Case studies

Brief

A US firm of about 50 people wins all its work through public-sector bids. A standard proposal takes about 40 hours, roughly $4,000 of staff time, and the firm submits around 93 a quarter. On a $30,000 task order, one lost bid can wipe out the profit of a won one. Bid / no-bid was a judgement call made by three to five people. The firm had recorded who beat it on just 13 of 1,404 lost bids.

What we did — the four Oper8 moves

1. 01

   Choose a process

   We chose the bid / no-bid decision and started with subcontract bids, about 75% of the firm’s volume. In a workshop with the team we designed the process first and added AI second, so the agents would not speed up a broken process.

   →Scoring candidate processes · AI on top of a broken process

2. 02

   Assess the economics

   We priced the outcome: what a won contract costs once the lost bids are counted. The discovery estimate: dropping half of the small, loss-making bids frees about $240,000 a year of staff time for larger bids.

   →Cost-per-Outcome

3. 03

   Test a pilot

   We built a Knowledge Base of the firm’s whole bid history: some 3,400 bids with outcomes, 575 delivered projects with fees, 535 partners and competitors, 215 client agencies. Quality was checked before launch: 1.9% unsupported facts on a random sample, scored against 46 real questions from the team. We replayed past bids whose outcome was already known. For every new bid, an agent now prepares a bid / no-bid brief and a pack: RFP breakdown, budget and staffing at the firm’s real rates, and interview prep.

   →Evals before the system · The Dual Run

4. 04

   Keep improving

   Every go or no-go is recorded with its reason, and outcomes are captured on wins as well as losses, so the system learns what separates a won bid. Since September it has run in the firm’s own SharePoint, with a weekly agent check of the Knowledge Base.

   →The Decision Capture pipeline · The Data Flywheel

What the client kept

• ✓

  Live bids run through agent-built packs. In one week in late September, the proposal manager opened three new packs and two more colleagues started working with the agents.

• ✓

  Bid know-how that no longer depends on one person.

• ✓

  For the first time, a record of bid decisions and outcomes — the basis for win rate by client and type of work.

Brief

One of the leading executive search and organisation design firms in its market, with a team of about 15, wanted AI to take on the routine of a search. Its material sat on a shared drive and in the firm’s ATS, and a precedent was found only if someone remembered it. The whole team also needed to learn to work this way.

What we did — the four Oper8 moves

1. 01

   Choose a process

   The people who run searches mapped the work themselves, in three groups, from client request to close: thirteen stages, each with what the team does, what it produces and what it needs to know. This gave 25 recurring tasks in six areas as candidates for AI.

   →Scoring candidate processes

2. 02

   Assess the economics

   The owner chose to start with one question that pays back on every new search: “Have we searched for a similar executive before? Show the past projects, the role requirements and the search strategy, with sources.” Everything else came after it.

3. 03

   Test a pilot

   We built a Knowledge Base of about 97,000 records from the firm’s archive: clients, projects, positions, people and long-list entries, each linked to its source document. On the firm’s own server, the team works in one shared AI workspace with one set of agent rules. Skills answer questions from the Knowledge Base with sources, read data from the ATS and record what a consultant knows about a client or candidate. Five workshops with the whole team ran alongside the build: between sessions we extended the system, and in sessions the team used it on their own work.

   →Context Engineering · The Decision Capture pipeline

4. 04

   Keep improving

   The team adds to the Knowledge Base itself, keeps its working rules for the firm, each client and each project, and turns a plain description of a task into a new skill. An internal champion looks after the system on the firm’s side.

   →The Rulebook · The Champions Network · The Data Flywheel

What the client kept

• ✓

  An agent system the team uses in its daily search work.

• ✓

  A search process built around AI and written down stage by stage.

• ✓

  A team that builds its own skills. Consultants have built skills that write a candidate report in the firm’s own three-part format and prepare reports for clients.

Brief

Haleon wanted employees in several functions to learn how to build agents for their own work, with support after the training ended.

What we did

1. 01

   Two-month intensive for 15 participants

   Champions from marketing, finance, R&D and operations. Each brought a real problem from their own backlog.

2. 02

   Weekly workshops

   Agent architecture, retrieval over internal documents, prompt design — worked through on the participants’ own cases.

3. 03

   Mentoring all the way through

   Our team worked with each champion, from the first project outline to presenting the work to its sponsor.

What the client kept

• ✓

  15 champions trained across several functions.

• ✓

  The first agents running in production — document processing among them.

• ✓

  Participants presented their projects to the sponsor at the end of the programme.

Brief

Work was spread across Google Sheets, Notion pages and Slack integrations. The goal was a single workspace for scrum masters and project managers — built by the people who would use it.

What we did

1. 01

   Seven vibe-coding modules

   Participants built automations with n8n and Make and connected them to Slack, Notion and Asana. Each module used tasks from their own work.

2. 02

   Mentoring from first automation to production

   Mentoring continued beyond the modules until an agent was running in the teams’ workflow.

What the client kept

• ✓

  An internal platform assembled by the participants rather than handed to them.

• ✓

  Scrum masters and project managers work in one place instead of a spread of sheets and integrations.

• ✓

  Participants report saving around ten hours a week on routine work.