AI Agents for Consulting Firms

At Ridgeline Advisory, a 22-consultant operations consultancy that runs $150k to $400k engagements for mid-market manufacturers and logistics operators, every new client starts the same way: a partner opens a blank document and writes the scoping brief from scratch. This is the exact gap AI agents for consulting firms are supposed to close, but most tools sold under that label just type the same blank-page draft a little faster. Objectives, hypotheses, phased approach, staffing plan, deliverables, all of it rebuilt from memory even though the firm has run eleven nearly identical supply-chain engagements in the past three years. The average scoping doc takes a partner 9 hours to draft across two or three revisions, and no two partners structure it the same way, so a client reading a Ridgeline proposal cannot tell from the format alone whether they are getting the firm's sharpest thinking or an improvised first draft. Meanwhile Elena, the senior consultant who led 11 of those supply-chain engagements over four years, gave notice in March. She took with her the sense of which diagnostic questions actually surfaced a client's real bottleneck, which framework steering committees pushed back on, and which one they signed off on in the first meeting. None of it lived anywhere but her laptop and her memory. And across the nine engagements running concurrently right now, engagement leads spend roughly 20 hours a week combined building status decks for weekly client steering committee meetings, time that bills at the same rate as real analysis but produces a slide deck instead of a client outcome.

What AI agents for consulting firms actually need to do

Ask ChatGPT or Claude.ai to draft a scoping doc and you get a competent-sounding document shaped entirely by what you typed in that one prompt. It has no memory of Ridgeline's last 40 proposals, no sense of which phased structure actually won client sign-off versus which one got redlined into oblivion, and no persistent record of the firm's own methodology. Every partner who uses it starts from the same blank slate the tool started from: nothing. Worse, it forgets everything between sessions, so the fifth partner drafting a supply-chain scoping doc this quarter gets no benefit from the fact that four other partners already solved variations of the same structuring problem this year. A generalist chatbot optimizes for a plausible answer to the prompt in front of it, not for consistency across a firm's entire proposal history. Ask it to help write a status update for a steering committee and it will produce something readable, but it has no connection to the engagement's actual billed hours, milestone dates, or the deliverable that shipped last week, so a partner still has to manually feed it every fact before it can write a sentence, which is often slower than just writing the update directly.

Cursor and GitHub Copilot solve a different problem for a different profession entirely. They are autocomplete layers tuned for source code inside an IDE, fast next-token suggestions for a developer writing functions. Consulting firms do not ship code. Their deliverables are scoping documents, phased engagement plans, client memos, and weekly steering committee updates, none of which live inside a code editor. Some firms have tried bolting a generic AI writing assistant onto their Word proposal template, and it behaves the way autocomplete always behaves: it completes the next paragraph based on pattern matching in the sentence, not based on what actually won the last four supply-chain mandates or which fee structure a client pushed back on.

The mismatch is structural. A consulting firm needs two things a single-shot generalist tool cannot offer: a standardized intake and scoping motion applied consistently across every partner and every engagement, and a compounding memory of what worked that survives any one consultant's departure. Both require persistence across sessions and across people, and a memoryless chat window resets to zero every time someone opens a new tab.

Helm, Atlas, and Lumen: an engagement team, not a chat window

Helm is the primary agent for a consultancy running this workflow inside Claude Code. Helm is built to write structured, decision-ready briefs, at Tonone that means product requirements, and the same discipline applies directly to a client scoping document. The helm-recon skill reads whatever intake material already exists, an RFP, discovery call notes, a client's own strategy deck, before Helm drafts anything, so the scoping doc is grounded in what the client actually said rather than a generic template. The helm-brief skill then produces the structured document itself: problem statement, working hypothesis, phased approach, deliverables per phase, staffing plan, and fee structure, in the same format every time regardless of which partner runs the engagement. That consistency is the point. A client comparing a Ridgeline proposal against a competitor's should see a document that reflects a firm-wide standard, not one partner's personal style on a good day.

Tonone's Helm turns a client scoping document from a 9-hour blank-page draft into an 85-minute partner review, without sacrificing firm-wide consistency across partners.

Once a client signs, the helm-plan skill turns the scoping document into a phased engagement roadmap with milestones, dependencies, and the exact points where the client steering committee needs to see and approve progress before the next phase starts. Instead of a partner reverse-engineering a project plan from the sales proposal after the fact, the roadmap already exists the day the engagement kicks off, and when two stakeholders on the client side disagree about scope mid-engagement, the helm-arbiter skill gives the partner a structured way to weigh the competing asks against the original signed brief rather than relitigating scope from memory in a tense call.

Atlas is the knowledge engineer that closes the loop Elena took out the door with her. When an engagement wraps, the atlas-onboard skill takes the actual approach that was used, the diagnostic questions that surfaced the real bottleneck, the framework the steering committee pushed back on, and the one it approved without a second look, and turns it into a structured playbook entry tagged by industry and problem type. The next consultant staffed on a similar mandate, whether that is Elena's replacement six weeks into the job or a consultant borrowed from another practice group, opens a document that already tells them what worked and what did not, instead of starting from the same blank page Elena started from four years ago.

Tonone's Atlas turns a completed engagement's actual approach into a structured playbook entry, so institutional knowledge survives a consultant's departure instead of leaving with them.

Lumen is the third leg: metrics and reporting for engagements already in flight. The lumen-metrics skill sets up a lightweight dashboard tracking utilization, realization rate (billed value against fee actually collected), and engagement health across every active mandate, pulling from the same structured data the scoping doc and roadmap already established. Instead of an engagement lead rebuilding a status deck from scratch every week for a steering committee meeting, three hours of slide-building becomes a live view the lead walks through, with commentary added only where judgment is actually needed.

Tonone's Lumen turns a weekly client status deck that took three hours to build into a live metrics view an engagement lead narrates instead of assembles from scratch.

A worked example: standardizing intake and capturing what Elena knew

Say a regional food distributor, call it Callahan Foods, sends Ridgeline an RFP for a warehouse throughput engagement, a problem shape close to three of the eleven supply-chain mandates Elena led before she left. A partner runs helm-recon against the RFP and Callahan's own operations deck. Helm then checks Atlas's playbook library and surfaces two prior engagements tagged supply chain, warehouse throughput, mid-market, plus the diagnostic questions and framework that worked on both. helm-brief drafts the scoping document using that grounding instead of a blank template.

Ridgeline Advisory, Scoping Brief: Callahan Foods Warehouse Throughput
Recon: RFP + ops deck reviewed. Playbook match: 2 prior engagements
(Danvers Foods 2024, Bristol Logistics 2025), same problem shape.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Problem statement
  Outbound throughput has degraded 18% over 6 months despite flat
  headcount. Client RFP frames it as a labor problem.

Working hypothesis (from playbook match)
  Danvers and Bristol both presented as labor problems and were
  actually slotting and pick-path problems. Recommend the diagnostic
  phase test this before proposing any staffing fix.

Phased approach
  Phase 1 (2 wks): Pick-path and slotting diagnostic, 3 warehouse
    shifts observed, WMS data pull.
  Phase 2 (4 wks): Redesign slotting, pilot on 1 zone, measure.
  Phase 3 (3 wks): Rollout plus steering committee handoff doc.

Staffing: 1 partner (20%), 2 consultants (80%), 1 analyst (100%)
Fee: $210k, phased billing at each gate

Draft time: 85 min partner review (vs. approx. 9 hrs from-scratch draft)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Next: partner approves brief, helm-plan generates the phase-gated
roadmap and steering committee calendar.

Callahan signs within the week, in part because the phased approach reads as considered rather than improvised, exactly the signal a mid-market buyer wants from a boutique firm competing against a bigger name. During delivery, the weekly steering committee update pulls from the Lumen dashboard instead of a rebuilt deck, and engagement lead time on reporting drops from roughly three hours a week to under thirty minutes of narration on top of the live view. When the engagement closes eight weeks later, Atlas's atlas-onboard skill files a third playbook entry under supply chain, warehouse throughput, mid-market, this time including the detail that Callahan's sponsor initially resisted the slotting hypothesis until shown week-one pick-path data, a negotiation detail that will matter to whichever consultant runs the next mandate in this pattern. Elena is gone. The pattern she built is not.

Scale that across a year and the numbers compound. Ridgeline runs roughly 40 new-engagement proposals annually. At 9 hours of from-scratch partner drafting per proposal and a blended partner rate of $450 an hour, that is around $162,000 of partner time spent writing documents instead of advising clients or developing new business, before Helm ever enters the picture. Cutting that to an 85-minute review frees most of those hours back to billable or business-development work. The same math applies to the 20 hours a week the firm spends on status decks across nine concurrent engagements: at a blended consultant rate of $340 an hour, that is roughly $6,800 a week, or over $340,000 a year, spent narrating progress instead of producing it. None of that requires hiring. It requires the firm's own scoping standard and its own engagement history to actually persist somewhere other than a partner's head.

Helm, Atlas, and Lumen vs the alternatives

Firms evaluating AI agents for consulting firms usually compare three categories: a generalist chatbot, an autocomplete layer bolted onto a document tool, and a purpose-built engagement team. The comparison only makes sense once you separate what each category was actually built to do. A generalist chatbot was built to answer a single prompt well. An autocomplete layer was built to speed up next-token suggestions inside an editor. Neither was built to hold a firm's scoping standard, its playbook of what worked on past engagements, or a live view of engagement health across nine concurrent client relationships.

Install and try

Tonone is free and MIT-licensed. Install it once and Helm, Atlas, and Lumen, along with the rest of the Tonone team, are available in your Claude Code session. You pay only for the Claude Code token usage during the work itself.