When to build vs buy AI capability in a portfolio company
Most operating partners approaching AI in a portfolio company ask the wrong question first. They ask "what AI tools should we be using?" before they have answered "what problem are we actually solving, and what does winning look like in our hold period?"
That sequencing error is expensive. It produces either premature investment in bespoke capability that the business cannot sustain, or a proliferation of off-the-shelf tools that never connect to the value creation plan. Both paths feel like progress. Neither produces returns.
The build vs buy decision in AI is not a technology question. It is a capital allocation question - and it deserves the same rigour you would apply to any other investment decision within the portfolio. This article gives you a framework for making that call correctly: when to buy commercial tools, when to build proprietary capability, and how to sequence the decision across a typical hold period.
Why the standard advice fails portfolio companies
The consulting orthodoxy on build vs buy tends toward "buy where possible, build only where you have genuine competitive advantage." That is reasonable guidance for a FTSE 100 with a ten-year technology roadmap. It is less useful for a £600m revenue industrials business eighteen months into a five-year hold with a transformation thesis that depends on margin expansion.
Portfolio companies face constraints that enterprise technology guidance rarely accounts for. Technical debt is often significant. Internal capability is thin. Management bandwidth is finite and already stretched. And the clock is running.
Against that backdrop, buying a commercial AI product feels like the safe choice. It is fast to deploy, the vendor handles maintenance and the board slide writes itself. But "safe" and "value-creating" are not the same thing. A generic AI tool pointed at a generic process produces generic efficiency gains. That is not what justifies a premium at exit.
The question is not whether to buy or build. The question is which capability creates asymmetric value - and whether that value can be captured within the hold period.
The three categories that determine your decision
Not all AI capability is equal in strategic terms. Before you can make the build vs buy call, you need to categorise what you are actually trying to do.
Commodity AI covers tasks where the underlying capability is widely available, the problem is well-defined and no competitive advantage flows from doing it yourself. Automating invoice processing, generating first-draft marketing copy, transcribing customer calls - these are commodity applications. Buy. The market has solved this. Building bespoke capability here is capital destruction.
Enabling AI covers capability that is not sector-specific but that materially improves the speed or quality of decisions the business makes. Commercial intelligence tools, demand forecasting platforms, pricing analytics - these exist as products, but the value comes from how deeply they are integrated into the operating model. This is where configuration and integration effort determines whether you get the return. Buy the platform, but invest in the integration.
Differentiating AI covers capability that depends on your data, your customer relationships or your operational context in a way that competitors cannot easily replicate. A logistics business with fifteen years of route and margin data. A specialist distributor whose pricing model reflects supplier relationships that no third party understands. A healthcare services business with proprietary outcome data. Here, buying a generic product either misses the opportunity entirely or hands your data advantage to a vendor. This is where building creates durable value - and where the investment is justifiable in exit multiples, not just operational savings.
Most portfolio companies have one, possibly two, genuine differentiating AI opportunities. The rest is commodity or enabling. The discipline is knowing which is which before you spend.
How hold period affects the decision
Sequencing matters as much as categorisation. A five-year hold creates a specific decision window. Year one is rarely the right time to build. Year four is rarely the right time to start.
A practical staging approach looks like this:
- Months 1–6: Run a structured AI readiness assessment. Understand the data assets, the technical infrastructure, the internal capability and - critically - where the value creation thesis actually touches AI. Do not start with use cases. Start with the thesis.
- Months 6–18: Deploy commodity and enabling tools. This generates early wins, builds internal confidence and, importantly, surfaces the data quality problems you will need to solve before any bespoke build is viable.
- Months 18–36: Invest in differentiating capability where the readiness assessment identified a genuine opportunity. This is when you build - with sufficient runway to embed, iterate and demonstrate impact before the exit process begins.
- Months 36 onwards: Shift focus to documentation, demonstrability and buyer narrative. Sophisticated acquirers and secondary buyers are increasingly pricing AI capability into valuations - but only if it is real, embedded and measurable.
A consumer healthcare business we worked with had strong NPS data stretching back a decade across a loyal repeat-purchase customer base. That data had never been properly structured or activated. In months one through twelve, they deployed a commercial BI and reporting platform to clean and consolidate it. In month eighteen, we built a proprietary churn prediction and personalisation layer on top. By month thirty-six, the capability was central to their trade sale narrative - and the acquirer paid for it.
The capability question most sponsors get wrong
Even when sponsors correctly identify a differentiating AI opportunity, they frequently make the wrong call about where the capability sits.
Buying a product to sit on top of your data is not building. It is licensing. If the vendor relationship ends, the capability ends with it. That matters because acquirers and strategic buyers increasingly want to own capability, not inherit vendor dependencies.
Equally, "build" does not always mean hiring a data science team. For most portfolio companies at this revenue level, that is neither fast enough nor cost-effective. The more appropriate model is building proprietary logic, models and data pipelines - with external support during the build phase - in a way that hands genuine IP into the business.
The test is simple: if you terminated every third-party AI contract tomorrow, what would remain? If the answer is nothing, you have bought operational tools. That is fine for commodity and enabling capability. It is not a strategic asset.
For differentiating capability, the goal is a system where the value lives in your data and your trained models - not in a vendor's platform.
What good looks like at the point of exit
Buyers are not yet consistent in how they value AI capability, but the direction is clear. The businesses achieving the strongest AI-related valuation uplift share three characteristics.
First, the capability is tied to a measurable commercial outcome - not a cost reduction, but a revenue, margin or retention metric that the acquirer can model.
Second, the capability is embedded in the operating model. The commercial team uses it. The finance team reports from it. It is not a proof of concept sitting in an engineering environment.
Third, the underlying IP is clear. The data is owned, the models are documented, the logic is transferable. A buyer can look at it and see something they are acquiring, not a set of running costs they are inheriting.
A portfolio company that enters exit with these three attributes in even one domain is materially better positioned than one that has deployed twelve SaaS tools and calls itself AI-enabled.
Make the decision before it makes itself
The most common outcome is not a bad build vs buy decision. It is no decision at all. Tools accumulate. Vendors are onboarded. A pilot runs and is quietly deprioritised. Eighteen months pass.
The cost of that drift is not just the wasted spend. It is the window that closes. Building differentiating capability takes time. The business that starts in month six reaches demonstrable maturity by exit. The business that starts in month thirty does not.
If you are an operating partner reviewing AI investment across your portfolio, the right starting point is a structured diagnostic: map the value creation thesis, assess data and technical readiness, identify which category each potential use case falls into. That scoping work takes weeks, not months, and it prevents the much more expensive mistake of building the wrong thing or buying a shelf-full of tools that never connect.
Rodan runs structured AI diagnostics with portfolio companies as a starting point into larger transformation work. If you are making build vs buy decisions across your portfolio - or want to pressure-test the AI strategy in a specific asset - contact the Rodan team to discuss a diagnostic engagement.
