Project controlling with AI: track costs and margins

Why a project can look healthy until the last week

A consultancy has invoiced half of a fixed-price assignment. Its dashboard shows steady revenue, yet the team has already used most of the planned hours. Neither figure is wrong. They describe different things. Project controlling brings scope, delivery progress, cost and expected remaining work into the same review.

This is a useful AI application because managers often spend their review time finding exceptions across spreadsheets. An AI data assistant can help prepare that review. It cannot inspect the customer's acceptance of a deliverable or decide that an unrecorded change request is billable.

Establish the project baseline

Record the agreed fee excluding VAT, approved scope, budgeted hours, internal cost rates and purchased work. Give every project a stable identifier and assign an owner to the budget. Record approved changes separately so the original estimate remains explainable.

Before connecting sources, check the integration catalogue. A project system may contain hours and invoices without the cost rates or remaining-work estimates needed for margin calculations. Supply missing inputs through an agreed process, for example a controlled CSV dataset. Never substitute the billing rate for the cost rate without explaining the difference.

Start with four calculations:

• Actual labour cost equals approved recorded hours multiplied by the applicable internal cost rate.
• Actual project cost adds purchased work and other included direct costs.
• Expected final cost adds the current estimate to complete to actual project cost.
• Expected project margin equals agreed revenue minus expected final cost.

These are management calculations, not instructions for statutory revenue recognition. Finance should determine which revenue basis is appropriate for the review.

Work through one example

Consider a hypothetical fixed-price assignment worth EUR 30,000. The team has recorded 200 hours at an internal cost of EUR 70 per hour and EUR 3,000 of purchased work. Actual cost is EUR 17,000.

The manager estimates another 100 hours and EUR 1,000 of purchases to finish. Expected final cost is EUR 25,000, leaving EUR 5,000 of expected margin, or 16.7% of the agreed fee. Looking only at the EUR 13,000 difference between the fee and cost to date would overstate the expected result.

If the remaining-hours estimate changes to 140, expected final cost becomes EUR 27,800 and margin falls to EUR 2,200. That change deserves a conversation about scope and delivery. It does not prove poor employee performance.

Give AI a bounded review task

Ask: "Compare expected margin with the approved project budget. Show which projects changed most, the reporting date and the inputs behind each change. Separate recorded costs from estimates."

In DigiData, define and review the required measurements before reusing them in dashboards and chat. Missing estimates should appear as missing. Treat a generated explanation as a hypothesis until the project owner confirms it. More hours can mean additional scope, delayed time entry or an incorrect project code.

A weekly review works only when someone updates the inputs. Set the time-entry cutoff before the review and check synchronization status. A recently refreshed dashboard can still contain late or incomplete source records.

Turn exceptions into decisions

Assign each material exception to the project manager. They should confirm remaining work, check whether scope changes were approved and record the chosen action. Options include discussing extra work with the customer, changing the delivery plan or accepting a lower margin. The assistant prepares evidence; people make commercial decisions.

Judge the first month by useful outcomes: fewer unreviewed exceptions, clearer remaining-work estimates and less preparation time. Do not count every flagged project as recovered revenue. If your current concern is repeated report preparation across teams, continue with automating management reporting.