Case Study
Forrest Co. Turns a Broad AI Ambition into a Scoped, Client-Facing Revenue Diagnostic Built on Claude
Clearscale ran a generative AI discovery and assessment engagement that took Forrest Co. from ten candidate AI use cases to two ranked pilots and a fixed-scope, four-week build plan for a new client-facing service offering, with nine objective success criteria, five milestones, and an explicit out-of-scope list.
Client Profile
Overview
Forrest Co. is a B2B technology marketing agency whose leadership set out to become an AI-enabled consultancy rather than a creative agency that uses AI at the edges. The firm already runs its own named agents in daily production, and its teams are fluent with Claude and MCP tooling. What it wanted next was
bigger: a client-facing, AI-powered service offering it could sell, and a defensible way to choose which one to build first.
Clearscale ran a generative AI discovery and assessment session
with Forrest’s CEO, Chief Creative Officer, and COO.
The engagement inventoried ten candidate use cases, scored each on
business value, fit for generative AI, and level of effort, and had the three leaders rank independently before comparing. All three converged on the same top two.
The output was a fixed-scope, four-week build plan for pilot one, the Marketing & Revenue Growth Assessment, sequenced across five milestones with nine objective success criteria, capped deliverables, and a firm scope freeze at the end of week two.
Clearscale funded the engagement as an investment in the
partnership.
Meet Our Hero
Forrest Co. builds brand, positioning, campaign, and demand-generation programs for B2B technology companies. Its differentiator has always been narrative: taking genuinely complex technology and turning it into a story a buyer acts on.
That market is changing quickly. Forrest’s own clients are adopting AI inside their marketing functions, and buyers increasingly expect their agency to bring strategy and diagnosis, not just execution. Forrest’s leadership saw the same shift and moved early: the firm already operates internal agents for pay-per-click research and project coordination, and its leaders work in Claude daily.
The ambition was to convert that fluency into something sellable: a repeatable, defensible diagnostic Forrest could run on a client’s own data and put in front of that client as strategy. The ideas were plentiful and the appetite was real. What was missing was a shared mechanism for choosing among them, and an engineering partner to turn the chosen one into a bounded build
The Goal
• Establish a shared, defensible way to rank AI opportunities across the leadership team
• dentify which candidates create new client-facing revenue, not only internal time savings
• Keep every artifact editable by a Forrest strategist without engineering support
• Fix data boundaries and a human validation gate before any client data moves
• Produce a fixed-scope plan with objective success criteria and a firm out-of-scope list
The Challenge
Challenge 01
Forrest’s leadership had accumulated more strong AI opportunities than a focused team could pursue at once, and
no shared, objective mechanism for ranking them against one another
Challenge 02
The most valuable candidates were client-facing revenue offerings rather than internal efficiency tools, so the choice
carried commercial consequences and could not be settled on intuition
Challenge 03
Forrest’s teams are fluent operators of Claude and MCP tooling, but the firm does not carry in-house AI engineering, so
whatever was built had to be something Forrest could run, change, and extend itself after handover
Challenge 04
Client marketing and sales data is commercially sensitive and arrives in a different shape from every client, so data boundaries, model access, and human review had to be settled before any build could begin
The Solution
Clearscale ran a five-stage assessment that worked from an open idea space down to one bounded,
fundable build. Every candidate was scored on the same axes, every deliverable was capped, and
everything not named was placed explicitly out of scope.
Step 01 | Current State & Art
of the Possible
• Opened with a structured review of the AI tooling Forrest already runs in production, including its named pay-per-click and project-coordination agents
• Presented where generative AI is genuinely strong and where it is the wrong instrument, so candidates were judged against real capability rather than expectation
• Walked the methodology ladder so the scoping conversation shared a vocabulary
Step 02 | Use-Case Ideation
• Captured ten candidate use cases live into a shared prioritization sheet with the CEO, Chief Creative Officer, and COO working in the same room
• Recorded the data sources and current manual workflow behind each candidate, not just the idea
• Included both internal-efficiency and client-facing candidates so the two could
be compared on the same axis
Step 03 | Stack-Ranking
• Scored every candidate on three axes: business value, fit for generative AI, and level of effort
• Ran head-to-head comparisons where scores tied, rather than letting a three-band scale decide
• Established that fit for generative AI was not the differentiator, because all ten candidates qualified; revenue impact and effort carried the decision
• Had each leader score independently before comparing; all three converged on the same top two
Step 04 | Pilot Deep Dive
• Selected the Marketing & Revenue Growth Assessment as pilot one and the Campaign Editor & Insights Agent as pilot two
• Mapped the diagnostic’s data inputs across ten funnel domains, from business strategy and brand position through pipeline, sales
execution, customer growth, and financial performance, then capped v1 at up to six
of those ten, so the first build could be proven end to end rather than spread thin
• Bounded CRM support to a single system, chosen at kickoff from Salesforce, HubSpot, or Zoho, the three that cover the majority of
Forrest’s client base, rather than building against every CRM on the market
• Sequenced pilot two and the remaining use cases as separate follow-on work rather than folding them into the same build
Step 05 | Fixed-Scope Build
Plan
• Converted pilot one into a four-week plan across five milestones: benchmark co-
design, scope freeze, build, pilot validation, and handover
• Wrote nine objective success criteria spanning three questions: does the diagnostic work, can Forrest run it, and is the output client-ready
How v1 Is Designed to Work
• A Revenue Growth Assessment Claude skill
family, to be authored against real client material Forrest already holds rather than synthetic examples
• Will run locally on Forrest’s own machines through Forrest’s own Claude client, with nothing deployed, no hosting, and no environments to operate
• A strategist-editable benchmark file, one row per funnel component, changeable without a code edit
• A strategist validation gate in the output template, so no finding reaches a client
unreviewed An evaluation harness and a golden set of up to 25 input- and-expected-finding pairs, producing a result a non-engineer can read, so Forrest can test the diagnostic itself after handover
• Intake is by documented export into a per-client folder rather than live connectors
into client systems: three export procedures and three parsers Forrest controls, no credentials held between engagements
• Scores are computed by deterministic rules in code against the benchmark file,
not by the model; the model drafts the narrative, it does not produce the numbers
What the Assessment Produced
A ranked portfolio of ten AI use cases the leadership team scored together and agrees on, with the unselected candidates sequenced rather than
discarded
Two pilots chosen by independent convergence:
three leaders scoring separately arrived at the same top two
A fixed-scope four-week build plan across five milestones, with nine objective success criteria replacing subjective sign-off at the end of the build
A build designed from the outset as a service offering rather than an internal tool; the diagnostic runs on a client’s own data and produces output
fit to put in front of that client as strategy
A requirement that per- assessment model cost be measured during the pilot runs, so Forrest can price the offering against a known unit cost rather than an estimate
Operational independence written in as a success criterion: a Forrest operator changes a benchmark value and a prompt and re-runs the
diagnostic without Clearscale involvement