HarmonyAI: an AI workplace for every enterprise role. Key Account Manager first, then other roles.

One agent in the middle. Ask it anything about your data and get an artifact — a table, a chart, a conclusion with its source — that you can save, pin and rebuild on schedule, so the monthly review pack assembles itself. Turn any question into an automation: a condition, a trigger, an action in your systems. And when a job needs a screen, open a built-in app. All of it on your data, on approved models, including local ones, inside your perimeter.

Early access · design partner programme 2026–2027

Apply for early access See how it works

What changes

A question becomes an answer, not a ticket.

"Why is my Q4 forecast below plan" returns a table, a chart and a conclusion with its source — in a minute, without an analyst.

A job becomes a draft you confirm.

The monthly review pack, a promo in the TPM, a letter to the retailer about a listing gap — done by the agent, confirmed by you.

Deviations arrive before the review.

Alerts with the money at stake and three buttons, a week after the promo started, not at the monthly meeting.

Your systems stay where they are.

One workplace for the role on top of them; nothing replaced, nothing moved, nothing written without confirmation.

Four things on one foundation

The agent

Ask in plain words: a fact, a trend, a comparison, "why". The agent resolves the question through your dictionary of metrics, runs the query under your access rights and returns an artifact: the number, the table, the chart, the conclusion, the source. If the data is not there, it says so.

Answers from a governed dictionary, not free SQL
Every number with source and calculation
Honest refusal when the data is missing

Artifacts

Every answer is an object. Save it, pin it to your cockpit as a live block, share it, compare it with last month's version, or put it on a schedule: the weekly run-rate every Monday, the quarterly pack on the first business day, on your template.

Save, pin, share, compare revisions
Rebuild on schedule
A pinned answer is how a widget appears

Automations and alerts

Any question can become a rule: a condition, a trigger — a schedule, a data update, a file uploaded, another alert — and an action: raise an alert, mail a digest, hand a task to the agent, create a draft in a system. Alerts carry the money at stake and three buttons.

Promo without approval → reminder → escalation
Contract threshold near → alert with two scenarios
End of quarter → accruals to the ERP, with confirmation

Built-in apps

When the job needs a screen: analytics apps — customer analytics, promo analytics, the monthly demand review — and process apps — a meeting pack, closing a listing gap, preparing a promo, mapping a new product's codes, a contract-threshold scenario.

Analytics and process apps on the same data
Built by us, your team or the agent on request
A manifest says who made it and what it touches

The foundation underneath

Connectors.ERP, TPM, CRM, planning model, retailer sell-out, market panel, cubes, Excel and CSV. Read by default; write through your systems' own APIs, with confirmation.
Metric dictionary and data passport.One dictionary of metrics, hierarchies and mappings across systems, with freshness per source. Owned by your analyst; a review queue for the rest.
Access and audit.Rights inherited from your identity provider; every action, rule, export and decision in the audit log.
LLM runtime.Approved models only: open-source models inside your perimeter, or your cloud provider's models where policy allows.

How it works

Connect2–3 sources, the dictionary for 20–30 metrics, rights from SSO.
AskThe role's real questions become the benchmark; answers with sources from day one.
Pin and scheduleAnswers become the cockpit; reports rebuild on schedule.
AutomateRules over the metrics; alerts with money at stake.
ActDrafts in your systems, confirmed by a person.

Who it is for

Key account manager — first role: plan and actuals for my customers, a report on request, actions without five systems.
Head of KAM / sales director — the team cockpit without asking for reports; one number at the review.
Supplier relations, category management, RGM, finance business partner — next roles on the same foundation.
Commercial director / function head — the buyer: faster decisions, people on their customers instead of spreadsheets.
IT and data — the gatekeeper: on-premise, SSO, audit, approved models, no new data warehouse.

What HarmonyAI is not

Not self-service BI: the cockpit is configured for the role; open-ended analysis goes through the agent.
Not a replacement for your ERP, TPM or CRM: they remain the systems of record and the executors of actions.
Not a data warehouse: we connect and build the dictionary; we do not own your data layer.
Not an ML platform: forecasts and elasticities come from your models; HarmonyAI shows, explains and acts on them.
Not a chat: if there is no data to answer, the agent says so.

What we measure with design partners

HarmonyAI is in early access. We do not publish effect figures yet; we agree what to measure with every design partner and publish when the numbers are real.

Time on data.Hours per week spent assembling numbers and reports, before and after; the share of questions answered without an analyst.
Speed of reaction.The share of deviations caught by an alert before the monthly review; time from alert to action.
Routine automated.The share of recurring reports and actions running as automations.
One number at the review.Not measured, but the first thing a commercial director notices.

Early access programme

You get

A 12-week pilot at a fixed price: one role, one customer, 2–3 sources, your users in the last weeks.
The metric dictionary built for your systems — it stays yours.
Influence on the roadmap: the next role template and apps are built with design partners.
Early-access pricing for production.

We ask

A sponsor on the business side and an analyst who will own the dictionary.
Access to 2–3 systems and a server with a GPU, or a cloud account under your control.
Thirty to fifty real questions from the role before we start.
Honest feedback and permission to describe the results, anonymised.

Questions we hear first

We already have Copilot in our CRM / Joule in SAP.

They see their own system. Your KAM lives in five. HarmonyAI starts from the role and connects all of them, whoever makes them.

Isn't this one more system?

It is the one your KAM will open instead of the others. It adds nothing to your data landscape, only a workplace on top of it.

Our data cannot go to the cloud.

It does not. Approved models run inside your perimeter; there is no outbound traffic to a model provider.

Can our team build apps and automations themselves?

Yes. Automations are configured on a canvas; apps carry a manifest and can be built by your team or requested from the agent.

Give one KAM a week without BI

One key account manager, one customer, one week in HarmonyAI instead of five systems. Apply, and we will tell you within a week whether you are a fit for the 2026–2027 cohort.

Apply for early access