurement. Early in the rollout, the client counted about 11 hours saved at the 1,000-task mark.

THE CHALLENGE

One shared inbox, dozens of emails, all by hand

One shared inbox, dozens of emails, all by hand

The firm helps private customers move their savings from one provider to another. Every provider replies into the same inbox, in its own wording, with no common format.

The firm helps private customers move their savings from one provider to another. Every provider replies into the same inbox, in its own wording, with no common format.

BEFORE · MANUAL

One person reads it

Find the deal by customer number. Work out what the email means. Create the follow-up task. Move the deal stage. Save the email on the record.

Slow, repetitive, and dependent on one person.

AFTER · THE AGENT

The team only handles judgement

Customer found. Task created in the right queue. Deal stage moved. Email saved on the record. Message colour-marked in Outlook.

The team only handles judgement.

Inbox rules could not solve this. The subject often says little, and the right action depends on reading the body and checking the deal in HubSpot.

THE SOLUTION

Five steps, every message

Five steps, every message

The AI decides only what an email means. Every HubSpot change follows rules the client agreed in advance.

The AI decides only what an email means. Every HubSpot change follows rules the client agreed in advance.

01

Reads

Takes the message and its attachments.

02

Understands

Sorts it into one of seven plain-language types.

03

Finds

Identifies the customer, three ways, tried in order.

04

Acts

Updates HubSpot by the agreed rule for that type.

05

Marks

Colours the message in Outlook.

WHAT HAPPENS IN HUBSPOT

EMAIL SAYS

Transfer rejected

A task in the queue for that reason. Seven reasons, seven queues, never a duplicate.

EMAIL SAYS

Transfer requested

Deal moved to “transfer request received”, with the provider recorded.

EMAIL SAYS

Customer changed their mind

Deal moved to “withdrawal received”, and the fourteen-day deadline checked.

From rulebook to live in five stages

  • Map
    Every email type and its rule, written down with the team.
  • Guardrails
    Which stages it may move, and what it must never touch.
  • Test
    292 real inbox emails before go-live.
  • Refine
    Colour and stage rules adjusted with the team.
  • Live
    2,489 messages in twelve weeks.

SAFE BY DESIGN

When the agent isn’t sure, a person decides

When the agent isn’t sure, a person decides

A wrong update is worse than no update. Every message ends in one of three outcomes, and the inbox shows which.

A wrong update is worse than no update. Every message ends in one of three outcomes, and the inbox shows which.

Green · handled

Closed. Nothing left to do.

Yellow · ready for you

Filed. A person decides.

Needs a person

Nothing changed.

Never guesses the customer

Anything short of one certain match goes to a person.

Never acts twice

Redelivered messages and open tasks are skipped.

Never answers for the advisor

When a provider asks a question, a person replies.

Never deletes anything

Messages are marked, never moved or removed.

Why it worked

Their process, not a template

Built on the firm’s own email types, queues, stages and rules.

Refined together

The colour model and stage rules were agreed before go-live.

Tested on real emails first

292 real messages showed what to expect on day one.

Room to grow

New email types can be added. The 403 unmatched messages are next.

If your process can be written as a rule, a conventional integration is cheaper and we will tell you so.

If your process can be written as a rule, a conventional integration is cheaper and we will tell you so.

If someone has to read it first and decide what it means, that is a different job and it is the one this page is about.

If someone has to read it first and decide what it means, that is a different job and it is the one this page is about.

Take the Process Check

Talk through your process