Make your business smoother, faster, and harder to break.

Backlog (name TBD) is an operations intelligence platform. An AI agent has ongoing conversations in scale with the people doing the work and surfaces a register of signals: from opportunities to consider, to risks to know, and to general information to keep in mind. Where action is possible, the agent estimates likely impact so the highest-impact signals rise first. Once action is taken, the agent follows up to see whether the change actually helped.

Already have an account? Sign in

01Conversation

02Observation

03Operating picture

04Feedback

01 / Conversation

It starts with a conversation.

A person talks with the agent about the work as it actually happens: what is slow, confusing, fragile, hard to learn, or harder than it should be. The agent asks one question at a time, in plain language, and the conversation stays private to that person.

ConversationPrivate to the employee

Backlog Agent

How did the marketing review go this week?

Employee

Honestly, slow. Every revision has to go through three people and the last reviewer is always booked back-to-back.

Backlog Agent

Got it — so a revision can't close until all three have looked at it? Roughly how many revisions land in a week?

Employee

Five or six. Each one adds a day or two of waiting, not actual work.

Logged as a process observation — kept at the work level, not about anyone by name.

02 / Observation

The agent captures the observation.

An observation is written at the work level, never as a judgement about a person: the situation, and what could change. The agent keeps discussing it until it can estimate what acting on it would be worth — time recovered, risk removed, confusion reduced — and records that in the detail. So a small comment becomes something you can actually weigh.

ObservationDetail view

PROC-12Raised in conversationUpdated 4 days ago

Marketing review takes three days because every revision touches three roles.

Estimated impact

9h / wk

DifficultyModerate

ConfidenceHigh

StatusShortlisted

CategoryProcess delays

The situation

Every revision of a marketing asset has to pass three roles before it can ship. The third reviewer is consistently the constraint — back-to-back booked on the two days revisions tend to land. Five to six revisions a week, each adding a day or two of waiting that isn't actual work.

What could change

Collapse the three sequential approvals into one async checkpoint with a clear owner, or shift the rarely-disputed checks to a post-publish review. Either gives the calendar time back without losing the quality gate.

03 / Operating picture

Everything lands in one operating picture.

Every observation across the organization collects in one place, ranked by likely impact — so the opportunity or risk worth doing next sits right at the top. Picking the highest-leverage move is a glance and a click, not an archaeology dig through threads. People still decide; the ranking just makes the best candidate obvious.

Operating pictureRanked by impact

Time reclaimable / week~17 hrs

High-impact risks3

People contributing11

PROC
Act nextMarketing review takes three days — every revision routes through three roles in sequence
The last reviewer is back-to-back booked on the days revisions tend to land.
~9 h / wk
PROC
Sales exports leads as CSV; ops hand-reformats every batch into the CRM's import format
Exporting the right format is near-free on the sales side — the conversion work is hours the ops team never needed to spend.
~6 h / wk
RSK
The production deploy ritual is undocumented
One engineer knows the order. Two weeks of vacation are booked in May.
High
TIME
Engineering spent 22% of last quarter on incident response
Up from 12% the quarter before — the trend held across four people.
Quarter / quarter
OPP
The same five onboarding questions land in #ops-help every Monday
A short FAQ surface would clear most of them before they're asked.
~2 h / wk

04 / Feedback

Feedback closes the loop.

When someone acts on an observation, the agent comes back later and asks whether the change matched the need, whether it helped, and what still feels wrong. That feedback closes the loop — and becomes context for the next conversation.

Follow-upAfter action

What leaders use it for.

The point is not another dashboard to admire. It is a way to keep the next operational decisions grounded in what people are actually experiencing.

Keep an ops roadmap that's actually current.

Leader question

Which problems are still worth attention after the work has changed?

What the agent surfaces

The agent keeps capturing observations as people do the work, estimates their impact, and follows up after action so stale priorities fall away.

A sample observation

PROC-12 · Marketing review takes three days because every revision touches three roles. Estimate: ~9 h/wk tied up in waiting.

Decide where AI actually fits, with evidence.

Leader question

Where would automation remove real friction instead of becoming theatre?

What the agent surfaces

The agent finds concrete tasks people already describe as repetitive, error-prone, or hard to route, then estimates what would change if they were automated.

A sample observation

OPP-007 · Convert client feedback DMs into Linear issues automatically. Estimate: ~1.5 h/wk reclaimed across customer ops.

See the risk building before it lands.

Leader question

What will break if one person is unavailable or one quiet habit stops working?

What the agent surfaces

The agent turns knowledge concentration, undocumented rituals, and fragile dependencies into observations while there is still time to act.

A sample observation

RSK-008 · Production deploy ritual is undocumented. One engineer knows the order. Two weeks of vacation booked in May.

Check whether fixes actually helped.

Leader question

Did the action solve the need, or just create a new workaround?

What the agent surfaces

After action, the agent follows up with the people affected and brings back what worked, what still feels wrong, and what should change next.

A sample observation

NOTE-014 · Async approvals helped, but legal review still waits on one calendar. Next check: move legal questions into the brief template.

Frequently asked.

How is this different from a yearly consulting audit?

A consulting audit gives you a slide deck after the interviews. Useful, but it is a snapshot. The agent keeps talking with people as the work changes, captures new observations, updates impact estimates, and follows up after action so the operating picture does not go stale the week after the project ends.

How is this different from running employee surveys?

Surveys ask everyone the same fixed questions and usually return aggregate sentiment. The agent has a real conversation: it probes, restates, and follows the thread until it can log a concrete work-level observation. Same familiar shape, different subject: how the work is structured, not how people feel.

How is this different from a BI dashboard?

A BI dashboard shows the numbers you already instrument: throughput, cycle time, ticket counts. It cannot see the undocumented deploy ritual, the approval that waits on one overloaded person, or the tool everyone quietly works around. Those live in people's heads until someone asks.

Can't I just ask people myself?

You should. Backlog is for the point where the operation is bigger than one person can hold in their head. Asking around tends to surface what was mentioned most recently or most loudly. The agent asks consistently, turns answers into observations, estimates impact, and keeps following up.

Why won't my employees just hate this?

Because it isn’t surveillance. The agent doesn’t observe anyone’s work — it only knows what each person chooses to tell it in their own conversation. There’s no tool integration, no tracking, no activity log. Whatever someone shares is translated into a work-level observation about a process, a tool, or a workflow — never about the person — and the original sentence isn’t stored. The conversation stays private to the individual; admins only ever see the depersonalised register.

Will my legal team or works council push back?

It's built so they don't have to. Backlog never infers emotion, never scores or ranks individuals, never predicts about people — the things the EU AI Act prohibits or treats as high-risk in a workplace. Items describe situations, not people; the conversation stays private to the employee; worker transparency is built in. A self-classification memo under EU AI Act Article 6(4) ships with the product, so the answer to legal is a document, not a promise.

We're in the middle of a merger / reorg — is now a bad time?

It's often the best time. A transformation is exactly when the map in everyone's head goes out of date fastest — duplicated work, unclear ownership, processes that broke in the reshuffle. Backlog gives you a live read of where the friction actually landed, updated weekly, instead of guessing or waiting for a post-mortem. It's just as useful for the steady-state operation, but mid-change is when leaders feel the gap most.

What if only some of my team engages?

We design for partial coverage. Thirty to forty per cent engagement still produces a richer picture than you have today, because the agent captures concrete observations from the people who do respond. If the same friction shows up independently from several places, that is useful signal; if it shows up once, it can still be worth understanding.

What does Request help do?

Every item carries a one-click route to Fixer of the North. Companies that can self-implement do so. Companies that can’t get a scoping reply within a couple of days, with the original situation already on the page.

What does it cost?

Personal workspace is free. Team, Company, and Enterprise workspaces are priced per seat with an included AI quota — overages billed per use, never shown as tokens.

See it on your operation.

Join the waitlist for early access. We’re working directly with our first teams — tell us a little about what you’d want it to surface, and we’ll bring you in as we open up.