I caught up with David Ullman of Robust Decisions the other day. Robust Decisions has an interesting product called Accord aimed at helping with decisions. While it is not aimed at exactly the same kind of decisions as the systems I usually review, I like the product and think there is some interesting synergy between it and technologies for operational decision making. First, a first look at Accord.
Process control, as often noted on this blog, leads to a need to focus on decisions. Most non-operational decisions in a business require lots of people, some management of uncertainty and a bunch of tacit knowledge. David has spent many years working with all sorts of companies and government agencies facing these kinds of problems. Which course of action is chosen is generally dependent on a mix of social and technical issues and decided in the face of uncertainty. This is where Accord comes in.
Accord is currently a desktop application and runs both standalone and in a server based configuration (the most common purchase by companies). The user, someone defining the decision making problem, picks from several templates designed to help with specific problems or frames a problem from scratch. For instance, there is a vendor selection template for choosing from several alternative vendors. Accord is built on a Bayesian platform and everything is done using probability so both qualitative and quantitative elements can be considered. A major feature is that there is no need for any statistical knowledge or probability estimation to use Accord. Multiple members or users can be added to the tool so that multiple evaluations of the problem can be considered. Initially all the alternatives (vendors say) are considered equal and the process of using the tool allows you to frame the problem and differentiate between the options. Throughout, the software tracks each of the options from which the selection is to be made using colors. This keeps everything in the tool coordinated so you can tell which evaluations or comments relate to which option.
In the assessment phase each person participating can weigh the available criteria either by ranking them or by giving each a relative weighting. Several viewpoints can be stored for a problem – either for the individuals participating or perspectives within the organization (development, marketing, customers etc). Each option can then be considered for each criterion from each of these perspectives. The assessment is on two dimensions – how well the alternative will satisfy the criterion and how certain the person making the assessment is that the assessment is accurate.
Accord allows various things to be specified. For quantitative criteria you can specify the values that will delight you or disgust you – the min/max thresholds within which you will be happy. For each alternative you can specify the range of values you think that each option might offer (a target value and range around it). And so on – a variety of sliders, data entry fields etc are provided to capture the information. As this data is entered the relative overall satisfaction with each alternative is recalculated (using the uncertainty built in) and shown for each perspective.
At any point you can click on a ‘What to do next’ button. This does a form of sensitivity analysis to recommend some actions that will increase the separation between the alternatives – to make them more easily differentiated. This is supported by measuring the degree of consensus on the criteria and a calculation of the risk in each alternative. The alternatives can also be assessed on the probability they will be the best.
The tool has some great characteristics. It does not force a particular degree of detail as it constantly recalculates the relative strength of your alternatives, so you could stop if one was clearly going to come out ahead. It lets you develop multiple perspectives and then combined them. And it does this in a way that considers how aligned the participants are (to assess risk) and how certain they are (to weight more certain assessments more strongly than guesses). It provides a great way to develop an answer that has some real structure and traceability without imposing a mechanical process.
Accord is also moving to a SaaS platform to increase its availability and Robust Decisions are increasing its focus on replacing meetings. After all, many meetings are aimed at decision making.
I think Accord has real value for those in operational decision making also. The idea of using Accord to help decide what changes to make to the rules or models in an operational decision is one area. Similarly in operational decisions where 5-20% are referred for manual review (a common situation) the use of Accord to make the manual decisions could be very interesting especially as the decision service could specify the allowed alternatives. In both cases some interesting integration is possible and Robust Decisions is interested. David and I would love to hear from any readers out there interested in either of these.
To learn more about the software visit Robust Decisions web site. To learn more about making good decisions see David’s book Making Robust Decisions: Decision Management For Technical, Business, & Service Teams, which you can get at amazon.com and which I reviewed previously.