Give senior hours back to client work.
Agencies, consultancies, law and accounting practices, and recruiters sell expertise. Yet a large share of senior time goes into intake, proposals, reports and billing clean-up. We look for the workflows where agents can assemble and check that work, so your people review it rather than build it, and every client fact stays with the right client.
Six workflows between the enquiry and the invoice.
These are the workflows we would examine first in a services firm. They run through the CRM or practice management system, document folders, timesheets, the billing system and many old files used as templates.
| Workflow | Manual today | AI-native version | Stays human | Verification check |
|---|---|---|---|---|
| Client intake and conflict checks | Staff re-key enquiry details into the practice system, run conflict searches by hand and draft the engagement letter from an old one. | An agent extracts the details, runs the conflict search on all name variants and related parties, and drafts the engagement letter from the current template. | Conflict decisions, accepting the client and the fee terms. | The search covers every listed party and variant, and its results are logged. No matter opens without sign-off. |
| Scoping and proposal assembly | A partner or director copies an old proposal, rewrites the scope, recalculates fees and hunts for credentials. | An agent assembles a draft from approved scope modules, the rate card and an approved credentials library. | Pricing, scope commitments and the final read. | Rates must match the rate card. Every credential and case reference must come from the approved library. |
| Client reporting | Account teams pull data from several platforms, paste it into a deck and write commentary every month. | An agent pulls and validates the data, builds the report and drafts commentary on what changed and why. | The interpretation, recommendations and the client conversation. | Every figure is reconciled to the source platform's totals. The report uses that client's data only. |
| Timesheet and billing reconciliation | Managers chase missing timesheets, review pre-bills line by line and argue about write-downs at month end. | An agent flags missing or unusual time, maps entries to engagement codes and drafts pre-bills against budgets and caps. | Write-offs, billing decisions and client discussions. | Billed hours must equal approved time. Rates and caps are checked. No invoice goes out without partner approval. |
| Client onboarding | Staff collect identity and company documents for anti-money-laundering checks, then set up access, folders and billing details. | An agent collects and checks documents against the required list, pulls registry data and prepares the onboarding pack. | The risk assessment and the decision to accept. | Evidence must be complete for the risk level before work starts. Screening hits always go to a person. |
| Candidate screening and scheduling (recruitment) | Recruiters read every CV against the brief and spend hours scheduling interviews by email. | An agent summarizes each CV against the stated must-have criteria and handles interview scheduling. | Every shortlisting and rejection decision. Nothing is rejected automatically. | The same criteria apply to every candidate and are logged. Summaries cite the CV text they rely on. |
AI used for recruitment and candidate selection is classed as high-risk under the EU AI Act. That is why we keep every hiring decision with a person and log the criteria used.
Six numbers that show whether it worked.
The Audit measures these before anything is built. The Pilot is measured against the same numbers.
| Metric | Definition |
|---|---|
| Enquiry to engagement letter | Working days from first enquiry to a signed engagement letter. |
| Proposal turnaround | Working days from a scoping call to a proposal sent. |
| Reporting hours per client | People time spent producing each recurring client report. |
| Unbilled time leakage | Hours worked but never billed, as a share of hours recorded. |
| Lock-up days | Days from work done to cash received, across work in progress and receivables. |
| Write-off rate | Value written off at pre-bill or invoice stage, as a share of standard value. |
What can go wrong in professional services, and the control for each.
In services firms, the worst failures involve trust: one client's information in another's document, a missed conflict, or a bill the client disputes. We design the controls around those first.
| Failure mode | Control | How we'd detect it |
|---|---|---|
| One client's information appears in another's document | Retrieval is scoped to the client or matter. Agents can't see other clients' files while drafting. | Access logs per document and a pre-send check for other client names. |
| Conflict of interest missed | The search runs on every party and name variant. A person decides every possible match. | Logged search terms and results for every new matter. |
| Wrong number in a client report | Every figure is reconciled to the source platform totals before the report is released. | Variance check between report figures and the source exports. |
| Client over- or under-billed | Pre-bills are checked against approved time, rate card and fee caps. A partner approves every invoice. | Billing disputes and credit notes tracked per month and per cause. |
| Proposal claims an unapproved credential | Credentials and case references come only from the approved library. | Provenance log for every claim in a sent proposal. |
| Biased or inconsistent candidate screening | No automated rejection. The same logged criteria apply to every candidate. A person decides. | Periodic review of shortlisting outcomes across candidate groups. |
Monthly client reporting, from diagnostic to pilot.
Illustrative example — not a client result. The numbers below are assumptions chosen to show the method and the arithmetic. Your Audit replaces them with measured values.
| 01 | Diagnostic | Suppose a performance marketing agency picks "monthly client reporting" on the General track. The Snapshot might show a High leverage signal with Medium confidence, friction at the Demand/Input and Processing stages, and Level 2 (workflow augmentation) as the indicated AI depth. |
| 02 | Audit | Suppose the Audit measures 45 client reports a month at 4 hours each, pulling data from four platforms and writing commentary. That is 45 × 4 = 180 hours a month, about 1.1 FTE, mostly account managers' time that could otherwise be client work. |
| 03 | Pilot | Suppose the agent builds every report and drafts the commentary, and an account manager spends 1 hour reviewing and editing each one. About 5 reports a month need 2 extra hours of rework because the data was unusual. The Pilot measures whether that holds. |
| Line | Assumption | Hours per month |
|---|---|---|
| Baseline | 45 reports × 4 h | 180.0 |
| Review, after | 45 reports × 1 h | 45.0 |
| Rework, after | 5 reports × 2 h | 10.0 |
| After, total | 45.0 + 10.0 | 55.0 |
| Difference | 180.0 − 55.0 (about 0.8 FTE at 160 h) | 125.0 |
At an assumed loaded cost of €65 an hour, 125 hours is about €8,125 a month. In a services firm the bigger question is often what those hours become: more clients per account manager, or more billable time. The Audit would state that as a separate assumption, and count the new costs: compute, platform connections and review time. If the numbers don't hold, the recommendation is "don't proceed".
Where our proof comes from.
We'll be plain about this. Our production proof is in AdTech, not professional services. We have not yet run these workflows for a services client.
The architecture is the one that runs in AdTech production: typed intake, deterministic checks before any action touches client data or an invoice, bounded retries, then a person. Our AdOps Copilot answers campaign questions from live data with every answer traced to its source, which is the same pattern as a verified client report. The AdTech page and case studies show that work.
Our engineering team and delivery network have built related systems: invoice and business document AI with layout-aware parsing, an ERP/CRM assistant on self-hosted open-source models, and personalised client communication drafted from CSV and PDF data. See Selected builds.
That is why services engagements start with the Audit. It measures your own baseline first, so any decision rests on your numbers, not on ours.
Questions services firms ask first.
Will clients notice the work is AI-assisted?
They should notice faster, more consistent work. A person still reviews and owns everything that goes to a client. Whether and how you disclose AI assistance is your policy decision, and we design to support it.
How do you keep client data confidential?
Agents only see the files for the client or matter in hand, and every access is logged. The stack is self-hostable, so it can run in your own environment.
Our work is bespoke. Is any of it a fit?
The advice is bespoke. The intake, assembly, reporting and billing around it usually are not. Those are the parts we examine.
Find out which non-billable workflow is worth redesigning.
Start with the free diagnostic on the Other / General Business track. Move to the Audit only if the Snapshot points to something worth measuring.
| 01 | Free AI Operational Diagnostic | 10 steps, 8–12 minutes, General track. Snapshot on screen and as a PDF. |
| 02 | Operational Leverage Audit, from $1,500 | Fixed fee, one workflow, 10–15 working days after data access. Measured baseline, benefit model with stated assumptions, go / no-go. |
| 03 | Pilot implementation, from $6,000 | Fixed scope. The workflow built and run in your environment, measured against the Audit baseline. Typically 4–8 weeks. |