A founder who had sold his business to a private-equity fund walked me through its portfolio: companies nearing exit, companies midway through the hold, and new acquisitions putting leadership in place. Each needed a different AI program.
Codos starts with one company whose leadership can change the work and demonstrate a result. That gives the next portfolio team something useful to reuse.
01The first company
Choose the company that can change.
Choose the first company against four conditions:
- Sponsor
- Will the CEO own the business goal and resolve decisions that cross functions?
- Economics
- Is the constraint visible in a number such as cost to serve, throughput, conversion, or time to market?
- Timing
- Can the company implement and demonstrate the change within its investment timetable?
- Transfer
- Will another portfolio company face a similar problem and benefit from what the first team learns?
02Impact by function
What changes across functions.
AI changes the economics of different functions in different ways. These examples show the workflows involved and the effects reported in client work.
| Function | Workflow | Reported effect |
|---|---|---|
| SalesSales & marketing | Self-service onboarding and application handling | 80%+of onboarding roles optimized |
| MarketingSales & marketing | Performance marketing optimization | ~30%lower customer acquisition cost |
| Content agents | Over halfof content-team roles optimized | |
| R&DResearch & development | Agents that write and run automated tests | ~80%of capacity unlocked |
| Engineering agents, design and frontend automation | ~75%of capacity unlocked | |
| HRGeneral & administrative | AI interviews, applicant handling and performance workflows | ~50%of capacity unlocked |
| FinanceGeneral & administrative | Invoice, payment and reporting agents | ~60%of capacity unlocked |
| Customer supportCost of service | Voice and text agents | ~75%of capacity unlocked |
| Across functionsMultiple cost lines | Company Brain | Hours backeach day for a subset of employees |
Capacity figures are estimates based on reported team changes, not measured increases in output. “Roles optimized” describes the scope of changed work. These figures are not additive.
03What we build
Codos PE Plays
Each play targets a completed piece of work. These examples follow one client engagement, with delivered systems, diagnostic findings and proposed next steps identified separately.
Sales automation
Handle more applications without making every new customer another sequence of calls, emails, and manual data entry.
- What Codos builds
- Codos builds self-service onboarding connected to the CRM. Agents extract application information from uploaded business documents, populate the customer record, and move the application through a defined process. Call analysis informs follow-up rules, so what the team learns from conversations changes how sales operates.
- From the work
- Customers uploaded documents once; the flow extracted the information into the application and CRM, with agreement drafts prepared for review. Codos reported application handling cost falling by roughly three-quarters and monthly application capacity increasing more than fourfold. Capacity describes what the workflow could handle; actual customer volume remains a separate measure.
- What to measure
- Track completed applications, time to completion, handling cost, and exceptions requiring a person.
Finance-HR-legal automation
Remove the drafting, information gathering, and coordination that accumulate between back-office functions.
- What Codos builds
- Codos runs AI interviews, builds a map of operations, and ranks queues that cross departmental boundaries. We then scope workflows such as document drafting and report assembly around their source records, review requirements, and owners.
- From the work
- A back-office diagnostic mapped hundreds of operations in about two weeks, then ranked queues crossing departmental boundaries. Contract drafting and signing became a priority because the work passed through accounting, HR and legal. The report identified potential capacity; an implemented workflow and a follow-up measurement were still needed to establish a realized result.
- What to measure
- Follow a document or report from request to acceptance. Measure waiting time, manual assembly, rework, and the recurring workload actually removed after implementation.
QA automation
Shorten the regression queue so engineering can release product changes without repeatedly rebuilding a manual testing effort.
- What Codos builds
- Codos deploys agents that write and run automated tests within the product’s testing workflow. The team reviews the results and focuses human attention on failures and cases needing judgment. The objective is a testing capability that keeps up with delivery as the product changes.
- From the work
- In a completed QA pilot, manual regression had occupied much of the release cycle. Agents took on writing and running automated tests, and the client recorded higher automated-test output. Its reporting tracked delivered effects separately from the benefits still planned for the wider rollout.
- What to measure
- Measure regression duration, test reliability, defects reaching production, and manual effort per release. More tests are useful when they produce faster, dependable releases.
E2E engineering agents
Turn a production signal or product request into a shipped change through one connected engineering workflow.
- What Codos builds
- Codos links monitoring and tickets to agents that plan, implement, review, and prepare changes for release. Accepted architecture and product documents govern the work. Automated checks block non-compliant changes, while the team retains its approval gate and visibility into agent activity.
- From the work
- In an engineering build, short tickets became plans with acceptance criteria and code references. Agents implemented against accepted documentation; automated review sent problem changes back before they reached a person. The team retained its approval gate and could inspect recorded agent sessions. The reusable asset is that connected delivery process, including the checks that stop a bad change.
- What to measure
- Track time from request to accepted release, review effort, failed changes, and customer-facing delivery.
Marketing funnel automation
Convert more of the demand a company already generates by fixing losses between first contact, payment, and activation.
- What Codos builds
- Codos maps and measures each funnel stage, then rebuilds follow-up, qualification, payment, and activation around the points where customers stall. Agents and workflow automation handle the repeatable steps; owners resolve the commercial decisions and exceptions that hold the journey up.
- From the work
- Codos reported that most self-service leads in one operating funnel reached payment, alongside released team capacity. A useful fix was small: stop sending several messages simultaneously so the application link was easier to find. A separate funnel rebuild proposed coordinated contact, payment links and earlier document checks. Its higher lead-to-live conversion remained a target.
- What to measure
- Follow the same lead population through to paid, active customers. Measure conversion at each transition and acquisition cost, including the cost of follow-up and service.
04A play in practice
Follow one customer all the way through.
The proposed second-funnel rebuild shows how a play becomes specific. Payment was a weak transition, while most won deals reached the partner. The response was to make payment easier where customers stalled and preserve the handoff that already worked.
The next loss was after the partner handoff, before the merchant could go live. That called for earlier document checks and an owner following the case through activation. The proposal concentrated work at those transitions; it did not assume that adding an agent everywhere would improve the funnel.
Unreached leads need a deliberate follow-up process. Decide who owns retries and when a human takes over.
A conversation needs a clear next step. Route ready customers to the right team and keep track of those who need more time.
A customer who agrees to buy can still stall. Put payment within the conversation and make outstanding payments visible.
Incomplete business documents create another queue. Validate the required information before passing the application to a partner.
Payment is not the end of the journey. Track the application until the customer is live, with an owner for unresolved cases.
An illustrative journey adapted from the proposed second-funnel rebuild. Select a stage to see the operating decision behind it.
05Management makes it move
Give the work an owner who can decide.
In the onboarding rollout, the signing step was waiting on legal and finance review. Engineers needed approved rules, an approver and an exception path before they could finish the workflow. Those were leadership decisions.
The fund can sponsor the work. The company’s leadership has to change how it runs.
06The economic result
Show how the operating change becomes value.
Agree the financial mechanism before building. Compare the result with its baseline and include the new workflow’s running cost.
- Expense removed
- Show which actual expense changed, when it changed, and the service or quality measure that held up alongside it.
- Capacity put to work
- Show where the released time went: more customers served, a backlog cleared, or hiring avoided against an agreed plan.
- Incremental revenue
- Follow the change through to paid business, account for delivery costs, and distinguish the effect from unrelated growth.
07From one company to the next
Carry the method across the portfolio.
For the next portfolio company, reuse the workflow design, document schema, exception tests and measurement method. Recheck the parts that depend on the business: which documents it needs, who can approve a case and what counts as a completed customer journey. Each company supplies its own records and rules within its own access boundaries.
A few practical questions
Before we start.
Should we start across the whole portfolio?
Start where the economic need and management authority are clearest. Use that company to establish the delivery method and evidence. Expand when another company has a relevant problem and a sponsor ready to act.
Can this work with our existing technology team?
Yes. Define who owns the systems, who will build and maintain each workflow, and where Codos adds capacity or expertise. The plan should fit the company’s operating team and approval process.
Does every company need to share its data with the fund?
No. Reusing a method does not require pooling company data. Agree each company’s deployment, model access, permissions, and reporting scope separately, including what evidence the fund is authorized to see.
What should we bring to the first conversation?
Bring one candidate company, the business number that needs to change, and the leader who owns it. An example of a slow or costly workflow helps us test whether a diagnostic is the right next step.
