BUILD VS. BUY
Enrichment, an orchestration tool, an AI copilot and a few good afternoons will get you most of an ABM system. But they won't get you the identity layer underneath it, and what you assemble won't still be running in six months.
THE HALF-LIFE
A stack you assemble yourself can give you everything you need to run ABM on day one. But while your system seems to be working well, it's quietly degrading over time. And with no identity graph of your own, you have nothing to reliably match audiences against. Vector lets you run a modern ABM program as a maintained product, so there's no inevitable decay.
Build at launch
Build at month 6
Nobody has revisited the scoring since launch
Licensed data has gone stale
A source API changed its schema
Credits ran out mid-quarter
Never actually got built
THE BUILD PATH
Each of these is a real piece of software that someone maintains full time, and an ABM system needs them all running at once, forever. All but one of them you could genuinely build.
An account model that stays honest
Scoring and staging are easy to write and hard to keep true. A model nobody revisits after launch quietly stops describing your pipeline.
A signal pipeline that survives
Every source is somebody else's API, with its own schema, rate limits and deprecation calendar. The pipeline works until one of them ships a change on a Thursday afternoon.
Audience sync that holds
Each ad platform hashes, matches and refreshes differently. Getting an audience in once is an afternoon. Keeping ten of them current as buying committees change is a standing job.
Proof that ties an ad to a deal
Connecting an impression to a named contact to a closed deal is the piece teams always defer, and the piece leadership always asks for. Bolting it on after the fact means rebuilding half the stack.
Someone who owns all of it
One item here isn't software at all. Not someone who owns it until launch, but someone who owns it for good, because the stack lasts exactly as long as that person stays.
An identity graph
This is the one you can't build. Ad platforms don't match on work email. They match on the personal email, phone number, or device a person uses on LinkedIn or Meta. Connecting a work identity to those personal identifiers isn't an API call, it's a dataset built up over years of matching. You can license contact records. You can't license the thing that makes an audience actually match.
THE AI QUESTION
This is the best version of the build argument. An agent can write most of this. Writing it was never the hard part, though.
Agents are good at exactly the parts that were already easy
Writing a connector, transforming a payload, scheduling a job. Sourcing an identity graph isn't a coding problem, so no amount of generation produces one.
Generated code is still code you own
Every integration an agent writes for you is one you still maintain, and the model isn't on call when the schema changes.
The failure mode is silence
Assembled systems rarely announce that they’ve stopped working. Campaigns get quieter, and someone figures out why a quarter later.
Assembly isn't accountability
When the audience stops matching two weeks before the quarter closes, someone has to fix it, and fast. An agent isn't that someone.
THE MATH
| Build it yourself | Vector From $3,000/mo + credits | |
|---|---|---|
| Identity resolution | Licensed per record, and no vendor sells the graph itself | Included, and proprietary rather than resold |
| Signal sources | Each one priced, integrated and maintained separately | 30+ native in platform, plus the ability to create custom signals |
| Engineering to build | Quarters before the first campaign runs | None |
| Engineering to maintain | Ongoing, forever, and invisible until it stops | None |
| Ad ops hours | Recurring list building, uploading and re-uploading | Audiences update themselves |
| Reporting | A project of its own, usually deferred indefinitely | Built into the platform |
| When it breaks | You're on the hook to find the problem and make the fix | A vendor and Customer Solutions team who will help make it right |
On the first estimate, usually. That estimate tends to include engineering time and data, and to leave out maintenance, ad ops hours, and the attribution work that gets deferred until leadership asks for it. For scale: Vector starts at a $3,000/mo platform fee. That is roughly what one engineer-week costs, and the build needs rather more than a week.
You probably can, and the first version will work. The longer-term question is whether you want your team spending the next three years maintaining it, or spending that time improving your own product.
Yes, and plenty of teams do. Vector runs the signals and data you trust. Your stack stays exactly where it is, and Vector takes over the hand-wiring underneath it.
